Air conditioner fault diagnosis method, air conditioner fault diagnosis device and air conditioner fault diagnosis system

Through the combination of autoregression method and big data technology, a potential fault diagnosis model is generated, which solves the problem that air conditioner fault diagnosis depends on predefined rules and thresholds, and realizes comprehensive fault monitoring and efficient fault handling of air conditioning systems.

CN118940010APending Publication Date: 2024-11-12GREE ELECTRIC APPLIANCE INC OF ZHUHAI +1
View PDF 0 Cites 3 Cited by

Patent Information

Application Number
CN202411177816.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-26
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

The existing air conditioner fault diagnosis methods rely on predefined rules and thresholds, and fail to fully tap the potential of large-scale timing data in the air conditioner system, resulting in insufficient comprehensiveness of fault diagnosis.

Method used

The autoregressive method is used to train the training data to generate potential fault diagnosis models, and adjust the model parameters through time alignment feature data. Combined with big data technology and large language model, the air conditioner operation status is monitored in real time, and time series feature data is extracted for faults. Identify.

Benefits of technology

It realizes comprehensive fault diagnosis of air conditioning systems, timely discovers and predicts potential faults, provides efficient fault treatment measures, and improves the accuracy and comprehensiveness of fault diagnosis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118940010A_ABST
    Figure CN118940010A_ABST
Patent Text Reader

Abstract

The invention provides an air conditioner fault diagnosis method, an air conditioner fault diagnosis device and a system, and the method comprises the steps: sequentially inputting multiple groups of training data into an initial potential fault diagnosis model for training through employing an autoregression method, and obtaining an intermediate potential fault diagnosis model and time alignment feature data; adjusting all model layer parameters in the intermediate potential fault diagnosis model by using the time alignment feature data to obtain a potential fault diagnosis model; acquiring real-time operation data of the air conditioner, extracting time sequence feature data in the real-time operation data, inputting the time sequence feature data into the potential fault diagnosis model, and outputting to obtain a current potential fault feature tag; and identifying and processing the fault according to the current potential fault feature tag. The problems that in the prior art, air conditioner fault diagnosis mainly depends on predefined rules and threshold values to conduct fault detection, the potential of a large amount of time sequence data in an air conditioner system cannot be fully excavated, and the comprehensiveness of fault diagnosis is insufficient are solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of air conditioning fault diagnosis, and in particular to an air conditioning fault diagnosis method, an air conditioning fault diagnosis device, a computer-readable storage medium and an air conditioning fault diagnosis system. Background Art

[0002] Air conditioning systems play an important role in modern life and industry, but their complex operating environment and diverse components also bring certain challenges to after-sales maintenance. Even an abnormality in one component of these complex structures may affect the normal use of the entire machine. These failures will not only affect the user's comfort experience, but may also cause energy waste, equipment damage, and even cause safety hazards. Many existing methods use existing data to diagnose air conditioning faults. However, they all use partial sensor data and rely on predefined rules and thresholds for fault detection. They fail to fully tap the potential of a large amount of time series data in the air conditioning system, and the comprehensiveness of fault diagnosis is insufficient. Summary of the invention

[0003] The main purpose of the present application is to provide an air conditioning fault diagnosis method, an air conditioning fault diagnosis device, a computer-readable storage medium and an air conditioning fault diagnosis system, so as to at least solve the problem that air conditioning fault diagnosis in the prior art mainly relies on predefined rules and thresholds for fault detection, fails to fully tap the potential of a large amount of time series data in the air conditioning system, and the fault diagnosis is not comprehensive enough.

[0004] In order to achieve the above-mentioned purpose, according to one aspect of the present application, an air conditioning fault diagnosis method is provided, comprising: using an autoregressive method to sequentially input multiple groups of training data into an initial potential fault diagnosis model for autoregressive training to obtain an intermediate potential fault diagnosis model and time-aligned feature data, each group of the training data includes univariate time series data within the first n time steps and a potential fault feature label of the n+1th time step, the univariate time series data includes at least one of the following: indoor temperature, indoor humidity, voltage and current, and the time-aligned feature data is data obtained after feature processing of the univariate time series data during the autoregressive training; using the time alignment feature data The method comprises the following steps: adjusting all model layer parameters in the intermediate potential fault diagnosis model by aligning feature data to obtain a potential fault diagnosis model, wherein the model layer parameters are at least parameters of an input layer, a convolution layer and an output layer in the intermediate potential fault diagnosis model; obtaining real-time operation data of the air conditioner, extracting time series feature data in the real-time operation data, inputting the time series feature data into the potential fault diagnosis model, and outputting a current potential fault feature label; identifying a fault according to the current potential fault feature label, obtaining a fault result, and executing corresponding fault handling measures according to the fault result, wherein the fault result at least includes a faulty component, a fault type and a fault probability of the air conditioner.

[0005] Optionally, an autoregressive method is used to sequentially input multiple groups of training data into the initial potential fault diagnosis model for autoregressive training to obtain an intermediate potential fault diagnosis model and time-aligned feature data, including a block division step, after sequentially inputting multiple groups of the training data into the initial potential fault diagnosis model, the univariate time series data in each group of the training data is divided into blocks according to a set time block to obtain multiple time blocks, wherein the set time block is a time length formed by a predetermined number of continuous time steps; a normalization step, performing instance normalization on the univariate time series data in each of the time blocks to obtain normalized time series data; a label encoding processing step, performing label encoding on the normalized time series data; The time series data is subjected to label coding processing through the convolutional layer of the initial potential fault diagnosis model to maintain the semantic information of the normalized time series data and obtain label coding feature data; in the time coding processing step, the label coding feature data is subjected to time coding processing using a two-stage aggregation method to obtain the time alignment feature data, and the time alignment feature data has a multi-scale time embedding feature; the blocking step, the normalization step, the label coding processing step and the time coding processing step are repeated at least once in sequence until the initial potential fault diagnosis model is trained with all the training data to complete the autoregressive training and obtain the intermediate potential fault diagnosis model.

[0006] Optionally, the time-aligned feature data is used to adjust all model layer parameters in the intermediate potential fault diagnosis model to obtain a potential fault diagnosis model, including: a training step, using the time-aligned feature data to perform linear detection training on the parameters of the output layer in the intermediate potential fault diagnosis model to preliminarily adjust the parameters of the output layer to obtain a preliminary training potential fault feature label; an adjustment step, using parameter efficient fine-tuning technology to adjust all model layer parameters in the intermediate potential fault diagnosis model; a denormalization step, denormalizing the preliminary training potential fault feature label to obtain denormalized time series data, and performing dimensionality conversion on the denormalized time series data to obtain a current training potential fault feature label; repeating the training step, the adjustment step and the denormalization step at least once in sequence until the current training potential fault diagnosis label is consistent with the potential fault feature label in the training data to obtain the potential fault diagnosis model.

[0007] Optionally, the denormalized time series data is dimensionally transformed to obtain a currently trained potential fault diagnosis label, including: converting the denormalized time series data into a one-dimensional data form to obtain one-dimensional time series data, wherein the one-dimensional data form is the output format of the intermediate potential fault diagnosis model; and using the output layer to linearly process the one-dimensional time series data to obtain the currently trained potential fault feature label.

[0008] Optionally, after adjusting all model layer parameters in the intermediate potential fault diagnosis model using the time-aligned feature data to obtain the potential fault diagnosis model, the method also includes: dividing the test set into multiple test subsets; an evaluation step, evaluating the potential fault diagnosis model using a target test subset to obtain evaluation indicators corresponding to the target test subset, the evaluation indicators including accuracy, recall, F1 score, mean square error and mean absolute error, and the target test subset is any one of the test subsets; repeating the evaluation step at least once until the evaluation is completed using all the test subsets to obtain the evaluation indicators corresponding to all the test subsets; calculating an average evaluation indicator based on the evaluation indicators corresponding to all the test subsets; when the average evaluation indicator is greater than the set evaluation indicator, determining that the potential fault diagnosis model has passed the test; when the average evaluation indicator is greater than the set evaluation indicator, determining that the potential fault diagnosis model has not passed the test, and retraining the potential fault diagnosis model.

[0009] Optionally, after identifying the fault according to the current potential fault feature label and obtaining the fault result, the method further includes: when there are multiple faults in the fault result, querying a fault handling priority table according to the fault type to obtain the fault priority corresponding to each fault, the fault handling priority table being a mapping relationship between fault type and fault handling priority, the fault handling priority being a priority set according to the degree of impact of each fault on the operation of the air conditioner when it occurs; determining a fault handling order according to the fault priority corresponding to each fault and the corresponding fault probability; and handling the fault according to the fault handling order.

[0010] Optionally, after identifying the fault according to the potential fault feature label and obtaining the fault result, the method also includes: continuously acquiring the real-time operating data and historical fault processing results of the air conditioner; regularly re-incrementally re-training or fully re-training the potential fault diagnosis model according to the real-time operating data and historical fault processing results; continuously monitoring the operating performance of the potential fault diagnosis model, the operating performance at least including the accuracy, false alarm rate and missed alarm rate of the potential fault diagnosis of the air conditioner; when the operating performance is lower than the performance threshold, controlling the adjustment of parameters of the potential fault diagnosis model or retraining the potential fault diagnosis model.

