Ventilation and air conditioning system sensor fault diagnosis method and system based on multi-source data

Abstract

The present invention proposes a ventilation and air-conditioning system sensor fault diagnosis method and system based on multi-source data, which relates to the field of quality control of ventilation and air-conditioning system sensors. The specific scheme includes: acquiring data of each sensor in the ventilation and air-conditioning system, and preprocessing it into multi-source data; using the constructed optimal classification hyperplane as the boundary of normal data, identifying anomalies in the multi-source data, and locating abnormal sensors; using a combination of linear and nonlinear methods to repair the abnormal data of the abnormal sensor; judging the fault type of the abnormal sensor based on the residual between the repaired data and the abnormal data; the present invention utilizes an intelligent diagnosis algorithm based on multi-sensor data correlation analysis to identify abnormal sensor data, and obtains the fault type of the abnormal sensor through the intelligent algorithm, and repairs the abnormal data.

Description

Technical Field

[0001] The present invention belongs to the field of quality control of ventilation and air-conditioning system sensors, and in particular, to a ventilation and air-conditioning system sensor fault diagnosis method and system based on multi-source data. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] The quality control of sensor data in ventilation and air conditioning systems is currently based on the threshold judgment method to identify abnormal sensors. The threshold judgment method requires setting a threshold range of normal values ​​for the sensor in advance. After each measurement data is obtained, it is compared with the threshold range. If it exceeds the range, it is considered that the data is abnormal. Although this method is simple and convenient, it is prone to misjudgment during actual use due to the subjective factors in the setting of the threshold and the large interval of the threshold range. In addition, the threshold judgment method only considers the operating status of the sensor itself, and does not consider the correlation between different sensor signals. It is easy to have the problem that although the sensor data is within the threshold range, the sensor test data is contradictory. Therefore, the current threshold-based sensor abnormal data identification method is inaccurate, which affects the quality control of the sensor data of the ventilation and air conditioning system. Summary of the invention

[0004] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides a ventilation and air-conditioning system sensor fault diagnosis method and system based on multi-source data, utilizes an intelligent diagnosis algorithm based on multi-sensor data correlation analysis to identify abnormal sensor data, and obtains the fault type of the abnormal sensor through an intelligent algorithm to repair the abnormal data.

[0005] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0006] A first aspect of the present invention provides a ventilation and air conditioning system sensor fault diagnosis method based on multi-source data.

[0007] A ventilation and air conditioning system sensor fault diagnosis method based on multi-source data includes:

[0008] Obtain data from various sensors in the ventilation and air conditioning system and preprocess it into multi-source data;

[0009] The constructed optimal classification hyperplane is used as the boundary of normal data to identify anomalies in multi-source data and locate abnormal sensors;

[0010] A combination of linear and nonlinear methods is used to repair abnormal data from abnormal sensors;

[0011] Based on the residual between the repaired data and the abnormal data, the fault type of the abnormal sensor is determined;

[0012] The optimal classification hyperplane is constructed by training the correlation between the sensor data, taking the origin of the high-dimensional space as the only negative point, and determining the optimal classification hyperplane in the high-dimensional space by maximizing the distance between the sample and the origin.

[0013] Furthermore, the ventilation and air-conditioning system includes various types of sensors such as temperature, humidity, pressure, and flow, and the preprocessing is to perform linear transformation on the data using deviation standardization to obtain normalized data.

[0014] Furthermore, the construction of the optimal classification hyperplane comprises the following specific steps:

[0015] Obtain sensor data under historical operating conditions and construct time series of multi-source data;

[0016] Preprocess the multi-source data at each moment in the time series;

[0017] Based on the preprocessed time series, the OCSVM model is trained to determine the boundary of normal samples and obtain the optimal classification hyperplane.

[0018] Furthermore, the OCSVM model maps multi-source data through a nonlinear function, performs linear regression in a high-dimensional space, and then uses the origin of the high-dimensional space as the only negative class point to determine the optimal classification hyperplane in the high-dimensional space by maximizing the distance between the sample and the origin.

[0019] Furthermore, the anomaly identification of multi-source data is to calculate the distance from the multi-source data to the center of the optimal classification hyperplane, and determine whether it is abnormal based on the distance from the center of the sphere and the radius of the sphere.

