Chiller sensor multi-source fault diagnosis method, system, equipment and medium

Abstract

The present invention discloses a multi-source fault diagnosis method, system, equipment and medium for chiller sensor. The method comprises: using historical sensor data of a normal chiller to train a Gaussian mixture model to obtain a multi-source fault pre-classification model; wherein the historical sensor data of the normal chiller includes normal data and preset fault data; using the pre-processed sensor data of the chiller to be tested as input of the multi-source fault pre-classification model, performing fault pre-classification through Gaussian distribution clustering, and obtaining a sensor fault pre-classification result of the chiller to be tested; using a K-means clustering algorithm to distinguish the sensor fault pre-classification results of the chiller to be tested, and obtaining a fault classification result of the chiller to be tested; the present invention realizes automatic fault detection through the overall operating status of the data; satisfies the diagnosis and detection of single-source or multi-source sensor faults of the chiller, has high fault detection efficiency, and obtains multi-source fault types.

Description

Technical Field

[0001] The present invention belongs to the technical field of air-conditioning sensor fault diagnosis, and in particular relates to a method, system, equipment and medium for diagnosing multi-source faults of chiller sensors. Background Art

[0002] Building energy is an important component of total energy consumption. Heating, ventilation and air conditioning (HVAC) systems account for approximately 30% of total building energy consumption. With the development of intelligent building technology, HVAC system energy-saving technologies based on control strategy optimization have played a significant role in building energy conservation. Sensors, as an important component of the control system, are prone to failure in high-temperature and high-salt working environments where they are in long-term contact with the outside world. In the event of a sensor failure, the HVAC system will not function properly, which will lead to a series of operational problems and many shortcomings, such as reduced system efficiency, wasted energy, poor indoor air quality, shortened equipment life, and even damage to components.

[0003] Therefore, the research on sensor fault diagnosis based on HVAC system is of great significance to building energy conservation and environmental protection; air conditioning system is usually composed of air system and water system; among them, the chiller in the water system (i.e. chiller) is the core component of the chiller; the exchange medium of cold water is usually composed of tap water. During long-term operation, the sensor is in a high temperature and high salt working environment for a long time, which can easily lead to sensor failure, and multiple failures can easily occur at the same time; existing sensor fault diagnosis methods are mostly limited to the failure of a single sensor in the system, and the diagnosis and detection methods for multiple sensor faults are still insufficient. Due to the high degree of multi-source fault coupling and other reasons, there is a lack of research on multi-source faults of chiller sensors. Summary of the Invention

[0004] In response to the technical problems existing in the prior art, the present invention provides a multi-source fault diagnosis method, system, equipment and medium for chiller sensors to solve the technical problem that the existing sensor fault diagnosis methods are mostly limited to the fault of a single sensor in the system and lack research on the multi-source fault of chiller sensors.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] The present invention provides a method for diagnosing multi-source faults of chiller sensors, which is characterized by comprising the following steps:

[0007] The Gaussian mixture model is trained using the historical sensor data of the normal chiller to obtain a multi-source fault pre-classification model; the historical sensor data of the normal chiller includes normal data and preset fault data;

[0008] The pre-processed sensor data of the chiller to be tested is used as the input of the multi-source fault pre-classification model, and the fault pre-classification is performed through Gaussian distribution clustering to obtain the sensor fault pre-classification results of the chiller to be tested;

[0009] The K-means clustering algorithm is used to distinguish the sensor fault pre-classification results of the chiller to be tested, and the fault classification results of the chiller to be tested are obtained.

[0010] Furthermore, the sensor history data of a normal chiller includes cooling water supply temperature sensor data, cooling water return temperature sensor data, chilled water supply temperature sensor data, and chilled water return temperature sensor data.

[0011] Furthermore, the preset fault data is collected by presetting fault forms for one or more sensors in a normal chiller; wherein the preset fault forms for one or more sensors in a normal chiller specifically include: single fault form, double fault form, triple fault form or four-method fault form;

[0012] Among them, single fault forms include: chilled water supply temperature sensor failure, chilled water return temperature sensor failure, cooling water supply temperature sensor failure and cooling water return temperature sensor failure;

[0013] Double fault modes include: chilled water supply temperature sensor failure-cooling water supply temperature sensor failure and chilled water return temperature sensor failure-cooling water return temperature sensor failure;

[0014] The three fault types include: chilled water supply temperature sensor failure-cooling water supply temperature sensor failure-cooling water return temperature sensor failure and chilled water return temperature sensor failure-cooling water supply temperature sensor failure-cooling water return temperature sensor failure;

[0015] The four types of faults include: chilled water supply temperature sensor failure - chilled water return temperature sensor failure - cooling water supply temperature sensor failure - cooling water return temperature sensor failure.

[0016] Furthermore, the Gaussian mixture module is trained using the historical sensor data of the normal chiller to obtain the multi-source fault pre-classification model. The details are as follows:

[0017] Place normal sensors or faulty sensors at sensor points of normal chillers at different time periods; collect data of normal sensors or faulty sensors at different time periods to obtain sensor historical data of normal chillers;

[0018] Using the EEMD-STD algorithm, denoising is performed on the sensor historical data of the normal chiller to obtain denoised samples;

[0019] The sensor data matrix Mar is constructed using the sensor data at the same time in the denoised samples in chronological order and normalized. According to the preset fault data form, the normalized sensor data matrix Mar is divided into a training data group Mar1 and a test data group Mar2.

