Air conditioning unit fault type diagnosis method and system

Abstract

The application relates to the technical field of air conditioning unit diagnosis, in particular to an air conditioning unit fault type diagnosis method and system, which comprises the following steps: based on a sensor group, the state of a target air conditioning unit is monitored according to a preset sampling period, so as to obtain a state data sequence of the target air conditioning unit; the state data sequence is subjected to data fitting, so as to obtain a state fitting curve; the state fitting curve is matched with a preset standard state fitting curve, so as to obtain the matching degree of the two; and the matching degree is compared with a preset matching degree threshold value.The application can timely find abnormal conditions by monitoring the state of the air conditioning unit through the sensor group, is convenient for early diagnosis and treatment, reduces the equipment fault risk, and can more accurately identify the running state of the equipment by combining the fitting of the state data sequence with the matching of the standard curve, so that the accuracy of fault detection is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of air conditioning unit diagnosis, in particular to an air conditioning unit fault type diagnosis method and system. BACKGROUND

[0002] An air conditioning unit is a device used to regulate air temperature, humidity, and quality, widely used in commercial buildings, industrial facilities, and homes, and its main functions include refrigeration, heating, ventilation, and air purification.

[0003] Traditional methods usually rely on manual inspection and experience-based judgment, which cannot achieve real-time monitoring, resulting in a longer reaction time for fault discovery and handling. Moreover, traditional methods rely on visual inspection or simple instrument measurement, which are easily affected by human factors and subjective judgment, which may lead to misdiagnosis or missed diagnosis. Moreover, traditional methods often do not systematically collect and analyze device operation data, lack scientific basis, and make fault diagnosis and prediction less accurate. Furthermore, the process of manual analysis and recording of fault information is tedious and inefficient, especially when a large number of devices need to be handled, which may cause information omission and delay. SUMMARY

[0004] The technical problem to be solved by the present application is to overcome the shortcomings of the prior art and provide an air conditioning unit fault type diagnosis method and system.

[0005] The technical solution adopted to solve the above technical problems is: an air conditioning unit fault type diagnosis method, comprising:

[0006] Based on the sensor group, the state of the target air conditioning unit is monitored according to a preset sampling period to obtain a state data sequence of the target air conditioning unit, and the state data sequence is fitted to obtain a state fitting curve.

[0007] The state fitting curve is matched with a preset standard state fitting curve to obtain a matching degree, and the matching degree is compared with a preset matching degree threshold.

[0008] If it is lower than the preset matching degree threshold, the sound of the target air conditioning unit is collected based on an audio collection device to obtain a sound signal of the target air conditioning unit, and the sound signal is preprocessed to obtain a standard sound signal.

[0009] The standard sound signal is extracted to obtain the mel-frequency cepstral coefficient of the standard sound signal, and the fault classification model is trained based on the mel-frequency cepstral coefficient to obtain the fault label of the target air conditioning unit.

[0010] The fault label is matched with a preset fault state atlas set to obtain a plurality of neighborhood fault labels corresponding to the fault label, and a repair work order set is constructed based on the fault label, the plurality of neighborhood fault labels and the state data sequence.

[0011] Preferably, the sound signal is preprocessed to obtain a standard sound signal, including:

[0012] The sound signal is emphasized, and the expression of the emphasis is as follows:

[0013]

[0014] Wherein H(z) represents the filter of the emphasis, a represents the emphasis parameter, and z represents the parameter of the filter of the emphasis;

[0015] The sound signal after the emphasis is divided into a plurality of sub-sound signals, and each sub-sound signal has an overlapping part, and the expression of the overlapping part is as follows:

[0016]

[0017] Wherein F represents the total number of sub-sound signals in the sound signal, w represents the length of the sub-sound signal, o represents the overlap rate, and n represents the total length in the sound signal;

[0018] The plurality of sub-sound signals are windowed to obtain a standard sound signal, and the expression of the windowing is as follows:

[0019]

[0020] Wherein ω(n) represents the windowing of the sub-sound signal, and L represents the window length of the windowing window.

