HVAC Fault Detection Method Based on Condition-Driven and Neural Component Analysis

Abstract

This invention discloses a fault detection method for HVAC systems based on condition-driven and neural component analysis, belonging to the field of HVAC system fault detection technology. The method involves acquiring real-time operating data of the HVAC system; inputting the real-time operating data into a trained fault detection model for processing to obtain fault detection results; wherein, the process of training the fault detection model includes: acquiring time-varying three-dimensional normal operating data of the HVAC system; reconstructing the time-varying three-dimensional normal operating data from the time axis to the temperature condition axis based on temperature condition-driven methods to obtain temperature condition slices; performing canonical variable analysis on the temperature condition slices to obtain multi-condition normal data based on the temperature condition slices; and inputting the multi-condition normal data into the neural component analysis model for processing to obtain statistical control limits for each condition. This method can improve the efficiency and accuracy of HVAC fault detection, solving the problem of reduced accuracy and efficiency in existing fault detection technologies.

Description

Technical Field

[0001] This invention relates to the field of HVAC system fault detection technology, and in particular to a fault detection method for HVAC systems based on condition-driven and neural component analysis. Background Technology

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

[0003] Heating, ventilation, and air conditioning (HVAC) systems refer to indoor systems or related equipment responsible for heating, ventilation, and air conditioning, improving indoor comfort by controlling indoor air temperature and humidity. Due to the complexity of equipment operation, fluctuating operating parameters, and the influence of external interference, HVAC malfunctions are unavoidable. Prolonged malfunctions in HVAC systems can lead to reduced user comfort, decreased operating efficiency, and increased energy consumption. Considering the complexity of HVAC systems with numerous coupled components and the complex interactions between HVAC systems, buildings, and residents, it is necessary to explore an efficient and accurate method and system for HVAC fault detection.

[0004] In the actual operation of HVAC systems, their operating conditions are affected by user temperature setpoints, indoor occupancy, and changes in the external environment. This results in widespread and frequent changes in operating conditions, and the switching process of operating conditions is random, exhibiting typical large-scale non-stationary characteristics in the time dimension. These non-stationary characteristics are usually reflected by time-varying mean and time-varying autocovariance. Furthermore, due to changes in external ambient temperature and humidity, indoor load and set parameters, and changes in the HVAC operating mode under closed-loop control, the process data exhibits strong correlation, obvious dynamic characteristics, and nonlinear features.

[0005] However, existing HVAC fault detection methods generally do not take into account the above characteristics, resulting in the following problems:

[0006] (1) In actual operation, the operating conditions of HVAC systems change frequently, exhibiting a wide range of non-stationary characteristics. Most existing fault detection methods are based on single-model monitoring methods, which do not take into account the frequent changes in operating conditions. This may cause fault information to be hidden in the normal changes of a wide range of non-stationary processes, resulting in a decrease in the accuracy of fault detection. Existing multi-model-based fault detection methods divide multiple operating conditions on the time axis based on time-driven methods. Since the working state of HVAC systems changes frequently and is unpredictable, it often brings great difficulties to the modeling and understanding of multiple operating conditions.

[0007] (2) When dividing the HVAC system into multiple operating condition datasets, it is necessary to extract the key features of the data for cluster analysis. However, existing clustering methods generally do not consider the correlation information and dynamic characteristics between various variables in the HVAC process data, and the data features are not fully extracted, resulting in inaccurate clustering of multiple operating condition datasets.

[0008] (3) After the multi-condition dataset is divided, a nonlinear monitoring model needs to be established to detect faults in each sub-condition based on the nonlinear characteristics of each dataset. However, the existing nonlinear fault detection model has a large dependence on the dimensionality of the training data for computational complexity, and lacks orthogonal constraints when extracting features, resulting in a lot of redundant information among the extracted nonlinear features, which affects the fault detection effect. Summary of the Invention

[0009] To address the shortcomings of existing technologies, this invention provides a HVAC fault detection method, system, electronic device, and computer-readable storage medium based on condition-driven and neural component analysis, thereby improving the accuracy and efficiency of HVAC fault detection.