[0020] Using the training data group Mar1 as a training sample of the Gaussian mixture model, training is performed to obtain a trained clustering model; wherein the trained clustering model is a Gaussian mixture model with trained parameters;

[0021] The test data group Mar2 is used as the test sample of the trained clustering model, and the parameters of the trained clustering model are evaluated and adjusted to obtain the maximum value of the likelihood function as the goal, thereby obtaining a Gaussian mixture model with optimal parameters, that is, the multi-source fault pre-classification model.

[0022] Furthermore, the optimal parameters in the Gaussian mixture model with optimal parameters include the weighting coefficients Average value μ i and covariance matrix ∑ i ;

[0023] The likelihood function is:

[0024]

[0025] in, is the log-likelihood function expectation of the training data; β is the initial parameter of the Gaussian mixture model; is the optimal parameter of the Gaussian mixture model; t is the sequence number of the training data in the training set; T is the length of the training set; X t is the tth training data in the training set; is the log-likelihood function expectation of each data in the training set; θ is an invisible variable that obeys the distribution of β probability; K is the number of Gaussian function models; P(X t ,θ / β) is the probability of the t-th training data in the training set under the condition that the hidden variable θ obeys the probability of β; P(X t / β) is the probability that the t-th training data in the training set obeys the β probability.

[0026] Furthermore, the pre-processed sensor data of the chiller to be tested is obtained by performing denoising on the sensor data of the chiller to be tested using the EEMD-STD algorithm.

[0027] Furthermore, the K-means clustering algorithm is used to distinguish the sensor fault pre-classification results of the chiller to be tested. In the process of obtaining the fault classification results of the chiller to be tested, the objective function of the K-means clustering algorithm is:

[0028]

[0029]

[0030] Where P is the Euclidean distance; r nk is the indicator matrix; x n is the sensor fault pre-classification result data of the chiller to be tested with number n; μ k is the cluster center; n is the number of the sensor fault pre-classification result data of the chiller to be tested; k is the cluster number; N is the total amount of sensor fault pre-classification result data of the chiller to be tested; K is the cluster category, that is, the category of the sensor fault pre-classification result data of the chiller to be tested.

[0031] The present invention also provides a multi-source fault diagnosis system for chiller sensors, comprising:

[0032] A pre-classification model module is used to train a Gaussian mixture model using historical sensor data of a normal chiller to obtain a multi-source fault pre-classification model; wherein the historical sensor data of the normal chiller includes normal data and preset fault data;

[0033] The fault pre-classification module is used to use the pre-processed sensor data of the chiller to be tested as the input of the multi-source fault pre-classification model, perform fault pre-classification through Gaussian distribution clustering, and obtain the sensor fault pre-classification results of the chiller to be tested;

[0034] The fault detection module is used to distinguish the sensor fault pre-classification results of the chiller to be tested by using the K-means clustering algorithm to obtain the fault classification results of the chiller to be tested.

[0035] The present invention also provides a multi-source fault diagnosis device for chiller sensors, comprising:

[0036] memory for storing computer programs;

[0037] A processor is used to implement the steps of the multi-source fault diagnosis method for chiller sensors when executing the computer program.

[0038] The present invention also provides a computer-readable storage medium, which stores a computer program, characterized in that when the computer program is executed by a processor, the steps of the multi-source fault diagnosis method of the chiller sensor are implemented.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] The present invention provides a multi-source fault diagnosis method and system for chiller sensors. By presetting fault data in the sensor historical data of a normal chiller, various fault states that may occur in the chiller are simulated; cluster analysis of the sensor data using a Gaussian mixture model is performed to achieve quantitative analysis of the sensor data, meeting the problem of pre-classifying multi-source fault data with a high degree of coupling, and combining the K-means clustering method to specifically classify the sensor fault pre-classification results, avoiding confusion of multi-source faults in the diagnosis process; there is no need to manually identify suspicious targets, and fault detection can be automatically performed through the overall operating status of the data; it meets the needs of diagnosing single-source or multi-source sensor faults in the chiller, and obtaining the fault area and the faulty sensor; the fault detection efficiency is high, and multi-source fault types are obtained, effectively filling the gap in multi-source fault diagnosis of sensors in the chiller.

[0041] Furthermore, the EEMD-STD algorithm is used to denoise the sensor data of the chiller to be tested, which effectively improves the signal-to-noise ratio of the sensor data, is conducive to clarifying the true distribution of the data, effectively improves the efficiency of multi-source fault diagnosis of sensors in the chiller, and improves the diagnosis success rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a flow chart of the multi-source fault diagnosis method for chiller sensors according to the present invention;

[0043] Figure 2 This is a diagram showing the data classification of test samples using the existing GMM algorithm;

[0044] Figure 3 This is a diagram showing the data classification of the fault diagnosis process using the existing GMM algorithm;

[0045] Figure 4 This is a diagram showing the effect of multi-source fault pre-classification using the existing GMM algorithm;

[0046] Figure 5 This is the fault diagnosis result diagram using the existing K-means clustering algorithm;

[0047] Figure 6 This is a diagram showing data classification of test samples in the embodiment;

[0048] Figure 7 A diagram showing data classification during a fault diagnosis process in an embodiment;