[0020] Furthermore, the abnormal data of the abnormal sensor is repaired, specifically:

[0021] The ARIMA model is used to predict the linear part of the normal data of the abnormal sensor;

[0022] The DBN network model is used to predict the nonlinear residual between the normal data of the abnormal sensor and the predicted value of the ARIMA model;

[0023] The linear prediction value and the nonlinear prediction value are added together to obtain the final repaired data of the abnormal sensor.

[0024] Furthermore, the fault types include deviation fault, drift fault, precision degradation fault and complete failure fault. Based on the residual between the repaired data and the abnormal data, the specific judgment method is as follows:

[0025] If the absolute value of the residual is a fixed constant, it is a deviation fault;

[0026] If the absolute value of the residual is greater than the threshold and changes linearly over time, it is a drift fault;

[0027] If the absolute value of the residual is greater than the threshold, the mean is equal, and the residual value follows a normal distribution, it is an accuracy degradation fault;

[0028] A hard fault identified by operation and maintenance personnel is a complete failure.

[0029] A second aspect of the present invention provides a ventilation and air conditioning system sensor fault diagnosis system based on multi-source data.

[0030] The sensor fault diagnosis system for ventilation and air conditioning systems based on multi-source data includes a data acquisition module, an abnormality identification module, an abnormality repair module and a fault diagnosis module:

[0031] The data acquisition module is configured to: acquire data from various sensors in the ventilation and air conditioning system and pre-process the data into multi-source data;

[0032] The anomaly identification module is configured to: use the constructed optimal classification hyperplane as the boundary of normal data, identify anomalies in multi-source data, and locate abnormal sensors;

[0033] The abnormality repair module is configured to: repair the abnormal data of the abnormal sensor by combining linear and nonlinear methods;

[0034] The fault diagnosis module is configured to: determine the fault type of the abnormal sensor based on the residual between the repaired data and the abnormal data;

[0035] The optimal classification hyperplane is constructed by training the correlation between the sensor data, taking the origin of the high-dimensional space as the only negative point, and determining the optimal classification hyperplane in the high-dimensional space by maximizing the distance between the sample and the origin.

[0036] A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps of the ventilation and air conditioning system sensor fault diagnosis method based on multi-source data as described in the first aspect of the present invention.

[0037] The fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the ventilation and air conditioning system sensor fault diagnosis method based on multi-source data as described in the first aspect of the present invention are implemented.

[0038] One or more of the above technical solutions have the following beneficial effects:

[0039] 1) The present invention proposes a new method for identifying abnormal sensor data based on the historical operation sensor data of the ventilation and air-conditioning system and the correlation of multiple sensor data. The method uses the correlation between the sensor data to establish the boundary of normal data, thereby identifying abnormal sensor data, greatly improving the accuracy of establishing the abnormal data identification model, and the model is more objective; it avoids the disadvantage that the threshold method is not suitable for online real-time identification of abnormal sensors. At the same time, the method is applicable to various types of sensors and is more scientific than the threshold method.

[0040] 2) Based on the correct identification of sensor data anomalies, the present invention further realizes the positioning of abnormal sensors and online repair of abnormal data; a model is established to predict the sensor data of the ventilation and air-conditioning system. When an abnormality occurs, only the predicted value needs to be used to repair the data, which is easy to operate; the prediction model has high accuracy and can improve the accuracy of abnormal data repair, thereby ensuring the normal operation of the ventilation and air-conditioning system.

[0041] 3) The present invention proposes a fault diagnosis method, which calculates the residual between the predicted value and the abnormal value, diagnoses the fault type according to the fault diagnosis rules, and the established fault diagnosis model can provide timely warnings for the staff, which has wide applicability.

[0042] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0044] Figure 1 This is a flow chart of the method of the first embodiment.

[0045] Figure 2 This is a flowchart of abnormality identification in the first embodiment.

[0046] Figure 3 This is the schematic diagram of the optimal classification hyperplane of the first embodiment.

[0047] Figure 4 This is a flow chart of repairing abnormal data of an abnormal sensor according to the first embodiment.

[0048] Figure 5 This is a flowchart of linear prediction of the first embodiment.

[0049] Figure 6 This is a diagram of the DBN network model structure of the first embodiment.

[0050] Figure 7 This is a flowchart of the nonlinear prediction of the first embodiment.