[0010] In a first aspect, the present invention provides a fault detection method for HVAC systems based on conditional driving and neural component analysis;

[0011] A fault detection method for HVAC systems based on condition-driven and neural component analysis includes:

[0012] Obtain real-time operating data for HVAC systems;

[0013] Real-time running data is input into the trained fault detection model for processing to obtain fault detection results;

[0014] The process of training the fault detection model includes:

[0015] Obtain time-varying three-dimensional normal operation data of HVAC system, and based on temperature condition drive, reconstruct the time-varying three-dimensional normal operation data from the time axis to the temperature condition axis to obtain temperature condition slices;

[0016] Perform canonical variable analysis on temperature condition slices to obtain normal data for multiple operating conditions based on temperature condition slices.

[0017] Normal data under multiple operating conditions are input into the neural component analysis model for processing to obtain the statistical control limits for each operating condition.

[0018] Furthermore, the process of reconstructing time-varying three-dimensional normal operation data from the time axis to the temperature axis based on temperature conditions, and obtaining temperature condition slices, includes:

[0019] The time-varying three-dimensional normal operation data is sliced ​​perpendicular to the time axis to obtain two-dimensional time slices at different times;

[0020] Based on two-dimensional time slices, the mean value of temperature variables in each two-dimensional time slice is calculated to obtain temperature condition slices;

[0021] Arrange the temperature condition slices in ascending order of the mean of the temperature variable to obtain temperature condition slices based on the temperature condition axis.

[0022] Furthermore, the canonical variable analysis of the temperature condition slices to obtain multi-condition normal data based on the temperature condition slice division includes:

[0023] Standardize the temperature condition slices to obtain standard temperature slices;

[0024] Canonical variable analysis was performed on standard temperature slices to obtain the correlation coefficients between adjacent standard temperature slices;

[0025] Based on the correlation coefficient and preset threshold, standard temperature slices are clustered to obtain normal data for multiple operating conditions based on temperature condition slices.

[0026] Furthermore, the step of inputting the normal data under multiple operating conditions into the neural component analysis model for processing includes:

[0027] Input the normal data under multiple operating conditions into the corresponding neural component analysis model to obtain the nonlinear principal component features of the normal data under each operating condition.

[0028] Based on the nonlinear principal component features, the Hotelling statistic and the squared prediction error statistic of the nonlinear latent features are constructed.

[0029] According to T 2 The control limits for the Hotelling statistic and the SPE statistic are constructed using kernel density estimation.

[0030] Furthermore, the step of inputting real-time operational data into the trained fault detection model for processing includes:

[0031] Based on real-time operational data, the kernel density estimates of the Hotelling statistic and the squared prediction error statistic are calculated and compared with preset statistical control limits to obtain fault detection results.

[0032] Preferably, if the kernel density estimate of the Hotelling statistic is greater than the preset control limit of the Hotelling statistic, or if the kernel density estimate of the squared prediction error statistic is greater than the preset control limit of the squared prediction error statistic, then the HVAC system malfunctions.

[0033] Furthermore, the fault detection model includes a nonlinear encoder and a decoder. The nonlinear encoder is used to extract nonlinear principal component features based on the input data; the decoder is used to decode the nonlinear principal component features into the original data space.

[0034] The fault detection model is trained by minimizing the reconstruction error between the input data and the decoded data.

[0035] Secondly, the present invention provides a HVAC fault detection system based on condition-driven and neural component analysis;

[0036] A HVAC fault detection system based on condition-driven and neural component analysis includes:

[0037] The data acquisition module is used to acquire real-time operating data of the HVAC system.

[0038] The fault detection module is used to input real-time running data into the trained fault detection model for processing in order to obtain fault detection results.

[0039] The process of training the fault detection model includes:

[0040] Obtain time-varying three-dimensional normal operation data of HVAC system, and based on temperature condition drive, reconstruct the time-varying three-dimensional normal operation data from the time axis to the temperature condition axis to obtain temperature condition slices;

[0041] Perform canonical variable analysis on temperature condition slices to obtain normal data for multiple operating conditions based on temperature condition slices.

[0042] Normal data under multiple operating conditions are input into the neural component analysis model for processing to obtain the statistical control limits for each operating condition.