[0049] Figure 8 This is a diagram showing the effect of multi-source fault pre-classification in the embodiment;

[0050] Figure 9 4 is a diagram showing the fault diagnosis results in the embodiment. DETAILED DESCRIPTION

[0051] In order to make the technical problems, technical solutions and beneficial effects solved by the present invention more clearly understood, the present invention is further described in detail in the following specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0052] As attached Figure 1 As shown, the present invention provides a method for diagnosing multi-source faults of chiller sensors, comprising the following steps:

[0053] Step 1: Use historical sensor data from a normal chiller to train a Gaussian mixture model to obtain a multi-source fault pre-classification model. Prior to training the Gaussian mixture model, the EEMD-STD algorithm is used to denoise the historical sensor data from the normal chiller, which includes both normal data and pre-set fault data.

[0054] The sensor historical data of a normal chiller include cooling water supply temperature sensor data, cooling water return temperature sensor data, chilled water supply temperature sensor data and chilled water return temperature sensor data; the preset fault data are collected after presetting the fault form of one or more sensors in a normal chiller; wherein, the preset fault form of one or more sensors in a normal chiller specifically include: single fault form, double fault form, triple fault form or four-method fault form; wherein, the single fault form includes: chilled water supply temperature sensor failure, chilled water return temperature sensor failure, cooling water supply temperature sensor failure and cooling water return temperature sensor failure. Water return temperature sensor failure; double failure modes include: chilled water supply temperature sensor failure-cooling water supply temperature sensor failure and chilled water return temperature sensor failure-cooling water return temperature sensor failure; triple failure modes include: chilled water supply temperature sensor failure-cooling water supply temperature sensor failure-cooling water return temperature sensor failure and chilled water return temperature sensor failure-cooling water supply temperature sensor failure-cooling water return temperature sensor failure; quadruple failure modes include: chilled water supply temperature sensor failure-chilled water return temperature sensor failure-cooling water supply temperature sensor failure-cooling water return temperature sensor failure.

[0055] In the present invention, the Gaussian mixture module is trained using the historical sensor data of the normal chiller to obtain the multi-source fault pre-classification model. The specific process is as follows:

[0056] Step 11: Place normal sensors or faulty sensors at sensor locations of normal chillers at different time periods; collect data of normal sensors or faulty sensors at different time periods to obtain historical sensor data of normal chillers;

[0057] Step 12: Using the EEMD-STD algorithm, denoising is performed on the sensor historical data of the normal chiller to obtain denoised samples;

[0058] Step 13: construct a sensor data matrix Mar using the sensor data at the same time in the denoised samples in chronological order and perform normalization processing; divide the normalized sensor data matrix Mar into a training data group Mar1 and a test data group Mar2 according to the preset fault data form;

[0059] Step 14: Use the training data group Mar1 as a training sample of the Gaussian mixture model to perform training and obtain a trained clustering model; wherein the trained clustering model is a Gaussian mixture model with trained parameters; the optimal parameters of the Gaussian mixture model with optimal parameters include weighting coefficients Average value μ i and covariance matrix ∑ i ; The optimal parameters are calculated using the maximum expectation algorithm EM.

[0060] Step 15: Using the test data group Mar2 as the test sample of the trained clustering model, the parameters of the trained clustering model are evaluated and adjusted to obtain the maximum value of the likelihood function as the goal, and a Gaussian mixture model with optimal parameters is obtained, that is, the multi-source fault pre-classification model is obtained; wherein the likelihood function is:

[0061]

[0062] in, is the log-likelihood function expectation of the training data; β is the initial parameter of the Gaussian mixture model; is the optimal parameter of the Gaussian mixture model; t is the sequence number of the training data in the training set; T is the length of the training set; X t is the tth training data in the training set; is the log-likelihood function expectation of each data in the training set; θ is an invisible variable that obeys the distribution of β probability; K is the number of Gaussian function models; P(X t ,θ / β) is the probability of the t-th training data in the training set under the condition that the hidden variable θ obeys the probability of β; P(X t / β) is the probability that the t-th training data in the training set obeys the β probability.

[0063] Step 2: Use the preprocessed sensor data of the chiller under test as input to the multi-source fault pre-classification model. Perform fault pre-classification using Gaussian distribution clustering to obtain sensor fault pre-classification results for the chiller under test. The preprocessed sensor data of the chiller under test is obtained by denoising the sensor data using the EEMD-STD algorithm.

[0064] Step 3: Use the K-means clustering algorithm to distinguish the sensor fault pre-classification structure of the chiller to be tested, and obtain the fault classification result of the chiller to be tested. The objective function of the K-means clustering algorithm is:

[0065]

[0066]

[0067] Where P is the Euclidean distance; r nk is the indicator matrix; x n is the sensor fault pre-classification result data of the chiller to be tested with number n; μ k is the cluster center; n is the number of the sensor fault pre-classification result data of the chiller to be tested; k is the cluster number; N is the total amount of sensor fault pre-classification result data of the chiller to be tested; K is the cluster category, that is, the category of the sensor fault pre-classification result data of the chiller to be tested.