[0011] According to another aspect of the present application, an air conditioning fault diagnosis device is provided, the device comprising: a training unit, for sequentially inputting multiple sets of training data into an initial potential fault diagnosis model for autoregressive training using an autoregressive method to obtain an intermediate potential fault diagnosis model and time-aligned feature data, each set of the training data comprising univariate time series data within the first n time steps and a potential fault feature label of the n+1th time step, the univariate time series data comprising at least one of the following: indoor temperature, indoor humidity, voltage and current, the time-aligned feature data being data obtained after feature processing of the univariate time series data during the autoregressive training process; an adjustment unit, for utilizing the time-aligned feature data to adjust the potential fault diagnosis model and the time-aligned feature data. The data adjusts all model layer parameters in the intermediate potential fault diagnosis model to obtain the potential fault diagnosis model, and the model layer parameters are at least the parameters of the input layer, the convolution layer and the output layer in the intermediate potential fault diagnosis model; an input-output unit is used to obtain the real-time operation data of the air conditioner, and extract the time series feature data in the real-time operation data, input the time series feature data into the potential fault diagnosis model, and output to obtain the current potential fault feature label; an identification processing unit is used to identify the fault according to the current potential fault feature label, obtain the fault result, and execute the corresponding fault handling measures according to the fault result, and the fault result at least includes the faulty component, fault type and fault probability of the air conditioner.

[0012] According to another aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute any one of the methods described.

[0013] According to another aspect of the present application, an air conditioning fault diagnosis system is provided, comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include methods for executing any one of the methods described.

[0014] By applying the technical solution of the present application, in an air conditioning fault diagnosis method, first, an autoregressive method is used to sequentially input multiple sets of training data into an initial potential fault diagnosis model for autoregressive training to obtain an intermediate potential fault diagnosis model and time-aligned feature data, each set of the above training data includes univariate time series data within the first n time steps and a potential fault feature label of the n+1th time step, the above univariate time series data includes at least one of the following: indoor temperature, indoor humidity, voltage and current, and the above time-aligned feature data is the data obtained after feature processing of the above univariate time series data during the above autoregressive training; then, the above time-aligned feature data is used to diagnose the air conditioning fault. All model layer parameters in the above-mentioned intermediate potential fault diagnosis model are adjusted to obtain a potential fault diagnosis model, and the above-mentioned model layer parameters are at least the parameters of the input layer, convolution layer and output layer in the above-mentioned intermediate potential fault diagnosis model; then, the real-time operation data of the air conditioner is obtained, and the time series feature data in the above-mentioned real-time operation data is extracted, and the above-mentioned time series feature data is input into the above-mentioned potential fault diagnosis model, and the current potential fault feature label is output; finally, the fault is identified according to the above-mentioned current potential fault feature label, and the fault result is obtained, so as to perform the corresponding fault handling measures according to the above-mentioned fault result, and the above-mentioned fault result at least includes the faulty component, fault type and fault probability of the above-mentioned air conditioner. This application collects, stores and preprocesses massive time series data by using big data technology as training data, and inputs the training data into the initial potential fault diagnosis model, which is a large language model, and undergoes time series alignment processing to obtain a model after preliminary training, that is, an intermediate potential fault diagnosis model, and fine-tunes the model parameters on the basis of the practical sequence alignment processing, and gradually adjusts the model to adapt to the complex operating environment and diverse time series data characteristics of the air conditioning system, so as to obtain the final potential fault diagnosis model. During the actual operation of the air conditioner, time series are extracted from the real-time operation data, and the time series feature data is input into the potential fault diagnosis model, and the current potential fault features are output. This not only takes advantage of the big data technology in data processing and management, but also utilizes the powerful capabilities of the large language model in data analysis and pattern recognition. It can monitor the air conditioner operation status in real time, detect and predict potential faults in a timely manner, and provide efficient fault handling measures. The present application solves the problem that the prior art air conditioner fault diagnosis mainly relies on predefined rules and thresholds for fault detection, fails to fully tap the potential of a large amount of time series data in the air conditioning system, and the fault diagnosis is not comprehensive enough. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A hardware structure block diagram of a mobile terminal for executing an air conditioner fault diagnosis method provided in an embodiment of the present application is shown;

[0016] Figure 2 A schematic diagram of a flow chart of an air conditioner fault diagnosis method provided according to an embodiment of the present application is shown;

[0017] Figure 3 A flow chart of a specific air conditioner fault diagnosis method provided according to an embodiment of the present application is shown;

[0018] Figure 4 A structural block diagram of an air conditioning fault diagnosis device provided according to an embodiment of the present application is shown.

[0019] The above drawings include the following reference numerals:

[0020] 102, processor; 104, memory; 106, transmission device; 108, input and output devices. DETAILED DESCRIPTION

[0021] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0022] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0023] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0024] For the convenience of description, some nouns or terms involved in the embodiments of the present application are explained below:

[0025] Parameter-Efficient Fine-Tuning (PEFT): including LayerNormalization Tuning and Low-Rank Adaptation (LoRA), which optimizes model performance without significantly increasing model parameters. PEFT fine-tunes a small number of key parameters to enable the pre-trained large language model to efficiently process the time series data of the air-conditioning system, improving the accuracy and efficiency of fault diagnosis and prediction.

[0026] As introduced in the background technology, many existing methods use existing data to diagnose air conditioner faults. However, they all use partial sensor data and rely on predefined rules and thresholds for fault detection, and fail to fully tap the potential of a large amount of time series data in the air conditioning system, and the comprehensiveness of fault diagnosis is insufficient. In order to solve the problem that air conditioning fault diagnosis in the prior art mainly relies on predefined rules and thresholds for fault detection, fails to fully tap the potential of a large amount of time series data in the air conditioning system, and the comprehensiveness of fault diagnosis is insufficient, the embodiments of the present application provide an air conditioning fault diagnosis method, an air conditioning fault diagnosis device, a computer-readable storage medium and an air conditioning fault diagnosis system.

[0027] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0028] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Taking running on a mobile terminal as an example, Figure 1 FIG. 1 is a hardware structure block diagram of a mobile terminal of an air conditioner fault diagnosis method according to an embodiment of the present invention. Figure 1 As shown, the mobile terminal may include one or more ( Figure 1 Only one is shown in the figure) a processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data, wherein the mobile terminal may also include a transmission device 106 and an input / output device 108 for communication functions. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the mobile terminal. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0029] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the display method of device information in the embodiment of the present invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, the above method is implemented. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories can be connected to the mobile terminal via a network. Examples of the above-mentioned network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof. The transmission device 106 is used to receive or send data via a network. The above-mentioned specific examples of the network may include a wireless network provided by a communication provider of the mobile terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0030] In this embodiment, an air conditioning fault diagnosis method running on a mobile terminal, a computer terminal or a similar computing device is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0031] Figure 2 FIG. 1 is a flow chart of an air conditioner fault diagnosis method according to an embodiment of the present application. Figure 2 As shown, the method comprises the following steps:

[0032] Step S201, using an autoregressive method to sequentially input multiple sets of training data into an initial potential fault diagnosis model for autoregressive training, to obtain an intermediate potential fault diagnosis model and time-aligned feature data, each set of the above training data includes univariate time series data within the first n time steps and a potential fault feature label of the n+1th time step, the above univariate time series data includes at least one of the following: indoor temperature, indoor humidity, voltage and current, and the above time-aligned feature data is data obtained after feature processing of the above univariate time series data during the above autoregressive training;

[0033] Specifically, sensors installed in air-conditioning equipment are used to collect operation data in real time, and the collected data is transmitted to the parsing server through wireless communication modules (wifi, 4G, etc.). After parsing, the data is sent to the big data platform cache service kafka, and the real-time processing engine is used to store the data in different databases (mysql, hive, kudu, hbase, etc.) according to different usage scenarios. Different processing engines (spark, MR, flink) are used on the big data platform according to different scenarios to normalize the data, eliminate the dimensional differences between different sensor data, remove outliers and missing values, ensure data quality and consistency, and ensure that all sensor data are synchronized in time, using a unified timestamp. There are many ways to normalize, for example, the maximum-minimum normalization formula: Where X max and X min They are the maximum and minimum values ​​of the data, the original data X, and the normalized data X. ′ , the inverse normalization formula can be obtained by the same logic: X = X ′ *(X max -X min )+X min . The training data is obtained by obtaining historical operation data from the database of the big data platform, extracting time series features from the historical operation data, dividing the time series data into sliding windows of fixed length, and calculating statistical features for the data in each window. Features: mean, standard deviation, maximum, minimum, median, quartile, etc. Example: For temperature sensor data, the above statistical features are calculated in each 5-minute sliding window. The autoregressive method is used to input multiple groups of training data into the initial potential fault diagnosis model in turn for autoregressive training to achieve time alignment, and obtain the intermediate potential fault diagnosis model after preliminary training and the time-aligned feature data during the training process.

[0034] Step S202, using the time-aligned feature data to adjust all model layer parameters in the intermediate potential fault diagnosis model to obtain a potential fault diagnosis model, wherein the model layer parameters are at least parameters of an input layer, a convolutional layer, and an output layer in the intermediate potential fault diagnosis model;

[0035] Specifically, the parameters of the model are adjusted based on time alignment to further improve the performance of the model in specific time series prediction tasks. The time alignment feature data can be used to adjust all model layer parameters in the above intermediate potential fault diagnosis model to obtain a potential fault diagnosis model.

[0036] Step S203, acquiring real-time operation data of the air conditioner, extracting time series feature data from the real-time operation data, inputting the time series feature data into the potential fault diagnosis model, and outputting a current potential fault feature label;

[0037] Specifically, after the training is completed, the potential fault diagnosis model can be deployed to the air-conditioning system to perform real-time fault detection and analysis, obtain the real-time operation data of the air conditioner, and extract the time series feature data in the above real-time operation data. The purpose is to convert it into input data that can be recognized by the potential fault diagnosis model, input the above time series feature data into the above potential fault diagnosis model, and output the current potential fault feature label.