[0051] Figure 8 This is a flowchart of fault diagnosis for the first embodiment. DETAILED DESCRIPTION

[0052] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used in the present invention have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0053] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0054] In view of the problems existing in the current method for identifying abnormal sensor data in ventilation and air conditioning systems, the present invention proposes an intelligent diagnosis algorithm based on multi-sensor data correlation analysis to identify abnormal sensor data. On this basis, the fault type of the abnormal sensor is obtained through an intelligent algorithm, and the abnormal data is repaired. The main contents of the invention are as follows:

[0055] 1) Based on the OCSVM model, the correlation between the sensor data of ventilation and air conditioning is learned, and a hyperplane of multi-source data in high-dimensional space is constructed; when performing data quality diagnosis, the abnormal data is located by comparing the distance from each data to the center of the hyperplane and the radius of the sphere, and the abnormal data is identified by the correlation of the data.

[0056] 2) When repairing abnormal data, the ARIMA model is used to predict the linear part of the time series, and the DBN network is used to predict the nonlinear part of the time series. The prediction results of the two models are combined to obtain the predicted value of the abnormal sensor and realize the repair of the abnormal data. In addition, the sensor fault type is determined based on the deviation between the predicted value and the abnormal value.

[0057] The present invention can improve the accuracy of abnormal data recognition and realize online repair of abnormal data, and provide a new method for abnormal data early warning and repair of ventilation and air conditioning system sensors.

[0058] Embodiment 1

[0059] In one embodiment of the present disclosure, a ventilation and air conditioning system sensor fault diagnosis method based on multi-source data is provided. Based on the historical operation data of the ventilation and air conditioning system sensor (i.e., the collected sensor data), the correlation between the multi-sensor data is analyzed, and abnormal data identification, repair and fault diagnosis are performed by establishing a model, including four steps of data acquisition, abnormality identification, abnormality repair and fault diagnosis, such as Figure 1 As shown, specifically:

[0060] Step S1: Acquire data from various sensors in the ventilation and air-conditioning system and pre-process them into multi-source data.

[0061] Furthermore, the ventilation and air-conditioning system includes various types of sensors such as temperature, humidity, pressure, and flow, and the preprocessing is to perform linear transformation on the data using deviation standardization to obtain normalized data.

[0062] Specifically, the sensors in the ventilation and air conditioning system are numbered, and the sensor data is collected at all times. The collected The data of each sensor is expressed as , ventilation and air conditioning system has If there are sensors, the collected raw sensor data can be expressed as:

[0063] (1)

[0064] Due to the variety of sensor types in the system, different sensors collect data with different dimensions and dimensional units, which will affect data analysis. In order to eliminate the impact of dimensions, it is necessary to pre-process the data to normalize it. In this embodiment, deviation standardization is used to perform linear transformation on the data, mapping the value to [0, 1]. The conversion formula is:

[0065] (2)

[0066] in, express After normalization, the data Indicated in The maximum value in the moment data, Indicated in The minimum value in the moment data.

[0067] After data normalization, we get the time The multi-source data is expressed as:

[0068] (3)

[0069] Step S2: Use the constructed optimal classification hyperplane as the boundary of normal data to identify anomalies in multi-source data and locate abnormal sensors.

[0070] Abnormal identification is essentially a data sample classification problem, that is, classifying the collected sensor data into normal data and abnormal data; since the ventilation and air-conditioning system includes multiple types of sensors such as temperature, humidity, pressure, flow, etc., there is a certain correlation between them, such as the inlet water temperature of the chiller is not lower than the outlet water temperature, and the inlet water temperature of the cooling tower is not lower than the outlet water temperature, etc., this embodiment adopts an unsupervised single classification learning algorithm-OCSVM model, such as Figure 2 As shown in the figure, firstly, the historical sensor data is trained for correlation to obtain the optimal hyperplane. Secondly, the newly collected sensor data is identified for anomalies based on the correlation between the sensor data. It is divided into three steps:

[0071] 1. Construction of the optimal classification hyperplane

[0072] The optimal classification hyperplane is constructed by training the correlation between the sensor data, taking the origin of the high-dimensional space as the only negative point, and determining the optimal classification hyperplane in the high-dimensional space by maximizing the distance between the sample and the origin, specifically:

[0073] (1) Obtain sensor data under historical operating conditions and construct a time series of multi-source data;

[0074] (2) Preprocessing the multi-source data at each moment in the time series;

[0075] (3) Based on the preprocessed time series, the OCSVM model is trained to determine the boundaries of normal samples and obtain the optimal classification hyperplane.