[0021] Preferably, the standard sound signal is extracted to obtain the mel-cepstral coefficient of the standard sound signal, including:

[0022] The standard sound signal is subjected to discrete Fourier transform to obtain frequency domain data of the standard sound signal, and the expression of the discrete Fourier transform is as follows:

[0023]

[0024] Wherein X(k) represents the complex spectrum component at frequency index k of the standard sound signal, N represents the total number of samples of the standard sound signal, x(n) represents the sample value of the time domain signal at time index n of the standard sound signal, and j represents the imaginary unit;

[0025] calculating a spectral line energy of the frequency domain data, wherein the spectral line energy is calculated according to the following formula:

[0026] P m = |X(k)| 2 ;

[0027] wherein P m represents a spectral line energy of a complex spectrum component at a frequency index k of a standard sound signal;

[0028] calculating a logarithmic spectrum of the frequency domain data based on the spectral line energy, wherein the logarithmic spectrum is calculated according to the following formula:

[0029]

[0030] wherein S m represents a logarithmic spectrum of a complex spectrum component at a frequency index k of a standard sound signal, H m (k) represents a triangular band-pass filter, and f(m) represents a center frequency;

[0031] performing a discrete cosine transform on the logarithmic spectrum of the frequency domain data to obtain a mel-frequency cepstral coefficient of the standard sound signal, wherein the mel-frequency cepstral coefficient is expressed according to the following formula:

[0032]

[0033] wherein C(n) represents a mel-frequency cepstral coefficient of a standard sound signal.

[0034] Preferably, the fault classification model adopts a feedforward neural network, and the fault classification model comprises an input layer, a hidden layer and an output layer, the input layer and the hidden layer are fully connected, and the hidden layer and the output layer are sparsely connected, wherein a training method of the fault classification model comprises:

[0035] randomly initializing a weight matrix W, initializing and assigning a learning rate a, and initializing a stability factor m;

[0036] inputting a feature vector corresponding to the mel-frequency cepstral coefficient into the input layer, extracting an i-th training sample x i in the feature vector, and calculating an Euclidean distance between a j-th neuron and the training sample x i , wherein the Euclidean distance is calculated according to the following formula:

[0037] d j = ||x i - ω ij ||2;

[0038] wherein d i represents the Euclidean distance between the jth neuron and the training sample x i , ω ij represents the weight between the jth neuron and the training sample x i .

[0039] Finding two hidden layer neurons j and k with the smallest Euclidean distance to the training sample x i , and marking the output types of j and k as c j and c k .

[0040] Determining whether the output types c j and c k are the same, if the output types c j and c k are not the same, determining whether the output types c j and the label type corresponding to the training sample x i are the same, if they are the same, updating the weight matrix W through a first weight update formula, otherwise, determining whether the output types c k and the label type corresponding to the training sample x i are the same, if they are the same, updating the weight matrix W through a second weight update formula.

[0041] If the Euclidean distance between the hidden layer neuron j and the training sample x i is smaller than the Euclidean distance between the hidden layer neuron j and the training sample x i , repeating the above operation until the Euclidean distance between the hidden layer neuron j and the training sample x i is greater than or equal to the Euclidean distance between the hidden layer neuron j and the training sample x i .

[0042] Based on the updated weight matrix W, information propagation is performed to obtain a trained fault classification model.

[0043] Preferably, the first weight update formula is as follows:

[0044]

[0045] wherein ω ik represents the weight between the kth neuron and the training sample x i , and β represents a bias coefficient.

[0046] The second weight update formula is as follows:

[0047]

[0048] Preferably, the state data comprises air conditioning unit name, monitoring data, weather data and energy consumption standardized value, wherein the monitoring data comprises cold and hot water valve opening degree feedback, fresh air temperature, supply air temperature, return air temperature, supply air humidity and return air humidity.

[0049] Preferably, the matching degree is calculated according to the following formula:

[0050]

[0051] wherein M represents the matching degree between the state fitting curve and the preset standard state fitting curve, y i represents the standard state curve data point, represents the to-be-state fitting curve data point, and B represents the total number of data points.