[0038] Step S204, identifying the fault according to the above-mentioned current potential fault feature label, obtaining the fault result, and executing corresponding fault handling measures according to the above-mentioned fault result, and the above-mentioned fault result at least includes the fault component, fault type and fault probability of the above-mentioned air conditioner.

[0039] Specifically, after the potential fault diagnosis model outputs the current potential fault feature label, the current potential fault feature label is analyzed to identify the fault, and the alarm mechanism is triggered to notify the relevant maintenance personnel. The alarm method may include SMS, email, APP push, etc. And the corresponding fault handling measures are determined according to the above fault results to execute the corresponding fault handling measures.

[0040] In this embodiment, first, an autoregressive method is used to sequentially input multiple groups of training data into an initial potential fault diagnosis model for autoregressive training to obtain an intermediate potential fault diagnosis model and time-aligned feature data, each group of the above training data includes univariate time series data within the first n time steps and a potential fault feature label of the n+1th time step, the above univariate time series data includes at least one of the following: indoor temperature, indoor humidity, voltage and current, and the above time-aligned feature data is the data obtained after feature processing of the above univariate time series data during the above autoregressive training; then, the above intermediate potential fault diagnosis model is analyzed using the above time-aligned feature data. All model layer parameters in the diagnosis model are adjusted to obtain a potential fault diagnosis model, and the above model layer parameters are at least the parameters of the input layer, convolution layer and output layer in the above intermediate potential fault diagnosis model; then, the real-time operation data of the air conditioner is obtained, and the time series feature data in the above real-time operation data is extracted, and the above time series feature data is input into the above potential fault diagnosis model, and the current potential fault feature label is output; finally, the fault is identified according to the above current potential fault feature label, and the fault result is obtained, so as to perform the corresponding fault handling measures according to the above fault result, and the above fault result at least includes the faulty component, fault type and fault probability of the above air conditioner. This application collects, stores and preprocesses massive time series data by using big data technology as training data, and inputs the training data into the initial potential fault diagnosis model, which is a large language model, and undergoes time series alignment processing to obtain a model after preliminary training, that is, an intermediate potential fault diagnosis model, and fine-tunes the model parameters on the basis of the practical sequence alignment processing, and gradually adjusts the model to adapt to the complex operating environment and diverse time series data characteristics of the air conditioning system, so as to obtain the final potential fault diagnosis model. During the actual operation of the air conditioner, time series are extracted from the real-time operation data, and the time series feature data is input into the potential fault diagnosis model, and the current potential fault features are output. This not only takes advantage of the big data technology in data processing and management, but also utilizes the powerful capabilities of the large language model in data analysis and pattern recognition. It can monitor the air conditioner operation status in real time, detect and predict potential faults in a timely manner, and provide efficient fault handling measures. The present application solves the problem that the prior art air conditioner fault diagnosis mainly relies on predefined rules and thresholds for fault detection, fails to fully tap the potential of a large amount of time series data in the air conditioning system, and the fault diagnosis is not comprehensive enough.

[0041] In order to enable those skilled in the art to more clearly understand the technical solution of the present application, the implementation process of the air conditioning fault diagnosis method of the present application will be described in detail below in combination with specific embodiments.

[0042] In order for the potential fault diagnosis model to better understand and process the time series information of the air-conditioning system, in an optional implementation manner, the above step S201 includes:

[0043] Step S2011, a block division step, after inputting the plurality of groups of the training data into the initial potential fault diagnosis model in sequence, the univariate time series data in each group of the training data is divided into blocks according to a set time block to obtain a plurality of time blocks, wherein the set time block is a time length formed by a predetermined number of continuous time steps;

[0044] Step S2012, a normalization step, performing instance normalization on the univariate time series data in each of the time blocks to obtain normalized time series data;

[0045] Step S2013, a label coding processing step, performing label coding processing on the normalized time series data through the convolutional layer of the initial potential fault diagnosis model to retain the semantic information of the normalized time series data and obtain label coding feature data;

[0046] Step S2014, a time coding processing step, using a two-stage aggregation method to perform time coding processing on the above-mentioned marked coding feature data to obtain the above-mentioned time alignment feature data, and the above-mentioned time alignment feature data has a multi-scale time embedding feature;

[0047] Step S2015, repeat the above-mentioned block division step, the above-mentioned normalization step, the above-mentioned mark coding processing step and the above-mentioned time coding processing step at least once in sequence, until the above-mentioned initial potential fault diagnosis model is trained with all the above-mentioned training data to complete the autoregressive training, and obtain the above-mentioned intermediate potential fault diagnosis model.

[0048] In the above embodiment, a pre-trained large language model (LLM) is selected as the above initial potential fault diagnosis model, such as Figure 3As shown in the figure, the LLM4TS framework was selected for model selection and training. Instance normalization is applied to the input time series data to enhance the consistency and reliability of the model in processing different time series data sets, and the multivariate time series data is split into multiple univariate time series data. Each univariate time series data is input as an independent time series. Continuous time step data is grouped into a block, for example, 5 minutes as a time block, to reduce the time dimension of the input sample and increase the feature dimension. Each time block contains data points of multiple time steps. For example, if each time step is 1 second, a 5-minute time block contains 300 data points. Autoregression is used for training, that is, the initial potential fault diagnosis model predicts the data of the current time step based on the previous time step data. Instance normalization is applied to the data in each time step to obtain normalized time series data to ensure the consistency and reliability of data processing. For the normalized time series data, a one-dimensional convolutional layer is used for label encoding processing to maintain the semantic information of the time series data and obtain label encoding feature data. The label encoding feature data is aggregated at two levels. In the first level of aggregation, the time coding features of each time unit in the marked coding feature data are weighted averaged to obtain the time embedding of each time unit. In the second level of aggregation, the time embeddings of adjacent time units are weighted averaged to obtain a higher level of time embedding. The final time embedding features are extracted by pooling method. The multi-scale time embedding features can be pooled by methods such as maximum pooling and average pooling to obtain the above-mentioned time alignment feature data. Through such a two-level aggregation and pooling method, multi-scale time embedding features can be extracted, so as to better characterize the time information and provide richer time information for subsequent tasks. The above-mentioned block division step, the above-mentioned normalization step, the above-mentioned mark coding processing step and the above-mentioned time coding processing step are repeated at least once in sequence until the above-mentioned initial potential fault diagnosis model is trained with autoregression using all the above-mentioned training data, so that the final potential fault diagnosis model can better understand and process time series information.

[0049] In order to efficiently process the time series data of the air conditioning system and improve the accuracy and efficiency of fault diagnosis, in an optional implementation manner, the above step S202 includes:

[0050] Step S2021, a training step, using the time-aligned feature data to perform linear detection training on the parameters of the output layer in the intermediate potential fault diagnosis model, so as to preliminarily adjust the parameters of the output layer and obtain a preliminarily trained potential fault feature label;

[0051] Step S2022, an adjustment step, adjusting all the model layer parameters in the intermediate potential fault diagnosis model by using a parameter efficient fine-tuning technique;

[0052] Step S2023, a denormalization step, denormalizing the above-mentioned preliminary training potential fault feature labels to obtain denormalized time series data, and performing dimension conversion on the above-mentioned denormalized time series data to obtain current training potential fault feature labels;

[0053] Step S2024, repeat the above training step, the above adjustment step and the denormalization step at least once in sequence until the above current training potential fault diagnosis label is consistent with the above potential fault feature label in the above training data, so as to obtain the above potential fault diagnosis model.

[0054] In the above embodiment, after completing the time series alignment, the prediction fine-tuning stage is entered. The main goal of this stage is to use the time series feature data that has been aligned and encoded to comprehensively fine-tune the intermediate potential fault diagnosis model to improve the accuracy of air conditioning fault diagnosis and prediction. First, only the parameters of the output layer are trained, and the time series feature data after the model is aligned is used for the prediction task of the training target. Training only the parameters of the output layer enables the model to quickly adapt to new prediction tasks while retaining most of the model's feature extraction capabilities. After the initial adaptation, the parameters of all layers are unfrozen and comprehensively fine-tuned to make the model better adapt to the time series prediction task, and all the above model layer parameters in the above intermediate potential fault diagnosis model are adjusted by using efficient parameter fine-tuning technology. In order to ensure that the data corresponding to the model output is consistent with the actual data distribution, the above-mentioned preliminary training potential fault feature labels need to be denormalized to obtain denormalized time series data. Normalization is achieved by calculating the mean μ and standard deviation σ of each feature as well as the original value χ, and then using Normalization is performed, and the denormalization process is: first perform training affine transformation (scaling and translation): affine =γχ norm +β, where γ and β are trainable parameters, and then the affine transformation value χ affine And all parameters are brought into the first formula for denormalization. The first formula is Reverse instance normalization is the inverse process of instance normalization, which rescales and translates the prediction results to make them as close as possible to the distribution of the original data. Then, the training step, the comparison step, the first adjustment step, the second adjustment step and the reverse normalization step are repeated at least once in sequence until the current training potential fault diagnosis label of the final model is consistent with the potential fault feature label in the training data, thereby obtaining the potential fault diagnosis model.