[0076] Specifically, according to the sensor number in step S1, data samples under the historical operating state are obtained, sorted by time, and it is determined whether the data at this time has abnormalities such as data missing. If not, the normalization preprocessing in step S1 is performed to obtain the time series of multi-source data:

[0077] (4)

[0078] in, , Respectively indicate time , multi-source data.

[0079] Import the normalized data into the OCSVM model.

[0080] The OCSVM (One Class SVM) model is an unsupervised single-classification learning algorithm that belongs to the large family of support vector machines. However, it is different from the traditional classification and regression support vector machine based on supervised learning. It is an unsupervised learning method that does not require the output labels of the training set to be marked.

[0081] The OCSVM model transforms multi-source data through nonlinear functions After mapping, linear regression is performed in a higher-dimensional space to train the correlation between the sensor data, such as Figure 3 As shown, the origin of the high-dimensional space is taken as the only negative class point, and the optimal classification hyperplane in the high-dimensional space is determined by maximizing the distance d between the sample and the origin. The formula is as follows:

[0082] (5)

[0083] In the formula, and is the hyperplane parameter; is the slack variable; To control the parameters; is the sample size; A mapping function that maps samples to high-dimensional space.

[0084] During the training process, the 10-fold cross-validation method was used to divide the data samples into ten equal parts, and each part was used as a test sample in turn to determine the optimal parameter value of the OCSVM model.

[0085] After completing the above steps, the boundary of normal samples, that is, the optimal classification hyperplane, can be determined.

[0086] 2. Anomaly Identification

[0087] Furthermore, the anomaly identification of multi-source data is to calculate the distance from the multi-source data to the center of the optimal classification hyperplane, and determine whether it is abnormal based on the distance from the center of the sphere and the radius of the sphere.

[0088] Specifically, when anomaly recognition is performed on newly collected sensor data, the distance from the multi-source data at that moment to the center of the model hyperplane is calculated. If the distance from the data point to the center of the sphere is less than or equal to the radius of the sphere, that is, , the data can be judged as normal data, where is a test sample; on the contrary, if the distance to the center of the sphere is greater than the radius of the sphere, it means that there is an abnormality in the correlation between the previous and subsequent data of the same type of sensor or the data of different types of sensors, and the data is judged to be an abnormal point.

[0089] 3. Positioning abnormal sensors

[0090] After using the OCSVM model to identify that a data anomaly occurred in the ventilation and air-conditioning system sensor at a certain moment, the abnormal sensor was further located. As can be seen from the above, the data recognition principle of the OCSVM model is to compare the distance from the newly collected data to the center of the hyperplane and the radius of the sphere. Therefore, it can be considered that the points falling within the hyperplane sphere represent normal values, and the points outside the hyperplane are abnormal points.

[0091] The data of the multi-source data with abnormalities is located outside the hyperplane, and the digital number of the abnormal coordinate point in the time series is compared with the time label of the historical data of the ventilation and air-conditioning system operation to determine the specific time when the sensor information pointed to by the abnormal point is obtained. Compare the collected data of the same sensor in the time series before and after this moment to determine the chronological relationship between the values. For example, the sampling value of the outlet water temperature sensor of the chiller at 12:00 on June 8 is recognized as abnormal by the model. If the difference with the sampling value of the same sensor at the previous sampling moment and the next sampling moment is greater than 3°C, it is considered to be abnormal; compare the collected data of different types of sensors at this moment to determine the logical relationship between the signals of different types of sensors. If the correlation is low, the sensor is regarded as an abnormal sensor. For example, the sampling value of the outlet water temperature sensor of the chiller at 12:00 on June 8 is recognized as abnormal by the model, and the sampling value is higher than the sampling value of the inlet water temperature sensor of the chiller at the same sampling moment, it is considered to be abnormal.

[0092] Step S3: Use a combination of linear and nonlinear methods to repair the abnormal data of the abnormal sensor.

