CN114941890A - Central air conditioner fault diagnosis method and system based on image and depth blurring - Google Patents

Abstract

The invention proposes a fault diagnosis method and system for a central air conditioner based on images and depth blur. The system mainly includes four modules: a data processing module, a digital-to-map conversion module, a residual depth fuzzy module, a fault diagnosis module, and a data processing module. Perform feature extraction and feature sorting on the dataset; the digital-to-map conversion module converts the feature-extracted dataset into a corresponding two-dimensional grayscale image, and uses the sliding window method to generate a rich image dataset; the residual depth blur module uses Residual deep fuzzy models are trained on a 2D grayscale image dataset for fault diagnosis. The real-time fault diagnosis module of the central air conditioner can collect data through multiple sensors and input it into the fault diagnosis model to judge whether the fault exists or not. The invention adopts the strategy of combining the kernel slow feature analysis algorithm, the digital image conversion algorithm and the residual deep fuzzy model to construct the fault diagnosis model of the central air conditioner, and can carry out the fault diagnosis efficiently and accurately.

Description

A fault diagnosis method and system for central air conditioner based on image and depth blur

technical field

The invention belongs to the field of air-conditioning fault diagnosis, and in particular relates to a central air-conditioning fault diagnosis method and system based on images and depth blur.

Background technique

Central air conditioners can provide a comfortable indoor environment by adjusting the air temperature and humidity in different rooms and areas. Due to many components, complex and changeable operating environment, improper operation or corrosion by natural factors, various failures of central air conditioners may occur. The long-term operation of the air conditioner in the fault state will not only shorten the service life of the equipment, but also cause serious energy waste. Therefore, it is necessary to explore an efficient and accurate central air-conditioning fault diagnosis method and system.

The central air conditioner presents a variety of operating modes with the change of the external environment, with strong temporal dynamic characteristics and nonlinear characteristics, and the correlation between various characteristic variables shows spatial correlation characteristics. The existing central air-conditioning fault diagnosis methods generally do not consider the above characteristics, and there are the following problems: (1) The unprocessed original data set is used to establish the fault diagnosis model, and the characteristic variables are complex and the fault characteristics are not obvious; (2) The numerical value is mainly used. The data builds a model, which destroys the spatial correlation between each characteristic variable; (3) The structure of the fault diagnosis model is complex and not interpretable.

SUMMARY OF THE INVENTION

In order to solve the above-mentioned problems in the prior art, a method and system for diagnosing faults of a central air conditioner based on images and depth blur are provided.

The technical scheme adopted by the present invention to solve its technical problems is:

This technical solution proposes a fault diagnosis method for central air conditioners based on images and depth blur, including the following steps:

Step (1): Data Processing

Collect and label the data of the normal operation of the central air conditioner and various faulty operations; preprocess the collected data; establish a kernel slow feature analysis model to extract and sort the features of the data, and obtain a kernel slow feature data set;

Step (2): Digital map conversion

Convert the kernel slow feature dataset after feature extraction into the corresponding two-dimensional grayscale image, and use the sliding window method to generate a rich image dataset;

Step (3): Build and train a residual deep fuzzy model

Establish a residual depth fuzzy model; through the acquired two-dimensional grayscale image data set, train the residual deep fuzzy model for fault diagnosis;

Step (4): Central Air Conditioning Fault Diagnosis

Measure and collect the data during the operation of the central air conditioner, obtain the newly collected data, perform steps 1 and 2 on the newly collected data, and input the processed data into the established residual depth fuzzy model, and carry out fault detection and verification. No judgment.