[0055] In order to improve the accuracy of the model output, in an optional implementation, step S2025 includes:

[0056] Step S20251, converting the above-mentioned denormalized time series data into a one-dimensional data form to obtain one-dimensional time series data, wherein the above-mentioned one-dimensional data form is the output format of the above-mentioned intermediate potential fault diagnosis model;

[0057] Step S20252, using the output layer to perform linear processing on the one-dimensional time series data to obtain the current training potential fault feature label.

[0058] In the above embodiment, the above-mentioned denormalized time series data is converted into a one-dimensional data form to obtain one-dimensional time series data to adapt to the output format of the model, and the high-dimensional features in the above-mentioned one-dimensional time series data are converted into specific prediction values, that is, the above-mentioned current training potential fault diagnosis label, and the above-mentioned high-dimensional features refer to the above-mentioned time embedding features.

[0059] In order to verify the operating performance of the potential fault diagnosis model, in an optional implementation, after the above step S2024, the method further includes:

[0060] Step S401, dividing the test set into multiple test subsets;

[0061] Step S402, an evaluation step, using a target test subset to evaluate the potential fault diagnosis model, to obtain evaluation indicators corresponding to the target test subset, the evaluation indicators including accuracy, recall, F1 score, mean square error and mean absolute error, the target test subset is any one of the test subsets;

[0062] Step S403, repeating the above evaluation step at least once until the evaluation is completed using all the above test subsets, and obtaining the above evaluation indicators corresponding to all the above test subsets;

[0063] Step S404, calculating an average evaluation index according to the above evaluation indexes corresponding to all the above test subsets;

[0064] Step S405, when the average evaluation index is greater than the set evaluation index, it is determined that the potential fault diagnosis model has passed the test;

[0065] Step S406: when the average evaluation index is greater than the set evaluation index, it is determined that the potential fault diagnosis model has not passed the test, and the potential fault diagnosis model is retrained.

[0066] In the above embodiment, the K-fold cross-validation method is used for verification, and the test set is divided into K test subsets, and one test subset is used as the verification set in turn, and the remaining K-1 test subsets are used as the training set. Through multiple iterations, the average performance of the potential fault diagnosis model on different test sets is obtained to reduce the risk of model overfitting. In each round of cross-validation, multiple evaluation indicators are used to quantify the model performance, and the average evaluation indicators of all rounds are calculated as the final evaluation result of the model, that is, the above average evaluation indicators, to ensure the stability and reliability of the model on different data sets. When the above average evaluation index is greater than the set evaluation index, it is determined that the above potential fault diagnosis model passes the test, and the above potential fault diagnosis model after the evaluation meets the requirements can be deployed. When the above average evaluation index is greater than the set evaluation index, it means that the above potential fault diagnosis model has not met the requirements and needs to be retrained. All training data can be replaced and completely retrained, or the model parameters can be adjusted and retrained. The above accuracy rate is used to measure the correct proportion of potential fault diagnosis model predictions, which directly reflects the reliability of the model in practical applications. Recall, also known as sensitivity, is used to measure the ability of a potential fault diagnosis model to identify all actual faults. A high recall means that the model can capture as many fault instances as possible and reduce missed faults. Recall = TP / (TP+FN), where TP is the number of faults that are actually faults and correctly predicted as faults, and FN is the number of faults that are actually faults but incorrectly predicted as normal. The F1 score is the harmonic mean of the precision and recall, which comprehensively evaluates the performance of the model. The mean square error is used to measure the average square error between the predicted value and the true value. The mean square error is an important regression evaluation indicator, especially suitable for time series prediction tasks, because the mean square error can reflect the accuracy of the predicted value. The mean absolute error is used to measure the average absolute error between the predicted value and the true value. Compared with the mean square error, the mean absolute error is not sensitive to outliers and is a more robust error metric.

[0067] In order to ensure the normal operation of the air conditioning system, in an optional implementation manner, after the above step S205, the method further includes:

[0068] Step S601, when there are multiple faults in the fault result, query the fault handling priority table according to the fault type to obtain the fault priority corresponding to each of the faults, the fault handling priority table is a mapping relationship between the fault type and the fault handling priority, and the fault handling priority is a priority set according to the degree of influence of each of the faults on the operation of the air conditioner;

[0069] Step S602, determining a fault processing order according to the fault priority and the fault probability corresponding to each of the faults;

[0070] Step S603: perform fault processing on the above fault according to the above fault processing sequence.

[0071] In the above embodiment, in order to ensure the overall stability of the system, the priority of fault handling is set according to the severity of the fault and the impact on the operation of the system. Priority is given to faults with greater impact. Therefore, in the case where there are multiple faults in the above fault results, the fault handling priority table is queried according to the above fault type, and the fault priority corresponding to each of the above faults existing in the current air conditioner is first determined. The fault handling order is determined according to the above fault priority corresponding to each of the above faults and the corresponding above fault probability; when the fault priority is lower than the priority threshold and the fault probability is lower than the probability threshold, the fault can be placed behind the fault handling order, and other faults with greater impact can be handled first; when the fault priority is higher than the above priority threshold or the fault probability is higher than the above probability threshold, the fault can be placed in front of the fault handling order to be handled first. Maintenance personnel can also adjust the fault handling order according to the specific air conditioning fault situation. Different processing methods are required for different types of faults, such as sensor failure, compressor failure, coolant leakage, etc. According to the fault type predicted by the potential fault diagnosis model, corresponding maintenance suggestions and guidance are provided so that corresponding maintenance and repair measures can be taken in time to ensure the normal operation of the air conditioning system. For example, a sensor failure can replace the sensor, and a compressor failure can check the compressor motor, etc. In the process of troubleshooting, it is necessary to record the process and results of each troubleshooting in detail, including the type of fault, handling measures, handling time, handling personnel, etc. Maintenance records can provide important data support for subsequent fault analysis and model optimization.

[0072] In order to improve the accuracy of the potential fault diagnosis model, in an optional implementation manner, after the above step S205, the method further includes:

[0073] Step S701, continuously acquiring the above-mentioned real-time operation data and historical fault processing results of the above-mentioned air conditioner;

[0074] Step S702, regularly re-incrementally train or fully train the potential fault diagnosis model based on the real-time operation data and historical fault processing results;

[0075] Step S703, continuously monitoring the operating performance of the potential fault diagnosis model, wherein the operating performance at least includes the accuracy rate, false alarm rate and missed alarm rate of the potential fault diagnosis of the air conditioner;

[0076] Step S704, when the operating performance is lower than the performance threshold, controlling to adjust the parameters of the potential fault diagnosis model or retraining the potential fault diagnosis model.

[0077] In the above embodiment, the operation data and fault handling results of the air-conditioning system are continuously collected to form a closed-loop feedback mechanism. Through feedback data, the model performance is continuously optimized and improved. According to the new data collected, the potential fault diagnosis model is retrained regularly. The new data is the real-time operation data and historical fault handling results. The incremental training refers to continuing to use new data for training on the basis of the already trained potential fault diagnosis model to further improve the performance of the potential fault diagnosis model. The full training refers to retraining the potential fault diagnosis model with all available data at each training to ensure that the model always maintains the best performance. In order to establish an effective feedback mechanism, the fault handling results are fed back to the model, the model performance is continuously optimized, and the prediction accuracy is improved. The operation performance of the model is continuously monitored, including indicators such as prediction accuracy, false alarm rate and missed alarm rate. When it is found that the performance drops below the performance threshold, the model parameters are adjusted in time or the model is retrained. For example, for fault types with low model prediction accuracy, the amount of training data can be increased or the model parameters can be adjusted. The fault handling records and operation data are fed back to the model training to form a closed-loop feedback. By analyzing this data, deficiencies in model predictions can be discovered and optimized accordingly to ensure that the model always maintains optimal performance. Based on the feedback data, the data acquisition and processing system is continuously improved. For example, new sensors are added or data preprocessing methods are improved to improve data quality and overall system performance.

[0078] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0079] The embodiment of the present application also provides an air conditioning fault diagnosis device. It should be noted that the air conditioning fault diagnosis device of the embodiment of the present application can be used to execute the air conditioning fault diagnosis method provided by the embodiment of the present application. The device is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware for a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0080] The following is an introduction to the air conditioning fault diagnosis device provided in the embodiment of the present application.

[0081] Figure 4is a structural block diagram of an air conditioner fault diagnosis device according to an embodiment of the present application. Figure 4 As shown, the device comprises:

[0082] The training unit 10 is used to sequentially input multiple sets of training data into the initial potential fault diagnosis model for autoregressive training using an autoregressive method to obtain an intermediate potential fault diagnosis model and time-aligned feature data, wherein each set of the above training data includes univariate time series data within the first n time steps and a potential fault feature label of the n+1th time step, wherein the above univariate time series data includes at least one of the following: indoor temperature, indoor humidity, voltage and current, and the above time-aligned feature data is data obtained after feature processing of the above univariate time series data during the above autoregressive training;

[0083] Specifically, sensors installed in air-conditioning equipment are used to collect operation data in real time, and the collected data is transmitted to the parsing server through wireless communication modules (wifi, 4G, etc.). After parsing, the data is sent to the big data platform cache service kafka, and the real-time processing engine is used to store the data in different databases (mysql, hive, kudu, hbase, etc.) according to different usage scenarios. Different processing engines (spark, MR, flink) are used on the big data platform according to different scenarios to normalize the data, eliminate the dimensional differences between different sensor data, remove outliers and missing values, ensure data quality and consistency, and ensure that all sensor data are synchronized in time, using a unified timestamp. There are many ways to normalize, for example, the maximum-minimum normalization formula: Where X max and X min They are the maximum and minimum values ​​of the data, the original data X, and the normalized data X. ′ , the inverse normalization formula can be obtained by the same logic: X = X ′ *(X max -X min )+X min . The training data is obtained by obtaining historical operation data from the database of the big data platform, extracting time series features from the historical operation data, dividing the time series data into sliding windows of fixed length, and calculating statistical features for the data in each window. Features: mean, standard deviation, maximum, minimum, median, quartile, etc. Example: For temperature sensor data, the above statistical features are calculated in each 5-minute sliding window. The autoregressive method is used to input multiple groups of training data into the initial potential fault diagnosis model in turn for autoregressive training to achieve time alignment, and obtain the intermediate potential fault diagnosis model after preliminary training and the time-aligned feature data during the training process.