[0052] Preferably, the fault label is matched with a preset fault state atlas set to obtain a plurality of neighborhood fault labels corresponding to the fault label, comprising:

[0053] The support degree of the fault label and a plurality of fault state atlases in the preset fault state atlas set is calculated, wherein the calculation formula of the support degree is as follows:

[0054]

[0055] wherein sup i represents the support degree of the fault label and the i-th fault state atlas in the preset fault state atlas set, represents the number of occurrences of the fault label in the i-th fault state atlas in the preset fault state atlas set, and n represents the total number of nodes in the fault state atlas;

[0056] The preset fault state atlas set is screened based on the support degree to obtain a plurality of matched preset fault state atlases;

[0057] The plurality of matched preset fault state atlases are compared in difference to obtain a difference coefficient set, wherein the calculation formula of the difference coefficient is as follows:

[0058]

[0059] wherein A i,j represents the difference coefficient between the i-th matched preset fault state atlas and the j-th matched preset fault state atlas, D i,j (v) represents the node difference number between the i-th matched preset fault state atlas and the j-th matched preset fault state atlas, D i,j(u) represents the difference in the number of edges between the i-th matched preset fault state graph and the j-th matched preset fault state graph, N n represents the sum of the number of vertices and the number of each edge label between the i-th matched preset fault state graph and the j-th matched preset fault state graph;

[0060] The difference coefficient set is sorted to obtain an ordered difference coefficient set, and the matched preset fault state graphs corresponding to the first, middle, and last difference coefficients in the ordered difference coefficient set are obtained.

[0061] The fault state nodes corresponding to the fault labels and the matched preset fault state graphs corresponding to the first, middle, and last difference coefficients are obtained, and the neighbor nodes of the fault state nodes are taken as the neighborhood fault labels to obtain a plurality of neighborhood fault labels.

[0062] Preferably, the expression of the repair work order set is as follows:

[0063]

[0064] wherein RE represents a repair work order set, represents the name of an air conditioning unit, τ represents monitoring data, and v represents weather data. represents an energy consumption standardization value, φ represents a fault label, and χ represents a plurality of neighborhood fault labels.

[0065] The technical solution adopted to solve the above technical problems is an air conditioning unit fault type diagnosis system, which is applicable to the air conditioning unit fault type diagnosis method and comprises:

[0066] A state monitoring unit is configured to perform state monitoring on a target air conditioning unit based on a sensor group according to a preset sampling period to obtain a state data sequence of the target air conditioning unit, and perform data fitting on the state data sequence to obtain a state fitting curve.

[0067] A curve matching unit is configured to match the state fitting curve with a preset standard state fitting curve to obtain a matching degree, and compare the matching degree with a preset matching degree threshold.

[0068] A sound acquisition unit is configured to, if the matching degree is lower than the preset matching degree threshold, perform sound acquisition on the target air conditioning unit based on an audio acquisition device to obtain a sound signal of the target air conditioning unit, and perform preprocessing on the sound signal to obtain a standard sound signal.

[0069] a fault diagnosis unit configured to perform feature extraction on the standard sound signal to obtain mel-frequency cepstral coefficients of the standard sound signal, and perform fault classification based on a trained fault classification model according to the mel-frequency cepstral coefficients to obtain a fault label of the target air conditioning unit;

[0070] a work order construction unit configured to match the fault label with a preset fault state atlas set to obtain a plurality of neighborhood fault labels corresponding to the fault label, and construct a repair work order set based on the fault label, the plurality of neighborhood fault labels, and the state data sequence.