其包含N天的数据，并且每天的数据有T个时刻样本和P个特征变量。因此，某一天的数据可以表示为

其中t∈[1,2,...,T]表示时间范围，

为连续T个时间序列中第p个特征变量对应的值；数据采集完成之后，将数据标注上对应的运行工况标签。In the step (1), the specific method of data collection and labeling is as follows: by installing sensors in multiple positions of the central air conditioner, the data of normal operation and various faulty operations are collected; the data collection is in units of one day, and the sampling time interval is One minute; represent the collected raw data set as

It contains N days of data, and each day's data has T time samples and P feature variables. Therefore, the data for a certain day can be expressed as

where t∈[1,2,...,T] denotes the time range,

is the value corresponding to the p-th characteristic variable in consecutive T time series; after the data collection is completed, the data is marked with the corresponding operating condition label.

In described step (1), the concrete method of data preprocessing is:

Use the Z-score algorithm to normalize the original data set X in units of each day, as follows:

Among them, mean(x _p (t)) is the mean of x _p (t), and std(x _p (t)) is the standard deviation of x _p (t);

其中

由此，对每一天的数据均进行标准化处理后可得到数据集

The normalized data for a day is defined as

in

Thus, the data set can be obtained by normalizing the data of each day

映射到高维数

然后，通过求解下述优化问题获得慢特征：In the step (1), the specific method of feature extraction is: use the normalized normal operation data to determine the optimal parameters, and establish a kernel slow feature analysis model; first, use the implicit nonlinear mapping function φ(·) to convert the data into

map to higher dimensions

Then, slow features are obtained by solving the following optimization problem:

是y_j(t)对时间变量t的一阶导数，<.>定义为

K为采样间隔；where the output data y _j (t) is the jth slow feature extracted from the input data x _φ (t),

is the first derivative of y _j (t) with respect to the time variable t, <.> is defined as

K is the sampling interval;

基于训练完成的核慢特征分析模型，从所有故障数据集中提取出变化缓慢的特征，得到N天核慢特征数据集，记为

Furthermore, the data after feature extraction is expressed as

Based on the trained kernel slow feature analysis model, the slowly changing features are extracted from all fault data sets, and the N-day kernel slow feature data set is obtained, denoted as

In the described step (2), the concrete method of digital image conversion is:

① Perform Min-Max normalization on the data of each day in the kernel slow feature dataset Y after feature extraction and multiply the output result by 255. The formula is as follows:

where min(y _η (t)) and max(y _η (t)) are the minimum and maximum values of the y _η (t) (η=1,2,...,R) vector, respectively, and will obtain The N-day dataset is denoted as

② Convert the data set Z obtained in the above steps into a corresponding two-dimensional grayscale image; in the image, arrange the feature variables according to the slow change degree of the feature variables; specifically, convert the slowest-changing feature variable into the image The pixels in one column are converted into pixels in the second column of the image, and the second slowest-changing feature variable is converted to obtain the converted two-dimensional grayscale image; wherein, the data-to-image conversion is performed by converting the input The data is converted into an unsigned 8 integer type, which is then performed using the data format conversion instruction; the larger the value, the darker the grayscale;

③Using the sliding window method to expand the generated image dataset; set the lag parameter to L, then the first image is generated from the data from the Mth row to (M+L-1) row; set the sliding distance of the sliding window If it is q, start intercepting from (M+q) line of data, and generate the second image at the end of (M+L+q-1) line of data, and so on; if the data in one day has a total of R characteristic variables and T For each sample, repeating the above operation can obtain (T-l)/q+1 images, all of which are L×R in size.

The residual depth fuzzy model in the step (3) includes an input layer, a hidden layer, and an output layer; the model is realized by stacking fuzzy inference modules in a bottom-up, layer-by-layer manner; the details of the model are The structure and principle are as follows:

1) Preliminarily obtain the input image information of each fault type from the input layer, and pass it to the hidden layer through the nodes of the equivalent mapping;

2) In the hidden layer, the fuzzy inference modules are stacked layer by layer, with a total of s layers; there are s fuzzy inference modules in the first hidden layer, and (s-1) fuzzy inference modules in the second hidden layer, and so on. There is only one fuzzy inference module in the sth hidden layer;

3) In the hidden layer, the output variables y ^l of all fuzzy inference modules in the previous layer and the residuals of the expected results are weighted as the input of the next layer of fuzzy inference modules, and the output of the first layer is:

为输出向量，

代表权重向量；Among them, l=1, 2,...,s, ε _l-1 represents the output of the layer,

is the output vector,

represents the weight vector;

得到各模糊推理模块中的模糊规则的权重向量值w^l＝[λ₁I+(y^l)^T(y^l)]^-1(y^l)^Tz_l，其中I为单位矩阵，λ₁代表正则化系数，z_l＝(z₁,z₂,…,z_l)为期望的输出残差向量，z₁＝y,y为期望输出，当l＝2,…,s时z_l＝z_l-1-ε_l-2；4) In the hidden layer, based on the image information of each fault type, the fuzzy C-means clustering algorithm is used to obtain the corresponding initial fuzzy rules, and the canonical optimization problem is solved by solving the problem.

Obtain the weight vector value of the fuzzy rules in each fuzzy inference module w ^l =[λ ₁ I+(y ^l ) ^T (y ^l )] ^-1 (y ^l ) ^T z _l , where I is the identity matrix, and λ ₁ represents the regular z _l =(z ₁ ,z ₂ ,...,z _l ) is the expected output residual vector, z ₁ =y,y is the expected output, when l=2,...,s, z _l =z _{l -1} -ε _l-2 ;

其中

代表最终输出的故障类型。5) In the output layer, the output layer is connected to the hidden layer, and the output result is obtained by accumulating all the intermediate weighted values layer by layer

in

Represents the fault type of the final output.

The present invention also proposes a central air-conditioning fault diagnosis system based on images and deep blur, including:

A data processing module, the module is used to execute the method of step (1);

A digital image conversion module, the module is used to execute the method of step (2);

Residual depth blur module, the module is used to perform the method of step (3);

A fault diagnosis module, which is used for executing the method of step (4).

Compared with the prior art, the present invention has the following advantages:

1. The method uses the kernel slow feature analysis algorithm to extract slowly changing features from the dynamic air-conditioning operation data, and sorts the feature variables according to their slow changing degree, so that the fault features are enhanced;

2. The digital map conversion method converts the feature-enhanced data into images, and fully exploits the neighborhood information and spatial correlation characteristics between the feature variables;

3. The sliding window method can generate rich image data sets, which provides necessary conditions for the accurate establishment of fault diagnosis models;

4. The residual deep fuzzy model can quickly and accurately identify images for fault diagnosis, and make the constructed fault diagnosis model interpretable.

Description of drawings

The above and/or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of embodiments taken in conjunction with the accompanying drawings, wherein:

Fig. 1 is a structural block diagram of the present invention.

FIG. 2 is a schematic diagram of the structure of a two-dimensional grayscale image converted from numerical data.

FIG. 3 is a schematic structural diagram of an example of using the sliding window method.

FIG. 4 is a schematic structural diagram of a residual depth blur model.

Detailed ways

The following describes in detail the embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals refer to the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary, only used to explain the present invention, and should not be construed as a limitation of the present invention.

As shown in Figures 1-4, in order to effectively solve the problems in the background technology, the present invention adopts the strategy of combining the kernel slow feature analysis algorithm, the digital image conversion algorithm and the residual deep fuzzy model to construct the fault diagnosis model of the central air conditioner.

Kernel slow feature analysis algorithms can derive slowly changing output variables from the input data, and rank the feature variables according to their slowness. The digital map conversion method can convert the obtained numerical data into corresponding two-dimensional grayscale images to mine the neighborhood information and spatial correlation of feature variables. The residual deep fuzzy model has good interpretability characteristics, and the output judgment of fault diagnosis results can be based on evidence.

In view of the above situation, the present invention proposes a fault diagnosis method and system for a central air conditioner based on images and depth blur. The system mainly includes four modules: data processing module, digital image conversion module, residual depth fuzzy module, and fault diagnosis module. Among them, the data processing module performs feature extraction and feature sorting on the data set; then the digital image conversion module converts the feature-extracted data set into a corresponding two-dimensional grayscale image, and uses the sliding window method to generate a rich image data set; The differential depth blur module is used for fault diagnosis by training a residual depth blur model using a 2D grayscale image dataset. The real-time fault diagnosis module of the central air conditioner can collect data through multiple sensors and input it into the fault diagnosis model to judge whether the fault exists or not.