[0084] An adjustment unit 20, used to adjust all model layer parameters in the intermediate potential fault diagnosis model using the time-aligned feature data to obtain a potential fault diagnosis model, wherein the model layer parameters are at least parameters of an input layer, a convolutional layer, and an output layer in the intermediate potential fault diagnosis model;

[0085] Specifically, the parameters of the model are adjusted based on time alignment to further improve the performance of the model in specific time series prediction tasks. The time alignment feature data can be used to adjust all model layer parameters in the above intermediate potential fault diagnosis model to obtain a potential fault diagnosis model.

[0086] The input-output unit 30 is used to obtain the real-time operation data of the air conditioner, extract the time series feature data in the real-time operation data, input the time series feature data into the potential fault diagnosis model, and output the current potential fault feature label;

[0087] Specifically, after the training is completed, the potential fault diagnosis model can be deployed to the air-conditioning system to perform real-time fault detection and analysis, obtain the real-time operation data of the air conditioner, and extract the time series feature data in the above real-time operation data. The purpose is to convert it into input data that can be recognized by the potential fault diagnosis model, input the above time series feature data into the above potential fault diagnosis model, and output the current potential fault feature label.

[0088] The identification processing unit 40 identifies the fault according to the above-mentioned current potential fault feature label and obtains the fault result, so as to perform corresponding fault handling measures according to the above-mentioned fault result. The above-mentioned fault result at least includes the faulty component, fault type and fault probability of the above-mentioned air conditioner.

[0089] Specifically, after the potential fault diagnosis model outputs the current potential fault feature label, the current potential fault feature label is analyzed to identify the fault, and the alarm mechanism is triggered to notify the relevant maintenance personnel. The alarm method may include SMS, email, APP push, etc. And the corresponding fault handling measures are determined according to the above fault results to execute the corresponding fault handling measures.

[0090] In this embodiment, a training unit is used to use an autoregressive method to sequentially input multiple sets of training data into an initial potential fault diagnosis model for autoregressive training to obtain an intermediate potential fault diagnosis model and time-aligned feature data, each set of the above training data includes univariate time series data within the first n time steps and a potential fault feature label of the n+1th time step, the above univariate time series data includes at least one of the following: indoor temperature, indoor humidity, voltage and current, and the above time-aligned feature data is data obtained after feature processing of the above univariate time series data during the above autoregressive training process; an adjustment unit is used to adjust the above intermediate potential fault diagnosis model using the above time-aligned feature data All model layer parameters in the model are adjusted to obtain a potential fault diagnosis model, wherein the model layer parameters are at least the parameters of the input layer, convolution layer and output layer in the intermediate potential fault diagnosis model; an input and output unit is used to obtain the real-time operation data of the air conditioner, and extract the time series feature data in the real-time operation data, input the time series feature data into the potential fault diagnosis model, and output the current potential fault feature label; an identification processing unit is used to identify the fault according to the current potential fault feature label, obtain the fault result, and execute the corresponding fault handling measures according to the fault result, wherein the fault result at least includes the faulty component, fault type and fault probability of the air conditioner. This application collects, stores and preprocesses massive time series data by using big data technology as training data, and inputs the training data into the initial potential fault diagnosis model, which is a large language model, and undergoes time series alignment processing to obtain a model after preliminary training, i.e., an intermediate potential fault diagnosis model, and fine-tunes the model parameters on the basis of the practical sequence alignment processing, and gradually adjusts the model to adapt to the complex operating environment and diverse time series data characteristics of the air conditioning system, so as to obtain the final potential fault diagnosis model. During the actual operation of the air conditioner, time series are extracted from the real-time operation data, and the time series feature data is input into the potential fault diagnosis model, and the current potential fault features are output. This not only takes advantage of the big data technology in data processing and management, but also utilizes the powerful capabilities of the large language model in data analysis and pattern recognition. It can monitor the air conditioner operation status in real time, detect and predict potential faults in a timely manner, and provide efficient fault handling measures. The present application solves the problem that the prior art air conditioner fault diagnosis mainly relies on predefined rules and thresholds for fault detection, fails to fully tap the potential of a large amount of time series data in the air conditioning system, and the fault diagnosis is not comprehensive enough.

[0091] In order for the potential fault diagnosis model to better understand and process the time series information of the air-conditioning system, in an optional implementation, the training unit includes:

[0092] A block division module is used to perform a block division step, after sequentially inputting the plurality of groups of the training data into the initial potential fault diagnosis model, divide the univariate time series data in each group of the training data into blocks according to a set time block to obtain a plurality of time blocks, wherein the set time block is a time length formed by a predetermined number of consecutive time steps;

[0093] A normalization module, used to execute a normalization step, perform instance normalization on the univariate time series data in each of the time blocks, and obtain normalized time series data;

[0094] A label coding processing module is used to execute the label coding processing step, and perform label coding processing on the normalized time series data through the convolution layer of the initial potential fault diagnosis model to maintain the semantic information of the normalized time series data and obtain label coding feature data;

[0095] A time coding processing module is used to execute the time coding processing step, and adopt a two-stage aggregation method to perform time coding processing on the above-mentioned marked coding feature data to obtain the above-mentioned time alignment feature data, and the above-mentioned time alignment feature data has a multi-scale time embedding feature;

[0096] The first repetitive module repeats the above-mentioned block division step, the above-mentioned normalization step, the above-mentioned mark coding processing step and the above-mentioned time coding processing step at least once in sequence, until the above-mentioned initial potential fault diagnosis model is trained with all the above-mentioned training data to complete the autoregressive training, and the above-mentioned intermediate potential fault diagnosis model is obtained.

[0097] In the above embodiment, a pre-trained large language model (LLM) is selected as the above initial potential fault diagnosis model, such as GPT-2. Instance normalization is applied to the input time series data to enhance the consistency and reliability of the model in processing different time series data sets, and the multivariate time series data is split into multiple univariate time series data. Each univariate time series data is input as an independent time series. Continuous time step data is grouped into a block, for example, 5 minutes as a time block, reducing the time dimension of the input sample and increasing the feature dimension. Each time block contains data points of multiple time steps. For example, if each time step is 1 second, a 5-minute time block contains 300 data points. Autoregression is used for training, that is, the initial potential fault diagnosis model predicts the data of the current time step based on the previous time step data. Instance normalization is applied to the data in each time step to obtain normalized time series data to ensure the consistency and reliability of data processing. A one-dimensional convolutional layer is used for label encoding processing of the normalized time series data to maintain the semantic information of the time series data and obtain label encoding feature data. Perform two-level aggregation on the labeled coded feature data. In the first level of aggregation, the time coding features of each time unit in the labeled coded feature data are weighted averaged to obtain the time embedding of each time unit. In the second level of aggregation, the time embeddings of adjacent time units are weighted averaged to obtain a higher level of time embedding. Use the pooling method to extract the final time embedding features. The multi-scale time embedding features can be pooled using methods such as maximum pooling and average pooling to obtain the above-mentioned time alignment feature data. Through such a two-level aggregation and pooling method, multi-scale time embedding features can be extracted, thereby better representing the time information and providing richer time information for subsequent tasks. Repeat the above-mentioned block division step, the above-mentioned normalization step, the above-mentioned label coding processing step and the above-mentioned time coding processing step at least once in sequence until the above-mentioned initial potential fault diagnosis model is completed with autoregressive training using all of the above-mentioned training data, so that the final potential fault diagnosis model can better understand and process time series information.

[0098] In order to efficiently process the time series data of the air conditioning system and improve the accuracy and efficiency of fault diagnosis, in an optional implementation manner, the adjustment unit includes:

[0099] A training module, used to execute a training step, using the time-aligned feature data to perform linear detection training on the parameters of the output layer in the intermediate potential fault diagnosis model, so as to preliminarily adjust the parameters of the output layer and obtain a preliminary training potential fault feature label;

[0100] An adjustment module, used for executing the adjustment step, adjusting all the model layer parameters in the intermediate potential fault diagnosis model by using the parameter efficient fine-tuning technology;

[0101] A denormalization module is used to perform a denormalization step, denormalize the above-mentioned preliminary training potential fault feature labels to obtain denormalized time series data, and perform dimension conversion on the above-mentioned denormalized time series data to obtain current training potential fault feature labels;

[0102] The second repetition module repeats the above training step, the above adjustment step and the denormalization step at least once in sequence until the above current training potential fault diagnosis label is consistent with the above potential fault feature label in the above training data, thereby obtaining the above potential fault diagnosis model.