[0071] The present application has the following advantages: (1) The present application can timely discover abnormal conditions by monitoring the state of the air conditioning unit through the sensor group, facilitate early diagnosis and processing, reduce the risk of equipment failure, and more accurately identify the running state of the equipment by combining the fitting of the state data sequence with the matching of the standard curve, thereby improving the accuracy of fault detection; (2) The present application can effectively capture the acoustic features of equipment failure by audio acquisition and sound signal processing using feature extraction technologies such as mel-frequency cepstral coefficients, which provides strong data support for fault classification, and can identify neighborhood faults related to the main fault by matching with the preset fault state atlas set, helping maintenance personnel to comprehensively understand potential problems and optimize maintenance strategies; (3) The present application can realize the automated processing of fault information by constructing a repair work order set based on the fault label and related data, improve work efficiency, reduce manual intervention, and reduce maintenance costs and downtime losses caused by sudden failures through accurate fault diagnosis and early warning, thereby prolonging the service life of the equipment. BRIEF DESCRIPTION OF DRAWINGS

[0072] Figure 1 A step flowchart of the overall method in one embodiment of the present application is shown in the figure;

[0073] Figure 2 A system architecture diagram of the overall system in one embodiment of the present application is shown in the figure.

[0074] Reference signs: 1, state monitoring unit; 2, curve matching unit; 3, sound acquisition unit; 4, fault diagnosis unit; 5, work order construction unit. DETAILED DESCRIPTION

[0075] Embodiment one, as shown in the figure, the air conditioning unit fault type diagnosis method proposed by the present application comprises: Figure 1

[0076] ​S1, based on the sensor group, the state of the target air conditioning unit is monitored according to the preset sampling period, so as to obtain the state data sequence of the target air conditioning unit, and the state data sequence is fitted to obtain a state fitting curve;

[0077] S2, the state fitting curve is matched with the preset standard state fitting curve to obtain the matching degree of the two, and the matching degree is compared with the preset matching degree threshold;

[0078] S3, if the matching degree is lower than the preset matching degree threshold, the sound signal of the target air conditioning unit is collected based on the audio collection device, and the sound signal is preprocessed to obtain a standard sound signal;

[0079] S4, the standard sound signal is extracted to obtain the mel frequency cepstral coefficient of the standard sound signal, and the trained fault classification model is used to classify the fault according to the mel frequency cepstral coefficient, so as to obtain the fault label of the target air conditioning unit;

[0080] S5, the fault label is matched with the preset fault state atlas set to obtain a plurality of neighborhood fault labels corresponding to the fault label, and the repair work order set is constructed based on the fault label, the plurality of neighborhood fault labels and the state data sequence.

[0081] In the present application, the standard state fitting curve refers to the ideal state curve set in advance, which is used for comparison with actual data; the audio collection device refers to the device for recording sound signal, such as microphone, which is used for collecting the running sound of air conditioning unit; the mel frequency cepstral coefficient (MFCC) refers to a common sound feature representation method, which is particularly suitable for speech and audio signal analysis, and can capture the frequency and energy characteristics of sound; the fault state atlas set refers to the set of various faults and their performance characteristics, which is used for comparison and analysis.

[0082] Embodiment two, compared with embodiment one, the air conditioning unit fault type diagnosis method provided by the present application further comprises: preprocessing the sound signal to obtain a standard sound signal, comprising:

[0083] A1, the sound signal is emphasized, and the expression of the emphasis is as follows:

[0084]

[0085] Wherein, H(z) represents the filter of the emphasis, a represents the emphasis parameter, and z represents the parameter of the filter of the emphasis;

[0086] A2, the sound signal after the emphasis is divided into a plurality of sub-sound signals, and each sub-sound signal has an overlapping part, and the expression of the overlapping part is as follows:

[0087]

[0088] wherein F represents the total number of sub-audio signals in the audio signal, w represents the length of the sub-audio signal, o represents the overlap rate, and n represents the total length in the audio signal;

[0089] A3, windowing the plurality of sub-audio signals to obtain a standard audio signal, wherein the expression of the windowing is as follows:

[0090]

[0091] wherein ω(n) represents the windowing of the sub-audio signal, and L represents the window length of the windowing window.