[0068] In the present invention, the sensor fault pre-classification results of the chiller to be tested are differentiated and diagnosed by the K-means clustering algorithm. As a classic unsupervised learning algorithm, the K-means clustering algorithm divides the data into various regions through iteration so that the sum of the distances from the data points to the centers of each region is minimized. In the iterative optimization, the cluster center μ k and indicator matrix r nk This is an NP problem that requires iterative optimization of the cluster center μ k and indicator matrix r nk , thus obtaining the optimal solution.

[0069] The specific iterative steps are as follows:

[0070] (1) Fixed cluster center μ k , optimize the indicator matrix r nk ; If the nth sample is closest to the kth center, assign r nk =1; otherwise r nk =0;

[0071] (2) Fixed indicator matrix r nk , optimize the cluster center μ k ,Depend on Yes, cluster center μ k Taking the derivative, we get Then we get the cluster center μ k ; Among them, the cluster center μ k for:

[0072]

[0073] Through iterative optimization, the goal of ultimately achieving the sum of the distances from the data points to the centers of each region is:

[0074]

[0075] After the above-mentioned iterative optimization process, the class centers of k regions can be finally obtained; wherein, the k value is the number of fault types in each fault group; the k regions are composed of k different fault pre-classification result data in the same group, and each region represents a fault type; the pre-processed sensor data of the chiller to be tested are pre-classified by the multi-source fault pre-classification model to generate multiple groups of sensor fault pre-classification result data of the chiller to be tested organized into clusters; the sensor fault pre-classification result data of the chiller to be tested are respectively input into the K-means clustering algorithm, and each group of sensor fault pre-classification result data of the chiller to be tested is subjected to the K-means clustering algorithm to obtain a specific fault type; wherein, k is the number of fault types in each fault group set artificially.

[0076] The present invention also provides a multi-source fault diagnosis system for chiller sensors, comprising a pre-classification model module, a fault pre-classification module and a fault detection module; the pre-classification model module is used to use the sensor historical data of a normal chiller to train a Gaussian mixture model to obtain a multi-source fault pre-classification model; wherein the sensor historical data of a normal chiller includes normal data and preset fault data; the fault pre-classification module is used to use the pre-processed sensor data of the chiller to be tested as the input of the multi-source fault pre-classification model, perform fault pre-classification through Gaussian distribution clustering, and obtain the sensor fault pre-classification result of the chiller to be tested; the fault detection module is used to use the K-means clustering algorithm to distinguish the sensor fault pre-classification result of the chiller to be tested, and obtain the fault classification result of the chiller to be tested.

[0077] The present invention also provides a multi-source fault diagnosis device for chiller sensors, comprising a memory and a processor: the memory is used to store a computer program; the processor is used to implement the steps of the multi-source fault diagnosis method for chiller sensors when executing the computer program.

[0078] When the processor executes the computer program, the steps of the above-mentioned multi-source fault diagnosis method for the chiller sensor are implemented, for example: using the sensor historical data of the normal chiller to train the Gaussian mixture model to obtain a multi-source fault pre-classification model; wherein the sensor historical data of the normal chiller includes normal data and preset fault data; using the preprocessed sensor data of the chiller to be tested as the input of the multi-source fault pre-classification model, performing fault pre-classification through Gaussian distribution clustering, and obtaining the sensor fault pre-classification result of the chiller to be tested; using the K-means clustering algorithm to distinguish the sensor fault pre-classification result of the chiller to be tested, and obtaining the fault classification result of the chiller to be tested.

[0079] Alternatively, when the processor executes the computer program, it implements the functions of each module in the above-mentioned system, for example: a pre-classification model module, which is used to use the sensor historical data of the normal chiller to train the Gaussian mixture model to obtain a multi-source fault pre-classification model; wherein the sensor historical data of the normal chiller includes normal data and preset fault data; a fault pre-classification module, which is used to use the pre-processed sensor data of the chiller to be tested as the input of the multi-source fault pre-classification model, perform fault pre-classification through Gaussian distribution clustering, and obtain the sensor fault pre-classification result of the chiller to be tested; a fault detection module, which is used to use the K-means clustering algorithm to distinguish the sensor fault pre-classification results of the chiller to be tested, and obtain the fault classification result of the chiller to be tested.

[0080] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing preset functions, and the instruction segments are used to describe the execution process of the computer program in the multi-source fault diagnosis device for chiller sensors. For example, the computer program can be divided into a pre-classification model module, a fault pre-classification module and a fault detection module, and the specific functions of each module are as follows: the pre-classification model module is used to use the sensor historical data of the normal chiller to train the Gaussian mixture model to obtain a multi-source fault pre-classification model; wherein the sensor historical data of the normal chiller includes normal data and preset fault data; the fault pre-classification module is used to use the pre-processed sensor data of the chiller to be tested as the input of the multi-source fault pre-classification model, perform fault pre-classification through Gaussian distribution clustering, and obtain the sensor fault pre-classification result of the chiller to be tested; the fault detection module is used to use the K-means clustering algorithm to distinguish the sensor fault pre-classification results of the chiller to be tested, and obtain the fault classification result of the chiller to be tested.

[0081] The multi-source fault diagnosis device for chiller sensors can be a computing device such as a desktop computer, laptop, PDA, or cloud server. The multi-source fault diagnosis device for chiller sensors may include, but is not limited to, a processor and memory. Those skilled in the art will appreciate that the above is merely an example of a multi-source fault diagnosis device for chiller sensors and does not constitute a limitation on the multi-source fault diagnosis device for chiller sensors. The multi-source fault diagnosis device for chiller sensors may include more components than shown, or a combination of certain components, or different components. For example, the multi-source fault diagnosis device for chiller sensors may also include input / output devices, network access devices, buses, and the like.