[0093] After realizing the abnormal identification of sensor data and the location of abnormal sensors, this implementation proposes a new method for online repair of abnormal sensor data. Since the time series of sensor data of ventilation and air conditioning system contains linear and nonlinear parts, such as Figure 4 As shown in the figure, the ARIMA model is used to predict the linear part of the normal data of the abnormal sensor, and the DBN network model is used to perform nonlinear prediction on the residual between the normal data of the abnormal sensor and the linear part prediction value. Finally, by fusing the linear prediction value and the nonlinear prediction value, the normal operating value of the sensor at the moment when the abnormal data appears is predicted, and the prediction value is used to perform online repair of the abnormal data. The specific steps are as follows:

[0094] 1. ARIMA model

[0095] The ARIMA (autoregressive integrated moving average model) model, whose full name is the differential autoregressive moving average model, is a commonly used time series model. Its advantage is that it can handle non-stationary time series. Its main steps are: first, perform a stationarity test, perform differential stabilization on the non-stationary part, then select the p and q coefficients according to the information criterion, and then predict the current series based on historical data.

[0096] The ARIMA model is used to perform linear prediction on the normal data of abnormal sensors, such as Figure 5 As shown, first determine the stationarity of the current time series. Non-stationary series need to perform data differentiation first. After ensuring that the input data is stable, it is used as the input of the ARIMA model.

[0097] The prediction principle of the ARIMA model is to treat the predicted value of the sensor at the desired moment as a linear function of the past sensor collection values, that is:

[0098] (6)

[0099] In the formula, Collect values ​​for historical sensors, is a measurement error with mean 0 and constant variance, p and q are the autoregressive and moving average orders, respectively, and their values ​​are determined during model training.

[0100] The model is tested for parameter significance and residual error. If the test passes, the linear part prediction value of the sensor at the current moment can be output.

[0101] 2. DBN network model

[0102] DBN (Deep Belief Network) network model, the full name is deep belief network, such as Figure 6 As shown in the figure, it contains multiple Restricted Boltzmann machines (RBM). The specific steps are: first, there is the pre-training stage, in which each RBM is trained separately. The last layer is the BP network, which receives the output of the RBM. After that, there is the fine-tuning stage, in which the error information is returned to each layer of RBM from top to bottom.

[0103] The DBN network model is used to predict the residual between the normal data of the abnormal sensor and the linear part prediction value, such as Figure 7As shown in the figure, firstly, the data is preprocessed, and the DBN network model is imported to complete the combined distribution between data and categories through the probability generation model. The network is trained and optimized layer by layer to generate a network structure with maximum probability training data. In the unsupervised training process, after multiple iterations of different batches, the optimal parameters of each layer are finally obtained. In the reverse fine-tuning process, the network model extracts the original data features of each layer, and then uses the top-level classifier for classification. Finally, the BP algorithm is used to combine the category information to fine-tune the relevant parameters from top to bottom to find the optimal solution.

[0104] By inputting the faulty sensor data into the trained model, the nonlinear part prediction value of the sensor at the current moment can be output.

[0105] 3. Prediction value fusion

[0106] The ARIMA model is used for linear part prediction, and the DBN network model is used for nonlinear part prediction. Through the comprehensive application of ARIMA and DBN network models, the linear prediction value and the nonlinear prediction value are added to obtain the final repaired data of the abnormal sensor, thereby improving the accuracy of the prediction effect.

[0107] Step S4: Based on the residual between the repaired data and the abnormal data, determine the fault type of the abnormal sensor.

[0108] Furthermore, the fault types include deviation fault, drift fault, precision degradation fault and complete failure fault. Based on the residual between the repaired data and the abnormal data, the specific judgment method is as follows:

[0109] If the absolute value of the residual is a fixed constant, it is a deviation fault;

[0110] If the absolute value of the residual is greater than the threshold and changes linearly over time, it is a drift fault;

[0111] If the absolute value of the residual is greater than the threshold, the mean is equal, and the residual value follows a normal distribution, it is an accuracy degradation fault;

[0112] A hard fault identified by operation and maintenance personnel is a complete failure.

[0113] Specifically, there are many types of sensors in ventilation and air conditioning systems, and the types of faults are not the same. Common sensor faults include deviation faults, drift faults, precision degradation faults, and complete failure faults. This embodiment proposes a new method for abnormal sensor fault diagnosis. By calculating the residual between the predicted value and the measured value, the fault type of the abnormal sensor is diagnosed, such as Figure 8 As shown:

[0114] (1) Deviation fault:

[0115] (7)

[0116] In the formula, is the threshold value.