[0092] In an optional embodiment, feature extraction is performed on the standard audio signal to obtain the mel-cepstral coefficients of the standard audio signal, comprising:

[0093] B1, performing discrete Fourier transform on the standard audio signal to obtain frequency domain data of the standard audio signal, wherein the expression of the discrete Fourier transform is as follows:

[0094]

[0095] wherein X(k) represents the complex spectrum component at the frequency index k of the standard audio signal, N represents the total number of samples of the standard audio signal, x(n) represents the sample value of the time domain signal at the time index n of the standard audio signal, and j represents the imaginary unit;

[0096] B2, calculating the spectral line energy of the frequency domain data, wherein the calculation formula of the spectral line energy is as follows:

[0097] P m =|X(k)| 2 ;

[0098] wherein P m represents the spectral line energy of the complex spectrum component at the frequency index k of the standard audio signal;

[0099] B3, calculating the logarithmic spectrum of the frequency domain data based on the spectral line energy, wherein the calculation formula of the logarithmic spectrum is as follows:

[0100]

[0101] wherein S m represents the logarithmic spectrum of the complex spectrum component at the frequency index k of the standard audio signal, H m (k) represents a triangular band-pass filter, and f(m) represents the center frequency;

[0102] B4, performing a discrete cosine transform on the logarithmic spectrum of the frequency domain data to obtain a mel cepstral coefficient of the standard sound signal, wherein the mel cepstral coefficient is expressed as follows:

[0103]

[0104] wherein C(n) represents the mel cepstral coefficient of the standard sound signal.

[0105] In an optional embodiment, the fault classification model adopts a feedforward neural network, and the fault classification model comprises an input layer, a hidden layer and an output layer, the input layer and the hidden layer are fully connected, and the hidden layer and the output layer are sparsely connected, wherein the training method of the fault classification model comprises:

[0106] C1, randomly initializing a weight matrix W, initializing and assigning a learning rate a, and initializing a stability factor m;

[0107] C2, inputting a feature vector corresponding to the mel cepstral coefficient to the input layer, extracting an i-th training sample x i in the feature vector, and calculating an Euclidean distance between a j-th neuron and the training sample x i , wherein the calculation formula of the Euclidean distance is as follows:

[0108] d j =||x i -ω ij ||2;

[0109] wherein d i represents the Euclidean distance between the j-th neuron and the training sample x i , and ω ij represents a weight between the j-th neuron and the training sample x i .

[0110] C3, finding two hidden layer neurons j and k with the smallest Euclidean distance from the training sample x i , and marking the output types of the two hidden layer neurons as c j and c k .

[0111] C4, judging whether the output types c j and c k are the same, if the output types c j and c k are not the same, judging whether the output types c j and a label type corresponding to the training sample x i are the same, if the output types c kand the training sample x i whether the corresponding label types are the same, and if so, updating the weight matrix W through a second weight update formula;

[0112] C5, if the Euclidean distance between the hidden layer neuron j and the training sample x i is less than the Euclidean distance between the hidden layer neuron j and the training sample x i , repeating the above operation until the Euclidean distance between the hidden layer neuron j and the training sample x i is greater than or equal to the Euclidean distance between the hidden layer neuron j and the training sample x i ;

[0113] C6, propagating information based on the updated weight matrix W to obtain a trained fault classification model.

[0114] In an optional embodiment, the first weight update formula is as follows:

[0115]

[0116] wherein ω ik represents the weight between the kth neuron and the training sample x i , and β represents a bias coefficient;

[0117] The second weight update formula is as follows:

[0118]

[0119] In an optional embodiment, the state data includes air conditioning unit name, monitoring data, weather data, and energy consumption standardized value, wherein the monitoring data includes cold and hot water valve opening feedback, fresh air temperature, supply air temperature, return air temperature, supply air humidity, and return air humidity.

[0120] In an optional embodiment, the calculation formula of the matching degree is as follows:

[0121]

[0122] wherein M represents the matching degree between the state fitting curve and the preset standard state fitting curve, y i represents a standard state curve data point, represents a to-be-state-fitted curve data point, and B represents the total number of data points.