[0082] The processor may be a central processing unit (CPU), or other general-purpose processors, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor, or the processor may be any conventional processor, etc. The processor is the control center of the multi-source fault diagnosis device for chiller sensors, and uses various interfaces and lines to connect various parts of the multi-source fault diagnosis device for chiller sensors.

[0083] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the multi-source fault diagnosis device of the chiller sensor by running or executing the computer program and / or module stored in the memory and calling the data stored in the memory.

[0084] The memory may mainly include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function (such as a sound playback function, an image playback function, etc.); the data storage area may store data created based on the use of the mobile phone (such as audio data, a phone book, etc.). In addition, the memory may include a high-speed random access memory and may also include a non-volatile memory, such as a hard disk, internal memory, a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0085] The present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps of the multi-source fault diagnosis method for a chiller sensor are implemented.

[0086] If the modules / units integrated in the multi-source fault diagnosis system for chiller sensors are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium.

[0087] Based on this understanding, the present invention implements all or part of the process of the above-mentioned multi-source fault diagnosis method for chiller sensors by using a computer program to instruct related hardware. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of the above-mentioned multi-source fault diagnosis method for chiller sensors. The computer program includes computer program code, which can be in source code form, object code form, executable file, or a preset intermediate form.

[0088] The computer-readable storage medium may include: any entity or device that can carry the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0089] It should be noted that the content contained in the computer-readable storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practices in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practices, computer-readable storage media do not include electrical carrier signals and telecommunication signals.

[0090] Example

[0091] Taking the fault diagnosis process of a chiller of an air-conditioning system as an example, the collected sensor data is the summer operating condition data of the air-conditioning system.

[0092] This embodiment provides a method for diagnosing multi-source faults of chiller sensors, including the following steps:

[0093] Step 1: Obtain historical sensor data of a normal chiller, use the EEMD algorithm to denoise the historical sensor data of the normal chiller, and use the denoised historical sensor data of the normal chiller to train a Gaussian mixture model to obtain a multi-source fault pre-classification model; wherein the historical sensor data of the normal chiller includes normal data and preset fault data.

[0094] The specific process is as follows:

[0095] Step 11. Place normal sensors or faulty sensors at sensor points of normal chillers at different time periods according to preset fault modes. In this embodiment, the fault mode is preset according to a 30% offset fault. The sensor historical data of the normal chiller includes cooling water supply temperature sensor data, cooling water return temperature sensor data, chilled water supply temperature sensor data, and chilled water return temperature sensor data.

[0096] The preset fault data is collected after presetting the fault form of one or more sensors in a normal chiller; wherein, the preset fault form of one or more sensors in a normal chiller specifically includes: single fault form, double fault form, triple fault form or four-method fault form.

[0097] Among them, single fault forms include: chilled water supply temperature sensor failure, chilled water return temperature sensor failure, cooling water supply temperature sensor failure and cooling water return temperature sensor failure; double fault forms include: chilled water supply temperature sensor failure-cooling water supply temperature sensor failure and chilled water return temperature sensor failure-cooling water return temperature sensor failure; triple fault forms include: chilled water supply temperature sensor failure-cooling water supply temperature sensor failure-cooling water return temperature sensor failure and chilled water return temperature sensor failure-cooling water supply temperature sensor failure-cooling water return temperature sensor failure; quadruple fault forms include: chilled water supply temperature sensor failure-chilled water return temperature sensor failure-cooling water supply temperature sensor failure-cooling water return temperature sensor failure.

[0098] In this embodiment, a total of nine preset fault modes are set; sensor data under normal conditions and at different time periods of the nine preset fault modes are used as experiments to obtain real-time collection of sensor parameters in the air-conditioning heat exchanger and obtain the sensor historical data of the normal chiller.

[0099] Step 12: Use the EEMD-STD algorithm to perform denoising on the sensor historical data of the normal chiller to obtain denoised samples.

[0100] In this embodiment, the EEMD-STD algorithm is based on an improved data denoising method based on the EEMD algorithm, which can effectively improve the accuracy of Gaussian mixture model fault pre-classification and increase the success rate of Gaussian mixture model fault pre-classification by removing noise. In this embodiment, the EEMD-STD algorithm uses sensors as units for denoising, that is, data collected by the same sensor over a period of time is treated as a data group for denoising.

[0101] Step 13: Construct a sensor data matrix Mar using the sensor data at the same time in the denoised samples in chronological order, and perform normalization processing to obtain normalized sensor data Mar; wherein, the sensor data matrix Mar is:

[0102] Mar=[x1 x2 … x i … x n ], n=1,2,3,…

[0103]

[0104] Among them, x i is the sensor data set at the i-th moment; a mi is the data collected by the mth sensor at the i-th moment, n is the number of sensor data groups, and m is the number of sensor points in the chiller.

[0105] Step 14: Divide the normalized sensor data matrix Mar into a training data group Mar1 and a test data group Mar2 according to the preset fault data form; wherein each training data group Mar1 and test data group Mar2 contains normal data and various preset fault data.