This embodiment proposes a fault diagnosis method for a central air conditioner based on images and depth blur, including the following steps:

Step (1): Data Processing

Collect and label the data of the normal operation of the central air conditioner and various faulty operations; preprocess the collected data; establish a kernel slow feature analysis model to extract and sort the features of the data, and obtain a kernel slow feature data set;

Step (2): Digital map conversion

Convert the kernel slow feature dataset after feature extraction into the corresponding two-dimensional grayscale image, and use the sliding window method to generate a rich image dataset;

Step (3): Build and train a residual deep fuzzy model

Establish a residual depth fuzzy model; through the acquired two-dimensional grayscale image data set, train the residual deep fuzzy model for fault diagnosis;

Step (4): Central Air Conditioning Fault Diagnosis

Measure and collect the data during the operation of the central air conditioner, obtain the newly collected data, perform steps 1 and 2 on the newly collected data, and input the processed data into the established residual depth fuzzy model, and carry out fault detection and verification. No judgment.

The following is explained in further detail:

1. Data processing module

This module is used to execute the method of step (1); the function of this module is to perform feature extraction and feature arrangement of data by establishing a kernel slow feature analysis model, so as to solve the temporal dynamic characteristics and nonlinear problems of central air conditioners. The module contains the following specific steps:

其包含N天的数据，并且每天的数据有T个时刻样本和P个特征变量。因此，某一天的数据可以表示为

其中t∈[1,2,...,T]表示时间范围，

为连续T个时间序列中第p个特征变量对应的值；数据采集完成之后，将数据标注上对应的运行工况标签。1.1 In step (1), the specific method of data collection and labeling is: by installing sensors in multiple positions of the central air conditioner (installing sensors at the core positions such as the supply and return air ducts, fresh air valves, return air valves and return fans of the central air conditioners) ) to collect data of normal operation and various faulty operations; data collection is in units of one day, and the sampling interval is one minute; the collected original data set is expressed as

It contains N days of data, and each day's data has T time samples and P feature variables. Therefore, the data for a certain day can be expressed as

where t∈[1,2,...,T] denotes the time range,

is the value corresponding to the p-th characteristic variable in consecutive T time series; after the data collection is completed, the data is marked with the corresponding operating condition label.

1.2 In step (1), the specific method of data preprocessing is:

Use the Z-score algorithm to normalize the original data set X in units of each day, as follows:

Among them, mean(x _p (t)) is the mean of x _p (t), and std(x _p (t)) is the standard deviation of x _p (t);

其中

由此，对每一天的数据均进行标准化处理后可得到数据集

The normalized data for a day is defined as

in

Thus, the data set can be obtained by normalizing the data of each day

映射到高维数

然后，通过求解下述优化问题获得慢特征：1.3 In step (1), the specific method of feature extraction is: use the normalized normal operation data to determine the optimal parameters, and establish a kernel slow feature analysis model; first, use the implicit nonlinear mapping function φ( ) to convert the data into

map to higher dimensions

Then, slow features are obtained by solving the following optimization problem:

是y_j(t)对时间变量t的一阶导数，<.>定义为

K为采样间隔；where the output data y _j (t) is the jth slow feature extracted from the input data x _φ (t),

is the first derivative of y _j (t) with respect to the time variable t, <.> is defined as

K is the sampling interval;

基于训练完成的核慢特征分析模型，从所有故障数据集中提取出变化缓慢的特征，得到N天核慢特征数据集，记为

Furthermore, the data after feature extraction is expressed as

Based on the trained kernel slow feature analysis model, the slowly changing features are extracted from all fault data sets, and the N-day kernel slow feature data set is obtained, denoted as