[0123] In an optional embodiment, the fault label is matched with a preset fault state atlas set to obtain a plurality of neighborhood fault labels corresponding to the fault label, including:

[0124] D1, calculate the support degree of the fault label and a plurality of fault state graphs in the preset fault state graph set, wherein the calculation formula of the support degree is as follows:

[0125]

[0126] Wherein, sup i represents the support degree of the fault label and the i-th fault state graph in the preset fault state graph set, represents the number of occurrences of the fault label in the i-th fault state graph in the preset fault state graph set, and n represents the total number of nodes in the fault state graph;

[0127] D2, screening the preset fault state graph set based on the support degree to obtain a plurality of matched preset fault state graphs;

[0128] D3, difference comparison of the plurality of matched preset fault state graphs to obtain a difference coefficient set, wherein the calculation formula of the difference coefficient is as follows:

[0129]

[0130] Wherein, A i,j represents the difference coefficient between the i-th matched preset fault state graph and the j-th matched preset fault state graph, D i,j (v) represents the number of node differences between the i-th matched preset fault state graph and the j-th matched preset fault state graph, D i,j (u) represents the number of associated edge differences between the i-th matched preset fault state graph and the j-th matched preset fault state graph, N n represents the sum of the vertex number and the associated edge label number between the i-th matched preset fault state graph and the j-th matched preset fault state graph;

[0131] D4, sorting the difference coefficient set to obtain a sequential difference coefficient set, obtaining the matched preset fault state graphs corresponding to the first, middle and last difference coefficients in the sequential difference coefficient set;

[0132] D5, obtaining the corresponding fault state nodes of the fault label and the matched preset fault state graphs corresponding to the first, middle and last difference coefficients, and taking the neighbor nodes of the fault state nodes as the neighborhood fault labels to obtain a plurality of neighborhood fault labels.

[0133] In an optional embodiment, the expression of the repair work order set is as follows:

[0134]

[0135] Where RE represents the set of repair work orders, This indicates the name of the air conditioning unit, τ represents the monitoring data, and v represents the weather data. φ represents the standardized energy consumption value, φ represents the fault label, and χ represents multiple neighboring fault labels.

[0136] Example 3, as Figure 2 As shown, the air conditioning unit fault type diagnosis system proposed in this invention, and the applicable air conditioning unit fault type diagnosis method, include:

[0137] Status monitoring unit 1 is used to monitor the status of the target air conditioning unit based on the sensor group according to the preset sampling period, so as to obtain the status data sequence of the target air conditioning unit, and to perform data fitting on the status data sequence to obtain the status fitting curve.

[0138] Curve matching unit 2 is used to match the state fitting curve with the preset standard state fitting curve to obtain the matching degree between the two, and compare the matching degree with the preset matching degree threshold.

[0139] The sound acquisition unit 3 is used to acquire sound from the target air conditioning unit based on the audio acquisition device if the sound level is lower than the preset matching threshold, so as to obtain the sound signal of the target air conditioning unit and preprocess the sound signal to obtain a standard sound signal.

[0140] Fault diagnosis unit 4 is used to extract features from standard sound signals to obtain the Mel-frequency cepstral coefficients of standard sound signals. Based on the trained fault classification model, fault classification is performed according to the Mel-frequency cepstral coefficients to obtain the fault label of the target air conditioning unit.

[0141] Work order construction unit 5 is used to match fault labels with a preset fault status map set to obtain multiple neighboring fault labels corresponding to the fault labels, and construct a repair work order set based on the fault labels, multiple neighboring fault labels and status data sequences.