[0103] In the above embodiment, after completing the time series alignment, the prediction fine-tuning stage is entered. The main goal of this stage is to use the time series feature data that has been aligned and encoded to comprehensively fine-tune the intermediate potential fault diagnosis model to improve the accuracy of air conditioning fault diagnosis and prediction. First, only the parameters of the output layer are trained, and the time series feature data after the model is aligned is used for the prediction task of the training target. Training only the parameters of the output layer enables the model to quickly adapt to new prediction tasks while retaining most of the model's feature extraction capabilities. After the initial adaptation, the parameters of all layers are unfrozen and comprehensively fine-tuned to make the model better adapt to the time series prediction task, and all the above model layer parameters in the above intermediate potential fault diagnosis model are adjusted by using efficient parameter fine-tuning technology. In order to ensure that the data corresponding to the model output is consistent with the actual data distribution, the above-mentioned preliminary training potential fault feature labels need to be denormalized to obtain denormalized time series data. Normalization is achieved by calculating the mean μ and standard deviation σ of each feature as well as the original value χ, and then using Normalization is performed, and the denormalization process is as follows: first perform training affine transformation (scaling and translation): affine =γχ norm +β, where γ and β are trainable parameters, and then the affine transformation value χ affine And all parameters are brought into the first formula for denormalization. The first formula is Reverse instance normalization is the inverse process of instance normalization, which rescales and translates the prediction results to make them as close as possible to the distribution of the original data. Then, the training step, the comparison step, the first adjustment step, the second adjustment step and the reverse normalization step are repeated at least once in sequence until the current training potential fault diagnosis label of the final model is consistent with the potential fault feature label in the training data, thereby obtaining the potential fault diagnosis model.

[0104] In order to improve the accuracy of the model output, in an optional implementation, the denormalization module includes:

[0105] A first conversion submodule converts the denormalized time series data into a one-dimensional data form to obtain one-dimensional time series data, wherein the one-dimensional data form is an output format of the intermediate potential fault diagnosis model;

[0106] The second conversion submodule uses the output layer to perform linear processing on the one-dimensional time series data to obtain the current training potential fault feature label.

[0107] In the above embodiment, the above-mentioned denormalized time series data is converted into a one-dimensional data form to obtain one-dimensional time series data to adapt to the output format of the model, and the high-dimensional features in the above-mentioned one-dimensional time series data are converted into specific prediction values, that is, the above-mentioned current training potential fault diagnosis label, and the above-mentioned high-dimensional features refer to the above-mentioned time embedding features.

[0108] In order to verify the operating performance of the potential fault diagnosis model, in an optional implementation, the device further includes:

[0109] A division unit, used for adjusting all model layer parameters in the intermediate potential fault diagnosis model by using the time alignment feature data to obtain the potential fault diagnosis model, and then dividing the test set into a plurality of test subsets;

[0110] An evaluation unit, used in the evaluation step, to evaluate the potential fault diagnosis model using a target test subset to obtain evaluation indicators corresponding to the target test subset, wherein the evaluation indicators include accuracy, recall, F1 score, mean square error and mean absolute error, and the target test subset is any one of the test subsets;

[0111] A repeating unit, used to repeat the above evaluation step at least once until the evaluation is completed using all the above test subsets, and obtain the above evaluation indicators corresponding to all the above test subsets;

[0112] A calculation unit, used for calculating an average evaluation index according to the above evaluation indexes corresponding to all the above test subsets;

[0113] A first determination unit, configured to determine that the potential fault diagnosis model passes the test when the average evaluation index is greater than a set evaluation index;

[0114] The second determination unit is used to determine that the potential fault diagnosis model fails the test and retrain the potential fault diagnosis model when the average evaluation index is greater than the set evaluation index.

[0115] In the above embodiment, the K-fold cross-validation method is used for verification, and the test set is divided into K test subsets, and one test subset is used as the verification set in turn, and the remaining K-1 test subsets are used as the training set. Through multiple iterations, the average performance of the potential fault diagnosis model on different test sets is obtained to reduce the risk of model overfitting. In each round of cross-validation, multiple evaluation indicators are used to quantify the model performance, and the average evaluation indicators of all rounds are calculated as the final evaluation result of the model, that is, the above average evaluation indicators, to ensure the stability and reliability of the model on different data sets. When the above average evaluation index is greater than the set evaluation index, it is determined that the above potential fault diagnosis model passes the test, and the above potential fault diagnosis model after the evaluation meets the requirements can be deployed. When the above average evaluation index is greater than the set evaluation index, it means that the above potential fault diagnosis model has not met the requirements and needs to be retrained. All training data can be replaced and completely retrained, or the model parameters can be adjusted and retrained. The above accuracy rate is used to measure the correct proportion of potential fault diagnosis model predictions, which directly reflects the reliability of the model in practical applications. Recall, also known as sensitivity, is used to measure the ability of a potential fault diagnosis model to identify all actual faults. A high recall means that the model can capture as many fault instances as possible and reduce missed faults. Recall = TP / (TP+FN), where TP is the number of faults that are actually faults and correctly predicted as faults, and FN is the number of faults that are actually faults but incorrectly predicted as normal. The F1 score is the harmonic mean of the precision and recall, which comprehensively evaluates the performance of the model. The mean square error is used to measure the average square error between the predicted value and the true value. The mean square error is an important regression evaluation indicator, especially suitable for time series prediction tasks, because the mean square error can reflect the accuracy of the predicted value. The mean absolute error is used to measure the average absolute error between the predicted value and the true value. Compared with the mean square error, the mean absolute error is not sensitive to outliers and is a more robust error metric.

[0116] In order to ensure the normal operation of the air conditioning system, in an optional implementation manner, the device further includes:

[0117] a query unit, for, after identifying the fault according to the current potential fault feature label and obtaining the fault result, if there are multiple faults in the fault result, querying the fault handling priority table according to the fault type to obtain the fault priority corresponding to each of the above faults, wherein the fault handling priority table is a mapping relationship between the fault type and the fault handling priority, and the fault handling priority is a priority set according to the degree of influence of each of the above faults on the operation of the air conditioner when the fault occurs;

[0118] A determination unit, configured to determine a fault processing order according to the fault priority level and the corresponding fault probability corresponding to each of the faults;

[0119] The fault handling unit is used to handle the fault according to the fault handling sequence.

[0120] In the above embodiment, in order to ensure the overall stability of the system, the priority of fault handling is set according to the severity of the fault and the impact on the operation of the system. Priority is given to faults with greater impact. Therefore, in the case where there are multiple faults in the above fault results, the fault handling priority table is queried according to the above fault type, and the fault priority corresponding to each of the above faults existing in the current air conditioner is first determined. The fault handling order is determined according to the above fault priority corresponding to each of the above faults and the corresponding above fault probability; when the fault priority is lower than the priority threshold and the fault probability is lower than the probability threshold, the fault can be placed behind the fault handling order, and other faults with greater impact can be handled first; when the fault priority is higher than the above priority threshold or the fault probability is higher than the above probability threshold, the fault can be placed in front of the fault handling order to be handled first. Maintenance personnel can also adjust the fault handling order according to the specific air conditioning fault situation. Different processing methods are required for different types of faults, such as sensor failure, compressor failure, coolant leakage, etc. According to the fault type predicted by the potential fault diagnosis model, corresponding maintenance suggestions and guidance are provided so that corresponding maintenance and repair measures can be taken in time to ensure the normal operation of the air conditioning system. For example, a sensor failure can replace the sensor, and a compressor failure can check the compressor motor, etc. In the process of troubleshooting, it is necessary to record the process and results of each troubleshooting in detail, including the type of fault, handling measures, handling time, handling personnel, etc. Maintenance records can provide important data support for subsequent fault analysis and model optimization.

[0121] In order to improve the accuracy of the potential fault diagnosis model, in an optional implementation manner, the device further includes:

[0122] an acquisition unit, configured to continuously acquire the real-time operation data and historical fault processing results of the air conditioner after identifying the fault according to the potential fault feature label and obtaining the fault result;

[0123] A retraining unit, used for periodically retraining the potential fault diagnosis model in increments or in full according to the real-time operation data and historical fault processing results;

[0124] A monitoring unit, used to continuously monitor the operating performance of the potential fault diagnosis model, wherein the operating performance at least includes the accuracy, false alarm rate and missed alarm rate of the potential fault diagnosis of the air conditioner;

[0125] A control unit is used to control the adjustment of parameters of the potential fault diagnosis model or retrain the potential fault diagnosis model when the operating performance is lower than a performance threshold.

[0126] In the above embodiment, the operation data and fault handling results of the air-conditioning system are continuously collected to form a closed-loop feedback mechanism. Through feedback data, the model performance is continuously optimized and improved. According to the new data collected, the potential fault diagnosis model is retrained regularly. The new data is the real-time operation data and historical fault handling results. The incremental training refers to continuing to use new data for training on the basis of the already trained potential fault diagnosis model to further improve the performance of the potential fault diagnosis model. The full training refers to retraining the potential fault diagnosis model with all available data at each training to ensure that the model always maintains the best performance. In order to establish an effective feedback mechanism, the fault handling results are fed back to the model, the model performance is continuously optimized, and the prediction accuracy is improved. The operation performance of the model is continuously monitored, including indicators such as prediction accuracy, false alarm rate and missed alarm rate. When it is found that the performance drops below the performance threshold, the model parameters are adjusted in time or the model is retrained. For example, for fault types with low model prediction accuracy, the amount of training data can be increased or the model parameters can be adjusted. The fault handling records and operation data are fed back to the model training to form a closed-loop feedback. By analyzing this data, deficiencies in model predictions can be discovered and optimized accordingly to ensure that the model always maintains optimal performance. Based on the feedback data, the data acquisition and processing system is continuously improved. For example, new sensors are added or data preprocessing methods are improved to improve data quality and overall system performance.