2. Digital image conversion module

This module is used to execute the method of step (2); the function of this module is to convert the numerical data into corresponding two-dimensional grayscale images in units of each day according to the data set provided by the data processing module, thereby enhancing the relationship between the characteristic variables. The neighborhood information and spatial correlation characteristics of . The module contains the following steps:

2.1 In step (2), the specific method of digital image conversion is:

① Perform Min-Max normalization on the data of each day in the kernel slow feature dataset Y after feature extraction and multiply the output result by 255. The formula is as follows:

where min(y _η (t)) and max(y _η (t)) are the minimum and maximum values of the y _η (t) (η=1,2,...,R) vector, respectively, and will obtain The N-day dataset is denoted as

② Convert the data set Z obtained in the above steps into a corresponding two-dimensional grayscale image; in the image, arrange the feature variables according to the slow change degree of the feature variables; specifically, convert the slowest-changing feature variable into the image The pixels in one column are converted into the pixels in the second column of the image, and the second-slowest-changing feature variables are converted into pixels in the second column of the image, and so on to obtain the converted two-dimensional grayscale image. Figure 2 shows the converted two-dimensional grayscale image. Example; where the data-to-image conversion is performed by converting the input data into an unsigned 8 integer type, followed by a data format conversion instruction; the larger the value, the darker the grayscale;

③Using the sliding window method to expand the generated image dataset; set the lag parameter to L, then the first image is generated from the data from the Mth row to (M+L-1) row; set the sliding distance of the sliding window If it is q, start intercepting from (M+q) line of data, and generate the second image at the end of (M+L+q-1) line of data, and so on; if the data in one day has a total of R characteristic variables and T By repeating the above operations, (T-l)/q+1 images can be obtained, all of which are L×R in size. Figure 3 shows an example of using the sliding window method.

3. Residual depth blur module

This module is used to execute the method of step (3); the function of this module is to train the residual depth fuzzy model for fault diagnosis through the acquired two-dimensional grayscale image data set. Figure 4 shows the structure of the model.

The residual depth fuzzy model in step (3) includes an input layer, a hidden layer, and an output layer; the model is realized by stacking fuzzy inference modules in a bottom-up, layer-by-layer manner; the detailed structure of the model and The principle is as follows:

1) The input image information of each fault type is initially obtained by the input layer, and is passed to the next layer (hidden layer) through the nodes of the equivalent mapping;

2) In the hidden layer, the fuzzy inference modules are stacked layer by layer, with a total of s layers; there are s fuzzy inference modules in the first hidden layer, and (s-1) fuzzy inference modules in the second hidden layer, and so on. There is only one fuzzy inference module in the sth hidden layer;

3) In the hidden layer, the output variables y ^l of all fuzzy inference modules in the previous layer and the residuals of the expected results are weighted as the input of the next layer of fuzzy inference modules, and the output of the first layer is:

为输出向量，Among them, l=1, 2,...,s, ε _l-1 represents the output of the layer,

is the output vector,

代表权重向量；

represents the weight vector;

得到各模糊推理模块中的模糊规则的权重向量值w^l＝[λ₁I+(y^l)^T(y^l)]^-1(y^l)^Tz_l，其中I为单位矩阵，λ₁代表正则化系数，z_l＝(z₁,z₂,…,z_l)为期望的输出残差向量，z₁＝y,y为期望输出，当l＝2,…,s时z_l＝z_l-1-ε_l-2；4) In the hidden layer, based on the image information of each fault type, the fuzzy C-means clustering algorithm is used to obtain the corresponding initial fuzzy rules, and the canonical optimization problem is solved by solving the problem.

Obtain the weight vector value of the fuzzy rules in each fuzzy inference module w ^l =[λ ₁ I+(y ^l ) ^T (y ^l )] ^-1 (y ^l ) ^T z _l , where I is the identity matrix, and λ ₁ represents the regular z _l =(z ₁ ,z ₂ ,...,z _l ) is the expected output residual vector, z ₁ =y,y is the expected output, when l=2,...,s, z _l =z _{l -1} -ε _l-2 ;

其中

代表最终输出的故障类型。5) In the output layer, the output layer is connected to the hidden layer, and the output result is obtained by accumulating all the intermediate weighted values layer by layer

in

Represents the fault type of the final output.