[0106] Step 15: Use the training data group Mar1 as a training sample of the Gaussian mixture model to perform training to obtain a trained clustering model; wherein the trained clustering model is a Gaussian mixture model with trained parameters;

[0107] Step 16: Use the test data group Mar2 as the test sample of the trained clustering model, evaluate and adjust the parameters of the trained clustering model, and obtain the maximum value of the likelihood function as the goal, and obtain a Gaussian mixture model with optimal parameters, that is, obtain the multi-source fault pre-classification model.

[0108] In this embodiment, normal data in the training sample and several vectors under preset fault data are used as the initial parameters β of the Gaussian mixture model, and the selection of the initial contribution rate is averaged according to the number of simulated fault conditions; the maximum expectation algorithm EM is used to estimate the optimal parameters of the Gaussian mixture model.

[0109] The specific process is as follows:

[0110] Take the training set X of length T in the training data set Mar1, then:

[0111]

[0112] Among them, p(X|β) is the probability that the training set X obeys the initial parameter β; p(X t|β) is the probability that the t-th training data obeys the initial parameter β; T is the length of the training set X, X t is the tth training data in the training set.

[0113] The optimal parameters of the Gaussian mixture model are obtained by training the initial parameters β So that it satisfies the following conditions:

[0114]

[0115] in, For the training set X to obey the optimal parameters probability.

[0116] Optimal parameters for Gaussian mixture models The training process is carried out until the model converges. The specific steps are as follows:

[0117] Define the likelihood function Likelihood function is the log-likelihood function expectation of the training data, the likelihood function The expression is as follows:

[0118]

[0119] Among them, X t is the tth training data in the training set; is the log-likelihood function expectation of each data in the training set; θ is an invisible variable that obeys the distribution of β probability; K is the number of Gaussian function models; P(X t ,θ / β) is the probability of the t-th training data in the training set under the condition that the hidden variable θ obeys the probability of β; P(X t / β) is the probability that the t-th training data in the training set obeys the β probability.

[0120] Using the Bayesian probability formula, the probability that the training data belongs to the assumed hidden state θ is:

[0121]

[0122] Where, P(θ / X t ,β) is the posterior probability under the current invisible variable θ; is the contribution rate in the hidden state.

[0123] Calculate the weighted coefficients separately Average value μ i and covariance matrix ∑ i The parameter values ​​at which the partial derivatives of the likelihood function for the three parameters are zero.

[0124] Repeat the above steps iteratively to obtain the trained fault pre-classification clustering model.

[0125] Step 2: Use the pre-processed sensor data of the chiller to be tested as the input of the multi-source fault pre-classification model, perform fault pre-classification through Gaussian distribution clustering, and obtain the sensor fault pre-classification result of the chiller to be tested; wherein, the pre-processed sensor data of the chiller to be tested is obtained by denoising the sensor data of the chiller to be tested using the EEMD-STD algorithm.

[0126] Step 3: Use the K-means clustering algorithm to distinguish the sensor fault pre-classification structure of the chiller to be tested, and obtain the fault classification result of the chiller to be tested; wherein the objective function of the K-means clustering algorithm is:

[0127]

[0128]

[0129] Where P is the Euclidean distance; r nk is the indicator matrix; x n is the sensor fault pre-classification result data of the chiller to be tested with number n; μ k is the cluster center; n is the number of the sensor fault pre-classification result data of the chiller to be tested; k is the cluster number; N is the total amount of sensor fault pre-classification result data of the chiller to be tested; K is the cluster category, that is, the category of the sensor fault pre-classification result data of the chiller to be tested.

[0130] Test results

[0131] In this embodiment, a total of 2,500 sets of sensor data of normal chillers at different time periods are collected, including normal data and nine preset fault data. Among them, 500 sets of sensor data are collected under each preset fault mode. The above data are processed using the existing Gaussian mixture model method and the multi-source fault diagnosis method described in the present invention, and the diagnosis results using the existing Gaussian mixture model method and the diagnosis results using the multi-source fault diagnosis method described in this embodiment are respectively obtained, as follows:

[0132] Attachment Figure 2-3 And attached Figure 6-7 Faults a, b, c, and d represent four types of faults: chilled water supply temperature sensor fault, chilled water return temperature sensor fault, cooling water supply temperature sensor fault, and cooling water return temperature sensor fault. Faults connected by "-" indicate that the faults occur simultaneously.

[0133] As attached Figure 2-5 As shown, attached Figure 2 The data classification diagram of the test sample using the existing GMM algorithm is given in the appendix. Figure 3The data classification diagram of the fault diagnosis process using the existing GMM algorithm is given in the appendix. Figure 4 The multi-source fault pre-classification effect diagram using the existing GMM algorithm is given in the appendix. Figure 5 The fault diagnosis results using the existing K-means clustering algorithm are given in the figure; Figure 2-5 As can be seen from the figure, while the K-means clustering method has a high success rate for fault diagnosis of pre-classified data, the existing Gaussian mixture model method for fault pre-classification has an overall accuracy rate of only 58.92% under normal circumstances. The dual-triple engine fault pattern has a high degree of coupling, resulting in relatively high false positives. This is clearly unusable in engineering applications and poses a significant safety hazard.