[0117] If the predicted value With actual value The absolute value of the residual is a fixed constant ,and , it is believed that the sensor has a deviation failure.

[0118] If the predicted value With actual value The absolute value of the residual is less than or equal to the threshold , it is considered that no fault has occurred and data collection and abnormality identification continue.

[0119] (2) Drift fault:

[0120] (8)

[0121] If the predicted value With actual value The absolute value of the residual is greater than , changes linearly with time, and it is believed that the sensor has a drift failure.

[0122] (3) Accuracy degradation fault:

[0123] (9)

[0124] If the predicted value With actual value The absolute value of the residual is greater than , and the means are equal, and the residual values ​​obey the normal distribution, so it is considered that the sensor has a precision degradation failure.

[0125] (4) Complete failure

[0126] Since the complete failure is a hard fault, it is believed that it can be diagnosed promptly and accurately by the operation and maintenance personnel, so the fault diagnosis algorithm is not used for diagnosis.

[0127] Embodiment 2

[0128] In one embodiment of the present disclosure, a ventilation and air conditioning system sensor fault diagnosis system based on multi-source data is provided, including a data acquisition module, an abnormality identification module, an abnormality repair module and a fault diagnosis module:

[0129] The data acquisition module is configured to: acquire data from various sensors in the ventilation and air conditioning system and pre-process the data into multi-source data;

[0130] The anomaly identification module is configured to: use the constructed optimal classification hyperplane as the boundary of normal data, identify anomalies in multi-source data, and locate abnormal sensors;

[0131] The abnormality repair module is configured to: repair the abnormal data of the abnormal sensor by combining linear and nonlinear methods;

[0132] The fault diagnosis module is configured to: determine the fault type of the abnormal sensor based on the residual between the repaired data and the abnormal data;

[0133] The optimal classification hyperplane is constructed by training the correlation between the sensor data, taking the origin of the high-dimensional space as the only negative point, and determining the optimal classification hyperplane in the high-dimensional space by maximizing the distance between the sample and the origin.

[0134] Embodiment 3

[0135] The purpose of this embodiment is to provide a computer-readable storage medium.

[0136] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps in the ventilation and air-conditioning system sensor fault diagnosis method based on multi-source data as described in the first embodiment of the present disclosure.

[0137] Embodiment 4

[0138] The purpose of this embodiment is to provide an electronic device.

[0139] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps in the ventilation and air conditioning system sensor fault diagnosis method based on multi-source data as described in the first embodiment of the present disclosure are implemented.

[0140] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A ventilation and air conditioning system sensor fault diagnosis method based on multi-source data, characterized in that: include: Obtain data from various sensors in the ventilation and air conditioning system and preprocess it into multi-source data; The constructed optimal classification hyperplane is used as the boundary of normal data to identify anomalies in multi-source data and locate abnormal sensors; A combination of linear and nonlinear methods is used to repair abnormal data from abnormal sensors; Based on the residual between the repaired data and the abnormal data, the fault type of the abnormal sensor is determined; The optimal classification hyperplane is constructed by training the correlation between the sensor data, taking the origin of the high-dimensional space as the only negative point, and determining the optimal classification hyperplane in the high-dimensional space by maximizing the distance between the sample and the origin; The construction of the optimal classification hyperplane includes the following specific steps: Obtain sensor data under historical operating conditions and construct time series of multi-source data; Preprocess the multi-source data at each moment in the time series; Based on the preprocessed time series, the OCSVM model is trained to determine the boundary of normal samples and obtain the optimal classification hyperplane; The ARIMA model is used to predict the linear part of the normal data of the abnormal sensor; The DBN network model is used to predict the nonlinear residual between the normal data of the abnormal sensor and the predicted value of the ARIMA model. Specifically, the data is preprocessed first, the DBN network model is imported, the combined distribution between data and categories is completed through the probability generation model, the network is trained and optimized layer by layer, and a network structure with maximum probability training data is generated. In the unsupervised training process, the optimal parameters are obtained through iterations. In the reverse fine-tuning process, the network model extracts the original data features, uses the top-level classifier for classification, and uses the BP algorithm combined with the category information to fine-tune the parameters from top to bottom to find the optimal solution. The linear prediction value and the nonlinear prediction value are added to obtain the data of the abnormal sensor after final repair; The fault types include deviation fault, drift fault, precision degradation fault and complete failure fault. Based on the residual between the repaired data and the abnormal data, the specific judgment method is as follows: If the absolute value of the residual is a fixed constant, it is a deviation fault; If the absolute value of the residual is greater than the threshold and changes linearly over time, it is a drift fault; If the absolute value of the residual is greater than the threshold, the mean is equal, and the residual value follows a normal distribution, it is an accuracy degradation fault; A hard fault identified by operation and maintenance personnel is a complete failure.