[0142] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A method of diagnosing a fault type of an air conditioning unit, characterized by, include: Based on the sensor group, the target air conditioning unit is monitored according to the preset sampling period to obtain the state data sequence of the target air conditioning unit. The state data sequence is then fitted to obtain the state fitting curve. The state fitting curve is matched with a preset standard state fitting curve to obtain the matching degree between the two, and the matching degree is compared with a preset matching degree threshold. If the match is below a preset threshold, the target air conditioning unit is sampled using an audio acquisition device to obtain the sound signal of the target air conditioning unit. The sound signal is then preprocessed to obtain a standard sound signal. Feature extraction is performed on the standard sound signal to obtain the Mel-frequency cepstral coefficients of the standard sound signal. Based on the trained fault classification model, fault classification is performed according to the Mel-frequency cepstral coefficients to obtain the fault label of the target air conditioning unit. The fault label is matched with a preset fault status map set to obtain multiple neighboring fault labels corresponding to the fault label. A repair work order set is constructed based on the fault label, the multiple neighboring fault labels and the status data sequence. The fault classification model employs a feedforward neural network. The model includes an input layer, hidden layers, and an output layer. The input layer and the hidden layers are fully connected, while the hidden layers and the output layer are sparsely connected. The training method for the fault classification model includes: For the weight matrix Perform random initialization and adjust the learning rate. Perform initialization assignment and initialize the stability factor. ; The feature vector corresponding to the Mel-frequency cepstral coefficients is input into the input layer, and the first feature vector is extracted from the feature vector. training samples Calculate the first Each neuron and training sample The Euclidean distance between them is given by the following formula: ； in, Indicates the first Each neuron and training sample The Euclidean distance between them Indicates the first Each neuron and training sample The weights between them; Find the training samples The two hidden layer neurons with the smallest Euclidean distance and and mark its output type as and ; Determine the type of the output and Are they the same, if the output type is the same? and If they are different, then determine the type of the output. and the training samples If the corresponding label types are the same, then the weight matrix is ​​updated using the first weight update formula. If an update is performed, then the type of the output is determined. and the training samples If the corresponding label types are the same, then the weight matrix is ​​updated using the second weight update formula. Update; If the hidden layer neurons With training samples The Euclidean distance is smaller than that of the hidden layer neurons. With training samples If the Euclidean distance is found, the above operation is repeated until the hidden layer neuron is found. With training samples The Euclidean distance is greater than or equal to that of the hidden layer neurons. With training samples Euclidean distance; Based on the updated weight matrix Information is disseminated to obtain a trained fault classification model; The fault label is matched with a preset fault state map set to obtain multiple neighboring fault labels corresponding to the fault label, including: Calculate the support between the fault label and multiple fault state maps in the preset fault state map set, wherein the formula for calculating the support is as follows: ； in, This indicates that the fault label is related to the first fault in the preset fault state map set. Support of each fault state map This indicates that the fault label is related to the first fault in the preset fault state map set. The number of times the fault label appears in each fault state map. This represents the total number of nodes within the fault state graph; Based on the support, the preset fault state map set is filtered to obtain multiple matching preset fault state maps. The multiple matching preset fault state maps are compared to obtain a set of difference coefficients, wherein the formula for calculating the difference coefficients is as follows: ； in, Indicates the first The first matched preset fault state map and the second The difference coefficient between a matching preset fault state map Indicates the first The first matched preset fault state map and the second The number of node differences between matching preset fault state maps. Indicates the first The first matched preset fault state map and the second The number of edge differences between matching preset fault state maps. Indicates the first The first matched preset fault state map and the second The sum of the number of vertices and the number of associated edge tags between a set of matching preset fault state graphs; The set of difference coefficients is sorted to obtain an ordered set of difference coefficients, and the preset fault state map corresponding to the first, middle and last difference coefficients in the ordered set of difference coefficients is obtained. Obtain the corresponding fault state node in the preset fault state map that matches the first, middle and last difference coefficients of the fault label, and take the neighboring nodes of the fault state node as the neighborhood fault label to obtain multiple neighborhood fault labels.

2. The method for diagnosing fault types of air conditioning units according to claim 1, characterized in that, The sound signal is preprocessed to obtain a standard sound signal, including: The sound signal is emphasized, and the expression for the emphasis process is as follows: ； in, The filter representing the emphasis processing, Indicates the emphasis parameter. The parameters of the filter used for emphasis processing; The emphasized sound signal is divided into multiple sub-sound signals, and each sub-sound signal has an overlapping portion, wherein the expression of the overlapping portion is as follows: ； in, This represents the total number of sub-sound signals in the sound signal. Indicates the length of the sub-sound signal. Indicates the overlap rate. Indicates the total length of the sound signal; The multiple sub-sound signals are windowed to obtain a standard sound signal, wherein the expression for the windowing process is as follows: ； in, This indicates windowing processing of the sub-sound signal. This indicates the length of the window to which a window has been added.