[0134] As attached Figure 6-9 As shown, attached Figure 6 The data classification diagram of the test samples in the embodiment is given in the appendix. Figure 7 The data classification diagram of the fault pre-classification process in the embodiment is given in the appendix. Figure 6 The fault pre-classification result diagram in the embodiment is given in FIG. Figure 7 The data classification diagram of the fault diagnosis process in the embodiment is given in the appendix. Figure 8 The multi-source fault pre-classification effect diagram in the embodiment is given in the appendix. Figure 9 The fault diagnosis result diagram in the embodiment is given in the following; Figure 6-9 It can be seen that the multi-source fault diagnosis method described in this embodiment has good diagnostic capabilities, and the pre-classification accuracy rates under normal conditions and the four fault groupings are 99%, 99.4%, 94.6%, 98.8% and 94.8% respectively; compared with the existing Gaussian mixture model fault classification method, this embodiment effectively improves the accuracy of fault pre-classification, especially in the fault groupings: chilled water supply temperature sensor fault + cooling water supply temperature sensor fault, chilled water supply temperature sensor fault + cooling water supply temperature sensor fault + cooling water return temperature sensor fault, the pre-classification accuracy is improved by 99%, which greatly reduces the diagnosis confusion, and the comprehensive fault diagnosis success rate reaches 97.6%.

[0135] The description of the relevant parts of the multi-source fault diagnosis system, equipment and medium of a chiller sensor provided in this embodiment can be found in the detailed description of the corresponding parts of the multi-source fault diagnosis method of a chiller sensor described in this embodiment, and will not be repeated here.

[0136] In this embodiment, the sensor data is preprocessed by the EEMD-STD algorithm, which can reduce the noise rate of the data, improve the signal-to-noise ratio of the data, and is conducive to clarifying the true distribution of the data, and can better fit the physical operating status of each part of the system in the chiller; each possible fault state in the system is simulated, and pre-classified using the Gaussian mixture model, and the misjudgment of multiple faults is reduced through pre-classification. The pre-classified data is classified using the K-means clustering method to obtain specific fault types; in this embodiment, based on the EEMD-STD algorithm, after the sensor data is denoised, redundant information is removed, and the signal-to-noise ratio is improved, thereby improving the diagnosis and detection of chiller system sensor faults by the multi-source fault diagnosis method of the chiller sensor.

[0137] The multi-source fault diagnosis method for chiller sensors described in the present invention avoids the current diagnosis of chiller and even air handling system sensor faults being limited to fault detection using neural networks or PCA methods; the fault diagnosis process does not require manual identification of suspicious targets, and does not require detection of suspicious targets, and can automatically perform fault detection through the overall operating status of sensor data; the multi-source fault diagnosis method described in the present invention can simultaneously meet the needs of single-source or multi-source sensor faults, realize the detection of multi-source sensor faults caused by harsh environments, and find out the fault area and faulty sensors; compared with the sensor fault detection and diagnosis method that only uses the existing Gaussian mixture model and K-means clustering, the present invention improves the fault detection rate, and adopts the concept of pre-classification to greatly reduce the confusion probability of complex multi-source fault diagnosis.

[0138] The above embodiment is only one of the implementation methods that can realize the technical solution of the present invention. The scope of protection claimed by the present invention is not limited only to this embodiment, but also includes changes, replacements and other implementation methods that can be easily thought of by any technician familiar with this technical field within the technical scope disclosed by the present invention.

Claims

1. A multi-source fault diagnosis method for chiller sensors, characterized in that: The following steps are involved: The Gaussian mixture model is trained using the historical sensor data of the normal chiller to obtain a multi-source fault pre-classification model; the historical sensor data of the normal chiller includes normal data and preset fault data; The pre-processed sensor data of the chiller to be tested is used as the input of the multi-source fault pre-classification model, and the fault pre-classification is performed through Gaussian distribution clustering to obtain the sensor fault pre-classification results of the chiller to be tested; The K-means clustering algorithm is used to distinguish the sensor fault pre-classification results of the chiller to be tested, and the fault classification results of the chiller to be tested are obtained; The preset fault data is collected by presetting the fault form of one or more sensors in a normal chiller; wherein the preset fault form of one or more sensors in a normal chiller specifically includes: a single fault form, a double fault form, a triple fault form or a four-method fault form; Among them, single fault forms include: chilled water supply temperature sensor failure, chilled water return temperature sensor failure, cooling water supply temperature sensor failure and cooling water return temperature sensor failure; Double fault modes include: chilled water supply temperature sensor failure-cooling water supply temperature sensor failure and chilled water return temperature sensor failure-cooling water return temperature sensor failure; The three fault types include: chilled water supply temperature sensor failure-cooling water supply temperature sensor failure-cooling water return temperature sensor failure and chilled water return temperature sensor failure-cooling water supply temperature sensor failure-cooling water return temperature sensor failure; The four fault types include: chilled water supply temperature sensor failure - chilled water return temperature sensor failure - cooling water supply temperature sensor failure - cooling water return temperature sensor failure; The K-means clustering algorithm is used to distinguish the sensor fault pre-classification results of the chiller to be tested. In the process of obtaining the fault classification results of the chiller to be tested, the objective function of the K-means clustering algorithm is: Where P is the Euclidean distance; r nk is the indicator matrix; x n is the sensor fault pre-classification result data of the chiller to be tested with number n; μ k is the cluster center; n is the number of the sensor fault pre-classification result data of the chiller to be tested; k is the cluster number; N is the total amount of sensor fault pre-classification result data of the chiller to be tested; K is the cluster category, that is, the category of the sensor fault pre-classification result data of the chiller to be tested.