[0127] The air conditioner fault diagnosis device includes a processor and a memory. The training unit, adjustment unit, input and output unit, etc. are stored in the memory as program units, and the processor executes the program units stored in the memory to implement corresponding functions. The modules are all located in the same processor; or, the modules are located in different processors in any combination.

[0128] The processor includes a kernel, which calls the corresponding program unit from the memory. One or more kernels can be set, and the kernel parameters can be adjusted to solve the problem that the air conditioner fault diagnosis in the prior art mainly relies on predefined rules and thresholds for fault detection, fails to fully tap the potential of a large amount of time series data in the air conditioner system, and the comprehensiveness of the fault diagnosis is insufficient.

[0129] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0130] An embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the air conditioning fault diagnosis method.

[0131] Specifically, the air conditioner fault diagnosis method includes:

[0132] Step S201, using an autoregressive method to sequentially input multiple sets of training data into an initial potential fault diagnosis model for autoregressive training, to obtain an intermediate potential fault diagnosis model and time-aligned feature data, each set of the above training data includes univariate time series data within the first n time steps and a potential fault feature label of the n+1th time step, the above univariate time series data includes at least one of the following: indoor temperature, indoor humidity, voltage and current, and the above time-aligned feature data is data obtained after feature processing of the above univariate time series data during the above autoregressive training;

[0133] Step S202, using the time-aligned feature data to adjust all model layer parameters in the intermediate potential fault diagnosis model to obtain a potential fault diagnosis model, wherein the model layer parameters are at least parameters of an input layer, a convolutional layer, and an output layer in the intermediate potential fault diagnosis model;

[0134] Step S203, acquiring real-time operation data of the air conditioner, extracting time series feature data from the real-time operation data, inputting the time series feature data into the potential fault diagnosis model, and outputting a current potential fault feature label;

[0135] Step S204, identifying the fault according to the above-mentioned current potential fault feature label, obtaining the fault result, and executing corresponding fault handling measures according to the above-mentioned fault result, and the above-mentioned fault result at least includes the fault component, fault type and fault probability of the above-mentioned air conditioner.

[0136] An embodiment of the present invention provides a processor, and the processor is used to run a program, wherein the air conditioner fault diagnosis method is executed when the program is run.

[0137] Specifically, the air conditioner fault diagnosis method includes:

[0138] Step S201, using an autoregressive method to sequentially input multiple sets of training data into an initial potential fault diagnosis model for autoregressive training, to obtain an intermediate potential fault diagnosis model and time-aligned feature data, each set of the above training data includes univariate time series data within the first n time steps and a potential fault feature label of the n+1th time step, the above univariate time series data includes at least one of the following: indoor temperature, indoor humidity, voltage and current, and the above time-aligned feature data is data obtained after feature processing of the above univariate time series data during the above autoregressive training;

[0139] Step S202, using the time-aligned feature data to adjust all model layer parameters in the intermediate potential fault diagnosis model to obtain a potential fault diagnosis model, wherein the model layer parameters are at least parameters of an input layer, a convolutional layer, and an output layer in the intermediate potential fault diagnosis model;

[0140] Step S203, acquiring real-time operation data of the air conditioner, extracting time series feature data from the real-time operation data, inputting the time series feature data into the potential fault diagnosis model, and outputting a current potential fault feature label;

[0141] Step S204, identifying the fault according to the above-mentioned current potential fault feature label, obtaining the fault result, and executing corresponding fault handling measures according to the above-mentioned fault result, and the above-mentioned fault result at least includes the fault component, fault type and fault probability of the above-mentioned air conditioner.

[0142] An embodiment of the present invention provides an air conditioning fault diagnosis system, the air conditioning fault diagnosis system comprising a processor, a memory, and a program stored in the memory and executable on the processor, and when the processor executes the program, at least the following steps are implemented:

[0143] Step S201, using an autoregressive method to sequentially input multiple sets of training data into an initial potential fault diagnosis model for autoregressive training, to obtain an intermediate potential fault diagnosis model and time-aligned feature data, each set of the above training data includes univariate time series data within the first n time steps and a potential fault feature label of the n+1th time step, the above univariate time series data includes at least one of the following: indoor temperature, indoor humidity, voltage and current, and the above time-aligned feature data is data obtained after feature processing of the above univariate time series data during the above autoregressive training;

[0144] Step S202, using the time-aligned feature data to adjust all model layer parameters in the intermediate potential fault diagnosis model to obtain a potential fault diagnosis model, wherein the model layer parameters are at least parameters of an input layer, a convolutional layer, and an output layer in the intermediate potential fault diagnosis model;

[0145] Step S203, acquiring real-time operation data of the air conditioner, extracting time series feature data from the real-time operation data, inputting the time series feature data into the potential fault diagnosis model, and outputting a current potential fault feature label;

[0146] Step S204, identifying the fault according to the above-mentioned current potential fault feature label, obtaining the fault result, and executing corresponding fault handling measures according to the above-mentioned fault result, and the above-mentioned fault result at least includes the fault component, fault type and fault probability of the above-mentioned air conditioner.

[0147] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing a program for initializing at least the following method steps:

[0148] Step S201, using an autoregressive method to sequentially input multiple sets of training data into an initial potential fault diagnosis model for autoregressive training, to obtain an intermediate potential fault diagnosis model and time-aligned feature data, each set of the above training data includes univariate time series data within the first n time steps and a potential fault feature label of the n+1th time step, the above univariate time series data includes at least one of the following: indoor temperature, indoor humidity, voltage and current, and the above time-aligned feature data is data obtained after feature processing of the above univariate time series data during the above autoregressive training;

[0149] Step S202, using the time-aligned feature data to adjust all model layer parameters in the intermediate potential fault diagnosis model to obtain a potential fault diagnosis model, wherein the model layer parameters are at least parameters of an input layer, a convolutional layer, and an output layer in the intermediate potential fault diagnosis model;

[0150] Step S203, acquiring real-time operation data of the air conditioner, extracting time series feature data from the real-time operation data, inputting the time series feature data into the potential fault diagnosis model, and outputting a current potential fault feature label;

[0151] Step S204, identifying the fault according to the above-mentioned current potential fault feature label, obtaining the fault result, and executing corresponding fault handling measures according to the above-mentioned fault result, and the above-mentioned fault result at least includes the fault component, fault type and fault probability of the above-mentioned air conditioner.

[0152] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0153] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0154] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0155] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0156] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0157] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0158] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0159] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0160] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0161] From the above description, it can be seen that the above embodiments of the present application achieve the following technical effects:

[0162] 1) The air conditioning fault diagnosis method of the present application firstly adopts an autoregressive method to sequentially input multiple sets of training data into an initial potential fault diagnosis model for autoregressive training to obtain an intermediate potential fault diagnosis model and time-aligned feature data, wherein each set of the above training data comprises univariate time series data within the first n time steps and a potential fault feature label of the n+1th time step, wherein the above univariate time series data comprises at least one of the following: indoor temperature, indoor humidity, voltage and current, and the above time-aligned feature data is data obtained after feature processing of the above univariate time series data during the above autoregressive training; then, the above time-aligned feature data is used to diagnose the intermediate potential fault diagnosis model and the time-aligned feature data. All model layer parameters in the intermediate potential fault diagnosis model are adjusted to obtain the potential fault diagnosis model, and the above model layer parameters are at least the parameters of the input layer, convolution layer and output layer in the above intermediate potential fault diagnosis model; then, the real-time operation data of the air conditioner is obtained, and the time series feature data in the above real-time operation data is extracted, and the above time series feature data is input into the above potential fault diagnosis model, and the current potential fault feature label is output; finally, the fault is identified according to the above current potential fault feature label, and the fault result is obtained, so as to perform the corresponding fault handling measures according to the above fault result, and the above fault result at least includes the faulty component, fault type and fault probability of the above air conditioner. This application collects, stores and preprocesses massive time series data by using big data technology as training data, and inputs the training data into the initial potential fault diagnosis model, which is a large language model, and undergoes time series alignment processing to obtain a model after preliminary training, that is, the intermediate potential fault diagnosis model, and fine-tunes the model parameters on the basis of the practical sequence alignment processing, and gradually adjusts the model to adapt to the complex operating environment and diverse time series data characteristics of the air conditioning system, so as to obtain the final potential fault diagnosis model. During the actual operation of the air conditioner, time series are extracted from the real-time operation data, and the time series feature data is input into the potential fault diagnosis model, and the current potential fault features are output. This not only takes advantage of the big data technology in data processing and management, but also utilizes the powerful capabilities of the large language model in data analysis and pattern recognition. It can monitor the air conditioner operation status in real time, detect and predict potential faults in a timely manner, and provide efficient fault handling measures. The present application solves the problem that the prior art air conditioner fault diagnosis mainly relies on predefined rules and thresholds for fault detection, fails to fully tap the potential of a large amount of time series data in the air conditioning system, and the fault diagnosis is not comprehensive enough.