3. The method for diagnosing fault types of air conditioning units according to claim 2, characterized in that, Feature extraction is performed on the standard sound signal to obtain its Mel-spectral coefficients, including: The standard sound signal is subjected to a Discrete Fourier Transform (DFT) to obtain its frequency domain data. The expression for the DFT is as follows: ； in, Frequency index representing a standard sound signal Complex spectral components at that location, This represents the total number of samples in the standard sound signal. Time index representing a standard sound signal The sample values ​​of the time-domain signal at that point. Represents the imaginary unit; Calculate the spectral line energy of the frequency domain data, wherein the formula for calculating the spectral line energy is as follows: ； in, Frequency index representing a standard sound signal The spectral energy of the complex spectral component at the location; The logarithmic spectrum of the frequency domain data is calculated based on the spectral line energy, wherein the formula for calculating the logarithmic spectrum is as follows: ； in, Frequency index representing a standard sound signal The logarithmic spectrum of the complex spectral components at point . This represents a triangular bandpass filter, and , Indicates the center frequency; The logarithmic spectrum of the frequency domain data is subjected to discrete cosine transform to obtain the Mel-frequency cepstral coefficients of the standard sound signal, wherein the expression for the Mel-frequency cepstral coefficients is as follows: ； in, The Mel-frequency cepstral coefficients represent the standard sound signal.

4. The method for diagnosing fault types of air conditioning units according to claim 3, characterized in that, The first weight update formula is as follows: ； in, Indicates the first Each neuron and training sample The weights between them Indicates the bias coefficient; The second weight update formula is as follows: 。 5. The method for diagnosing fault types of air conditioning units according to claim 1, characterized in that, The status data includes the name of the air conditioning unit, monitoring data, weather data, and standardized energy consumption values. The monitoring data includes feedback on the opening degree of the hot and cold water valves, fresh air temperature, supply air temperature, return air temperature, supply air humidity, and return air humidity.

6. The method for diagnosing fault types of air conditioning units according to claim 1, characterized in that, The formula for calculating the matching degree is as follows: ； in, This indicates the degree of matching between the state-fitted curve and the preset standard state-fitted curve. Represents the standard state curve data points. This represents the data points of the curve to be fitted to the desired state. This indicates the total number of data points.

7. The method for diagnosing fault types of air conditioning units according to claim 1, characterized in that, The expression for the set of repair work orders is as follows: ； in, This represents a collection of repair work orders. Indicates the name of the air conditioning unit. Indicates monitoring data, Represents weather data. This represents the standardized energy consumption value. Indicates fault label, This indicates multiple neighboring fault labels.

8. An air conditioning unit fault type diagnosis system, applicable to the air conditioning unit fault type diagnosis method according to any one of claims 1-7, characterized in that, include: The status monitoring unit (1) is used to monitor the status of the target air conditioning unit based on the sensor group according to the preset sampling period, so as to obtain the status data sequence of the target air conditioning unit, and to perform data fitting on the status data sequence to obtain the status fitting curve. Curve matching unit (2), the curve matching unit (2) is used to match the state fitting curve with the preset standard state fitting curve to obtain the matching degree between the two, and compare the matching degree with the preset matching degree threshold; The sound acquisition unit (3) is used to acquire sound from the target air conditioning unit based on the audio acquisition device if the sound level is lower than the preset matching threshold, so as to obtain the sound signal of the target air conditioning unit and preprocess the sound signal to obtain a standard sound signal. Fault diagnosis unit (4) is used to extract features from the standard sound signal to obtain the Mel-Cepstral Coefficient of the standard sound signal, and to classify the fault based on the Mel-Cepstral Coefficient of the trained fault classification model to obtain the fault label of the target air conditioning unit. Work order construction unit (5) is used to match the fault label with a preset fault status map set to obtain multiple neighboring fault labels corresponding to the fault label, and construct a repair work order set based on the fault label, the multiple neighboring fault labels and the status data sequence.

Images