2. The ventilation and air conditioning system sensor fault diagnosis method based on multi-source data according to claim 1, characterized in that: The ventilation and air-conditioning system includes various types of sensors such as temperature, humidity, pressure, and flow. The preprocessing is to perform linear transformation on the data using deviation standardization to obtain normalized data.

3. The ventilation and air conditioning system sensor fault diagnosis method based on multi-source data according to claim 1, characterized in that: The OCSVM model maps multi-source data through a nonlinear function, performs linear regression in a high-dimensional space, and then uses the origin of the high-dimensional space as the only negative class point to determine the optimal classification hyperplane in the high-dimensional space by maximizing the distance between the sample and the origin.

4. The ventilation and air conditioning system sensor fault diagnosis method based on multi-source data according to claim 1, characterized in that: The anomaly identification of multi-source data is to calculate the distance from the multi-source data to the center of the optimal classification hyperplane, and determine whether it is abnormal through the distance from the center of the sphere and the radius of the sphere.

5. A ventilation and air conditioning system sensor fault diagnosis system based on multi-source data, characterized in that: It includes data acquisition module, anomaly identification module, anomaly repair module and fault diagnosis module: The data acquisition module is configured to: acquire data from various sensors in the ventilation and air conditioning system and pre-process the data into multi-source data; The anomaly identification module is configured to: use the constructed optimal classification hyperplane as the boundary of normal data, identify anomalies in multi-source data, and locate abnormal sensors; The abnormality repair module is configured to: repair the abnormal data of the abnormal sensor by combining linear and nonlinear methods; The fault diagnosis module is configured to: determine the fault type of the abnormal sensor based on the residual between the repaired data and the abnormal data; The optimal classification hyperplane is constructed by training the correlation between the sensor data, taking the origin of the high-dimensional space as the only negative point, and determining the optimal classification hyperplane in the high-dimensional space by maximizing the distance between the sample and the origin; The construction of the optimal classification hyperplane includes the following specific steps: Obtain sensor data under historical operating conditions and construct time series of multi-source data; Preprocess the multi-source data at each moment in the time series; Based on the preprocessed time series, the OCSVM model is trained to determine the boundary of normal samples and obtain the optimal classification hyperplane; The ARIMA model is used to predict the linear part of the normal data of the abnormal sensor; The DBN network model is used to predict the nonlinear residual between the normal data of the abnormal sensor and the predicted value of the ARIMA model. Specifically, the data is preprocessed first, the DBN network model is imported, the combined distribution between data and categories is completed through the probability generation model, the network is trained and optimized layer by layer, and a network structure with maximum probability training data is generated. In the unsupervised training process, the optimal parameters are obtained through iterations. In the reverse fine-tuning process, the network model extracts the original data features, uses the top-level classifier for classification, and uses the BP algorithm combined with the category information to fine-tune the parameters from top to bottom to find the optimal solution. The linear prediction value and the nonlinear prediction value are added to obtain the data of the abnormal sensor after final repair; The fault types include deviation fault, drift fault, precision degradation fault and complete failure fault. Based on the residual between the repaired data and the abnormal data, the specific judgment method is as follows: If the absolute value of the residual is a fixed constant, it is a deviation fault; If the absolute value of the residual is greater than the threshold and changes linearly over time, it is a drift fault; If the absolute value of the residual is greater than the threshold, the mean is equal, and the residual value follows a normal distribution, it is an accuracy degradation fault; A hard fault identified by operation and maintenance personnel is a complete failure.

6. An electronic device, comprising: a memory for non-transitory storage of computer readable instructions; a processor for executing the computer-readable instructions; Wherein, when the computer-readable instructions are executed by the processor, the method according to any one of claims 1 to 4 is executed.

7. A storage medium, characterized in that: The computer-readable instructions are non-transitorily stored, wherein when the computer-readable instructions are executed by a computer, the method according to any one of claims 1 to 4 is performed.

Images