4. Fault diagnosis module

The module is used to execute the method of step (4); the function of the module is to perform fault diagnosis on the central air conditioner. The data during the operation of the air conditioner is measured and collected by multiple sensors, and input into the established fault diagnosis model. First, the data is converted into a 2D grayscale image, followed by a residual depth blur model for fault diagnosis.

Through the above embodiments, it can be seen that: (1) the method uses the kernel slow feature analysis algorithm to extract slowly changing features from the dynamic air-conditioning operation data, and sorts the feature variables according to the degree of slow change, so that the fault features are enhanced ; (2) The digital-to-map conversion method converts the feature-enhanced data into images, and fully exploits the neighborhood information and spatial correlation characteristics between the feature variables; (3) The sliding window method can generate rich image datasets for fault diagnosis. The accurate establishment of the model provides the necessary conditions; (4) The residual deep fuzzy model can quickly and accurately identify images for fault diagnosis, and make the constructed fault diagnosis model interpretable.

Although embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, The scope of the invention is defined by the claims and their equivalents.

Claims (7)

1. a central air-conditioning fault diagnosis method based on image and deep blur, is characterized in that, comprises the steps: Step (1): Data Processing Collect and label the data of the normal operation of the central air conditioner and various faulty operations; preprocess the collected data; establish a kernel slow feature analysis model to extract and sort the features of the data, and obtain a kernel slow feature data set; Step (2): Digital map conversion Convert the kernel slow feature dataset after feature extraction into the corresponding two-dimensional grayscale image, and use the sliding window method to generate a rich image dataset; Step (3): Build and train a residual deep fuzzy model Establish a residual depth fuzzy model; through the acquired two-dimensional grayscale image data set, train the residual deep fuzzy model for fault diagnosis; Step (4): Central Air Conditioning Fault Diagnosis Measure and collect the data during the operation of the central air conditioner, obtain the newly collected data, perform steps 1 and 2 on the newly collected data, and input the processed data into the established residual depth fuzzy model, and carry out fault detection and verification. No judgment. 2. A central air-conditioning fault diagnosis method based on image and depth blur according to claim 1, characterized in that, in the step (1), the specific method for data collection and labeling is: Sensors are installed at the location to collect data of normal operation and various faulty operations; data collection is in units of one day, and the sampling time interval is one minute; the collected raw data set is expressed as

It contains N days of data, and each day's data has T time samples and P feature variables. Therefore, the data for a certain day can be expressed as

where t∈[1,2,...,T] denotes the time range,

is the value corresponding to the p-th characteristic variable in consecutive T time series; after the data collection is completed, the data is marked with the corresponding operating condition label. 3. a kind of central air-conditioning fault diagnosis method based on image and depth blur according to claim 1, is characterized in that, in described step (1), the concrete method of data preprocessing is: Use the Z-score algorithm to normalize the original data set X in units of each day, as follows:

Among them, mean(x _p (t)) is the mean of x _p (t), and std(x _p (t)) is the standard deviation of x _p (t); The normalized data for a day is defined as

in

Thus, the data set can be obtained by normalizing the data of each day

4. A central air-conditioning fault diagnosis method based on image and depth blur according to claim 1, characterized in that, in the step (1), the specific method for feature extraction is: using standardized normal operation data to determine The optimal parameters are used to establish the kernel slow feature analysis model; first, the data is converted into

map to higher dimensions

Then, slow features are obtained by solving the following optimization problem:

where the output data y _j (t) is the jth slow feature extracted from the input data x _φ (t),

is the first derivative of y _j (t) with respect to the time variable t, <.> is defined as

K is the sampling interval; Furthermore, the data after feature extraction is expressed as