2. A multi-source fault diagnosis method for chiller sensors according to claim 1, characterized in that: The sensor historical data of a normal chiller includes cooling water supply temperature sensor data, cooling water return temperature sensor data, chilled water supply temperature sensor data, and chilled water return temperature sensor data.

3. A multi-source fault diagnosis method for chiller sensors according to claim 1, characterized in that: The process of training the Gaussian mixture module using the historical sensor data of the normal chiller to obtain the multi-source fault pre-classification model is as follows: Place normal sensors or faulty sensors at sensor points of normal chillers at different time periods; collect data of normal sensors or faulty sensors at different time periods to obtain sensor historical data of normal chillers; Using the EEMD-STD algorithm, denoising is performed on the sensor historical data of the normal chiller to obtain denoised samples; The sensor data matrix Mar is constructed using the sensor data at the same time in the denoised samples in chronological order and normalized; According to the preset fault data form, the normalized sensor data matrix Mar is divided into a training data group Mar1 and a test data group Mar2; Using the training data group Mar1 as a training sample of the Gaussian mixture model, training is performed to obtain a trained clustering model; wherein the trained clustering model is a Gaussian mixture model with trained parameters; The test data group Mar2 is used as the test sample of the trained clustering model, and the parameters of the trained clustering model are evaluated and adjusted to obtain the maximum value of the likelihood function as the goal, thereby obtaining a Gaussian mixture model with optimal parameters, that is, the multi-source fault pre-classification model.

4. A chiller sensor multi-source fault diagnosis method according to claim 3, characterized in that: The optimal parameters in the Gaussian mixture model with optimal parameters include the weighting coefficients Average value μ i and covariance matrix ∑ i ; The likelihood function is: in, is the log-likelihood function expectation of the training data; β is the initial parameter of the Gaussian mixture model; is the optimal parameter of the Gaussian mixture model; t is the sequence number of the training data in the training set; T is the length of the training set; X t is the tth training data in the training set; is the log-likelihood function expectation of each data in the training set; θ is an invisible variable that obeys the distribution of β probability; K is the number of Gaussian function models; P(X t ,θ / β) is the probability of the t-th training data in the training set under the condition that the hidden variable θ obeys the probability of β; P(X t / β) is the probability that the t-th training data in the training set obeys the β probability.

5. The multi-source fault diagnosis method for chiller sensors according to claim 1, characterized in that: The pre-processed sensor data of the chiller to be tested is obtained by denoising the sensor data of the chiller to be tested using the EEMD-STD algorithm.

6. A multi-source fault diagnosis system for chiller sensors, characterized in that: include: A pre-classification model module is used to train a Gaussian mixture model using historical sensor data of a normal chiller to obtain a multi-source fault pre-classification model; wherein the historical sensor data of the normal chiller includes normal data and preset fault data; The fault pre-classification module is used to use the pre-processed sensor data of the chiller to be tested as the input of the multi-source fault pre-classification model, perform fault pre-classification through Gaussian distribution clustering, and obtain the sensor fault pre-classification results of the chiller to be tested; A fault detection module is used to distinguish the sensor fault pre-classification results of the chiller to be tested using the K-means clustering algorithm to obtain the fault classification results of the chiller to be tested; The preset fault data is collected by presetting the fault form of one or more sensors in a normal chiller; wherein the preset fault form of one or more sensors in a normal chiller specifically includes: a single fault form, a double fault form, a triple fault form or a four-method fault form; Among them, single fault forms include: chilled water supply temperature sensor failure, chilled water return temperature sensor failure, cooling water supply temperature sensor failure and cooling water return temperature sensor failure; Double fault modes include: chilled water supply temperature sensor failure-cooling water supply temperature sensor failure and chilled water return temperature sensor failure-cooling water return temperature sensor failure; The three fault types include: chilled water supply temperature sensor failure-cooling water supply temperature sensor failure-cooling water return temperature sensor failure and chilled water return temperature sensor failure-cooling water supply temperature sensor failure-cooling water return temperature sensor failure; The four fault types include: chilled water supply temperature sensor failure - chilled water return temperature sensor failure - cooling water supply temperature sensor failure - cooling water return temperature sensor failure; The K-means clustering algorithm is used to distinguish the sensor fault pre-classification results of the chiller to be tested. In the process of obtaining the fault classification results of the chiller to be tested, the objective function of the K-means clustering algorithm is: Where P is the Euclidean distance; r nk is the indicator matrix; x n is the sensor fault pre-classification result data of the chiller to be tested with number n; μ k is the cluster center; n is the number of the sensor fault pre-classification result data of the chiller to be tested; k is the cluster number; N is the total amount of sensor fault pre-classification result data of the chiller to be tested; K is the cluster category, that is, the category of the sensor fault pre-classification result data of the chiller to be tested.

7. A multi-source fault diagnosis device for chiller sensors, characterized in that: include: memory for storing computer programs; A processor is configured to implement the steps of the chiller sensor multi-source fault diagnosis method according to any one of claims 1 to 5 when executing the computer program.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, wherein when the computer program is executed by a processor, the steps of the multi-source fault diagnosis method for a chiller sensor are implemented as described in any one of claims 1 to 5.

Images