[0163] 2) The air conditioning fault diagnosis device of the present application, the training unit, is used to use the autoregressive method to sequentially input multiple sets of training data into the initial potential fault diagnosis model for autoregressive training to obtain an intermediate potential fault diagnosis model and time-aligned feature data, each set of the above training data includes univariate time series data within the first n time steps and potential fault feature labels of the n+1th time step, the above univariate time series data includes at least one of the following: indoor temperature, indoor humidity, voltage and current, the above time-aligned feature data is the data obtained after feature processing of the above univariate time series data during the above autoregressive training process; the adjustment unit is used to adjust the above intermediate potential fault diagnosis model using the above time-aligned feature data In the fault diagnosis model, all model layer parameters are adjusted to obtain a potential fault diagnosis model, wherein the model layer parameters are at least the parameters of the input layer, convolution layer and output layer in the intermediate potential fault diagnosis model; an input and output unit is used to obtain the real-time operation data of the air conditioner, and extract the time series feature data in the real-time operation data, input the time series feature data into the potential fault diagnosis model, and output the current potential fault feature label; an identification processing unit is used to identify the fault according to the current potential fault feature label, obtain the fault result, and execute the corresponding fault handling measures according to the fault result, wherein the fault result at least includes the faulty component, fault type and fault probability of the air conditioner. This application collects, stores and preprocesses massive time series data by using big data technology as training data, and inputs the training data into the initial potential fault diagnosis model, which is a large language model, and undergoes time series alignment processing to obtain a model after preliminary training, i.e., an intermediate potential fault diagnosis model, and fine-tunes the model parameters on the basis of the practical sequence alignment processing, and gradually adjusts the model to adapt to the complex operating environment and diverse time series data characteristics of the air conditioning system, so as to obtain the final potential fault diagnosis model. During the actual operation of the air conditioner, time series are extracted from the real-time operation data, and the time series feature data is input into the potential fault diagnosis model, and the current potential fault features are output. This not only takes advantage of the big data technology in data processing and management, but also utilizes the powerful capabilities of the large language model in data analysis and pattern recognition. It can monitor the air conditioner operation status in real time, detect and predict potential faults in a timely manner, and provide efficient fault handling measures. The present application solves the problem that the prior art air conditioner fault diagnosis mainly relies on predefined rules and thresholds for fault detection, fails to fully tap the potential of a large amount of time series data in the air conditioning system, and the fault diagnosis is not comprehensive enough.

[0164] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for diagnosing air conditioner faults, characterized in that: include: An autoregressive method is used to sequentially input multiple groups of training data into an initial potential fault diagnosis model for autoregressive training, to obtain an intermediate potential fault diagnosis model and time-aligned feature data, each group of the training data includes univariate time series data within the first n time steps and a potential fault feature label of the n+1th time step, the univariate time series data includes at least one of the following: indoor temperature, indoor humidity, voltage and current, and the time-aligned feature data is data obtained after feature processing of the univariate time series data during the autoregressive training; Using the time-aligned feature data, all model layer parameters in the intermediate potential fault diagnosis model are adjusted to obtain a potential fault diagnosis model, wherein the model layer parameters are at least parameters of an input layer, a convolutional layer, and an output layer in the intermediate potential fault diagnosis model; Acquire real-time operation data of the air conditioner, extract time series feature data from the real-time operation data, input the time series feature data into the potential fault diagnosis model, and output a current potential fault feature label; The fault is identified according to the current potential fault feature label to obtain a fault result, so as to perform corresponding fault handling measures according to the fault result, and the fault result at least includes the fault component, fault type and fault probability of the air conditioner.

2. The method according to claim 1, characterized in that The autoregressive method is used to sequentially input multiple sets of training data into the initial potential fault diagnosis model for autoregressive training, and an intermediate potential fault diagnosis model and time-aligned feature data are obtained, including: A block division step, after inputting the plurality of groups of training data into the initial potential fault diagnosis model in sequence, the univariate time series data in each group of the training data is divided into blocks according to a set time block to obtain a plurality of time blocks, wherein the set time block is a time length formed by a predetermined number of continuous time steps; a normalization step of performing instance normalization on the univariate time series data in each of the time blocks to obtain normalized time series data; a label coding processing step, performing label coding processing on the normalized time series data through the convolution layer of the initial potential fault diagnosis model to maintain the semantic information of the normalized time series data and obtain label coding feature data; A time coding processing step, using a two-stage aggregation method to perform time coding processing on the marked coded feature data to obtain the time alignment feature data, wherein the time alignment feature data has a multi-scale time embedding feature; The blocking step, the normalization step, the label encoding processing step and the time encoding processing step are sequentially repeated at least once until the autoregressive training of the initial potential fault diagnosis model is completed using all the training data to obtain the intermediate potential fault diagnosis model.

3. The method according to claim 2, characterized in that The time-aligned feature data is used to adjust all model layer parameters in the intermediate potential fault diagnosis model to obtain a potential fault diagnosis model, including: A training step, using the time-aligned feature data to perform linear detection training on the parameters of the output layer in the intermediate potential fault diagnosis model to preliminarily adjust the parameters of the output layer to obtain a preliminarily trained potential fault feature label; An adjustment step, using a parameter efficient fine-tuning technique to adjust all model layer parameters in the intermediate potential fault diagnosis model; A denormalization step, performing denormalization processing on the preliminary training potential fault feature labels to obtain denormalized time series data, and performing dimension conversion on the denormalized time series data to obtain current training potential fault feature labels; The training step, the adjustment step and the denormalization step are sequentially repeated at least once until the current training potential fault diagnosis label is consistent with the potential fault feature label in the training data, thereby obtaining the potential fault diagnosis model.

4. The method according to claim 3, characterized in that The denormalized time series data is dimensionally transformed to obtain the current training potential fault diagnosis label, including: Converting the denormalized time series data into a one-dimensional data form to obtain one-dimensional time series data, wherein the one-dimensional data form is an output format of the intermediate potential fault diagnosis model; The one-dimensional time series data is linearly processed using the output layer to obtain the current training potential fault feature label.

5. The method according to claim 1, characterized in that After adjusting all model layer parameters in the intermediate potential fault diagnosis model by using the time alignment feature data to obtain the potential fault diagnosis model, the method further includes: Divide the test set into multiple test subsets; An evaluation step, using a target test subset to evaluate the potential fault diagnosis model, to obtain evaluation indicators corresponding to the target test subset, the evaluation indicators including accuracy, recall, F1 score, mean square error and mean absolute error, the target test subset being any one of the test subsets; Repeat the evaluation step at least once until the evaluation is completed using all the test subsets to obtain the evaluation indicators corresponding to all the test subsets; Calculate an average evaluation index according to the evaluation indexes corresponding to all the test subsets; When the average evaluation index is greater than the set evaluation index, determining that the potential fault diagnosis model passes the test; When the average evaluation index is greater than the set evaluation index, it is determined that the potential fault diagnosis model has failed the test, and the potential fault diagnosis model is retrained.

6. The method according to claim 1, characterized in that After identifying the fault according to the current potential fault feature label and obtaining the fault result, the method further includes: In the case where there are multiple faults in the fault result, a fault handling priority table is queried according to the fault type to obtain the fault priority corresponding to each fault, wherein the fault handling priority table is a mapping relationship between the fault type and the fault handling priority, and the fault handling priority is a priority set according to the degree of influence of each fault on the operation of the air conditioner when it occurs; Determine a fault processing order according to the fault priority and the corresponding fault probability corresponding to each of the faults; The fault is processed according to the fault processing sequence.

7. The method according to any one of claims 1 to 6, characterized in that: After identifying the fault according to the potential fault feature label and obtaining the fault result, the method further includes: Continuously acquiring the real-time operation data and historical fault handling results of the air conditioner; Regularly retrain the potential fault diagnosis model incrementally or fully based on the real-time operation data and historical fault processing results; Continuously monitoring the operating performance of the potential fault diagnosis model, wherein the operating performance at least includes the accuracy rate, false alarm rate and missed alarm rate of the potential fault diagnosis of the air conditioner; In a case where the operating performance is lower than a performance threshold, control is performed to adjust parameters of the potential fault diagnosis model or to retrain the potential fault diagnosis model.

8. An air conditioner fault diagnosis device, characterized in that: The device comprises: A training unit, used for sequentially inputting a plurality of sets of training data into an initial potential fault diagnosis model for autoregressive training using an autoregressive method, to obtain an intermediate potential fault diagnosis model and time-aligned feature data, wherein each set of the training data comprises univariate time series data within the first n time steps and a potential fault feature label of the n+1th time step, wherein the univariate time series data comprises at least one of the following: indoor temperature, indoor humidity, voltage and current, and the time-aligned feature data is data obtained after feature processing is performed on the univariate time series data during the autoregressive training process; An adjustment unit, used to adjust all model layer parameters in the intermediate potential fault diagnosis model using the time-aligned feature data to obtain a potential fault diagnosis model, wherein the model layer parameters are at least parameters of an input layer, a convolutional layer, and an output layer in the intermediate potential fault diagnosis model; An input-output unit, used to obtain real-time operation data of the air conditioner, extract time series feature data from the real-time operation data, input the time series feature data into the potential fault diagnosis model, and output a current potential fault feature label; An identification and processing unit is used to identify the fault according to the current potential fault feature label and obtain a fault result, so as to execute corresponding fault handling measures according to the fault result, and the fault result at least includes the faulty component, fault type and fault probability of the air conditioner.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the method according to any one of claims 1 to 7.

10. An air conditioning fault diagnosis system, characterized in that: include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include methods for executing any one of claims 1 to 7.

Citation Information

Cited By

  • Mine card air conditioner fault diagnosis method and system based on data analysis

    CN120800824A

  • A Data-Based Method and System for Fault Diagnosis of Air Conditioners in Mining Trucks

    CN120800824B

  • Air conditioner operation fault alarm method and system

    CN121388713A