Based on the trained kernel slow feature analysis model, the slowly changing features are extracted from all fault data sets, and the N-day kernel slow feature data set is obtained, denoted as

5. a kind of central air-conditioning fault diagnosis method based on image and depth blur according to claim 1, is characterized in that, in described step (2), the concrete method of digital map conversion is: ① Perform Min-Max normalization on the data of each day in the kernel slow feature dataset Y after feature extraction and multiply the output result by 255. The formula is as follows:

where min(y _η (t)) and max(y _η (t)) are the minimum and maximum values of the y _η (t) (η=1,2,...,R) vector, respectively, and will obtain The N-day dataset is denoted as

② Convert the data set Z obtained in the above steps into a corresponding two-dimensional grayscale image; in the image, arrange the feature variables according to the slow change degree of the feature variables; specifically, convert the slowest-changing feature variable into the image The pixels in one column are converted into pixels in the second column of the image, and the second slowest-changing feature variable is converted to obtain the converted two-dimensional grayscale image; wherein, the data-to-image conversion is performed by converting the input The data is converted into an unsigned 8 integer type, which is then performed using the data format conversion instruction; the larger the value, the darker the grayscale; ③Using the sliding window method to expand the generated image dataset; set the lag parameter to L, then the first image is generated from the data from the Mth row to (M+L-1) row; set the sliding distance of the sliding window If it is q, start intercepting from (M+q) line of data, and generate the second image at the end of (M+L+q-1) line of data, and so on; if the data in one day has a total of R characteristic variables and T For each sample, repeating the above operation can obtain (T-l)/q+1 images, all of which are L×R in size. 6. a kind of central air-conditioning fault diagnosis method based on image and depth blur according to claim 1, is characterized in that, the residual depth fuzzy model in described step (3) comprises input layer, hidden layer and output layer ; The model is implemented by stacking fuzzy inference modules in a bottom-up, layer-by-layer manner; the detailed structure and principles of the model are as follows: 1) Preliminarily obtain the input image information of each fault type from the input layer, and pass it to the hidden layer through the nodes of the equivalent mapping; 2) In the hidden layer, the fuzzy inference modules are stacked layer by layer, with a total of s layers; there are s fuzzy inference modules in the first hidden layer, and (s-1) fuzzy inference modules in the second hidden layer, and so on. There is only one fuzzy inference module in the sth hidden layer; 3) In the hidden layer, the output variables y ^l of all fuzzy inference modules in the previous layer and the residuals of the expected results are weighted as the input of the next layer of fuzzy inference modules, and the output of the first layer is:

Among them, l=1, 2,...,s, ε _l-1 represents the output of the layer,

is the output vector,

represents the weight vector; 4) In the hidden layer, based on the image information of each fault type, the fuzzy C-means clustering algorithm is used to obtain the corresponding initial fuzzy rules, and the canonical optimization problem is solved by solving the problem.

Obtain the weight vector value of the fuzzy rules in each fuzzy inference module w ^l =[λ ₁ I+(y ^l ) ^T (y ^l )] ^-1 (y ^l ) ^T z _l , where I is the identity matrix, and λ ₁ represents the regular z _l =(z ₁ ,z ₂ ,...,z _l ) is the expected output residual vector, z ₁ =y,y is the expected output, when l=2,...,s, z _l =z _{l -1} -ε _l-2 ; 5) In the output layer, the output layer is connected to the hidden layer, and the output result is obtained by accumulating all the intermediate weighted values layer by layer

in

Represents the fault type of the final output. 7. A central air-conditioning fault diagnosis system based on image and deep blur is characterized in that, when being used for implementing the method for diagnosing a central air-conditioning fault based on image and deep blur according to any one of claims 1-6 steps, including: A data processing module, the module is used to execute the method of step (1); A digital-to-map conversion module, which is used to execute the method of step (2); Residual depth blur module, the module is used to perform the method of step (3); A fault diagnosis module, which is used for executing the method of step (4).

Images