[0043] Thirdly, the present invention provides an electronic device;

[0044] An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the steps of the above-described HVAC fault detection method based on condition-driven and neural component analysis.

[0045] Fourthly, the present invention provides a computer-readable storage medium;

[0046] A computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps of the above-described HVAC fault detection method based on condition-driven and neural component analysis.

[0047] Compared with the prior art, the beneficial effects of the present invention are:

[0048] 1. This invention employs a temperature-variable-driven method to reconstruct the three-dimensional normal operation data of HVAC systems. Temperature is selected as the indicator variable for condition-driven reconstruction, and the three-dimensional modeling data of the HVAC system is reconstructed from the time axis to the temperature axis, resulting in multiple temperature condition slices arranged in ascending order of temperature mean. This condition-driven method effectively solves the problem that frequent dynamic changes in operating conditions can mask fault information due to the large-scale non-stationary operation of HVAC systems between and within batches, thus significantly improving the accuracy of fault detection.

[0049] 2. This invention uses canonical variable analysis to cluster temperature condition slice data with high similarity. By maximizing the correlation of canonical variables in two adjacent temperature condition slices arranged in ascending order of temperature mean, the similarity between the two temperature condition slices is characterized. Temperature condition slices with high similarity are clustered into the same working condition. That is, the process characteristics of temperature condition slices under the same working condition are highly similar, while the process characteristics of temperature condition slices under different working conditions are significantly different, thereby achieving accurate division of multi-working-condition data.

[0050] 3. This invention employs neural component analysis to establish nonlinear fault detection models for different operating condition datasets, extracting independent nonlinear principal component features from multiple operating condition datasets. The nonlinear network of the neural component analysis model does not depend on the dimensionality of the data samples, and the use of orthogonal constraints greatly reduces the correlation between data. Therefore, by training a neural component analysis network with orthogonal constraints, the problems of nonlinearity and data redundancy in HVAC system data are solved. Attached Figure Description

[0051] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0052] Figure 1 This is a flowchart provided for an embodiment of the present invention;

[0053] Figure 2 This is a schematic diagram of the process for obtaining temperature condition slices according to an embodiment of the present invention;

[0054] Figure 3 A schematic diagram of the multi-condition division provided in an embodiment of the present invention;

[0055] Figure 4 This is a schematic diagram of the fault detection process provided in an embodiment of the present invention. Detailed Implementation

[0056] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0057] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments of the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. Furthermore, it should be understood that the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0058] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0059] Example 1

[0060] Currently, due to the large-scale non-stationary characteristics of HVAC systems during operation, the accuracy of existing fault detection methods is reduced, there is a lot of redundant information, and the efficiency is reduced. Therefore, this invention provides a fault detection method for HVAC systems based on condition-driven and neural component analysis, which detects faults based on the process characteristics and data features of HVAC operation.

[0061] Next, combined Figures 1-4 This embodiment discloses a HVAC fault detection method based on condition-driven and neural component analysis, which includes the following steps:

[0062] S1. Obtain real-time operating data of HVAC systems.

[0063] S2. Input the real-time running data into the trained fault detection model for processing to obtain fault detection results.

[0064] The fault detection model is a neural component analysis model, which includes a nonlinear encoder and a decoder. The nonlinear encoder is used to extract nonlinear principal component features based on the operating condition data; the decoder is used to decode the nonlinear principal component features into the original data space; the fault detection model is trained by minimizing the reconstruction error between the operating condition data and the decoded data.

[0065] Furthermore, the process of training the fault detection model includes:

[0066] S201. Obtain time-varying three-dimensional normal operation data of HVAC systems. Based on temperature conditions, reconstruct the time-varying three-dimensional normal operation data from the time axis to the temperature condition axis, and obtain temperature condition slices. Specific steps include:

[0067] S2011. Obtain time-varying three-dimensional normal operation data of HVAC system, slice the time-varying three-dimensional normal operation data perpendicular to the time axis, and obtain two-dimensional time slices at different times.

[0068] S2012. Based on two-dimensional time slices, calculate the mean value of temperature variables in each two-dimensional time slice to obtain temperature condition slices.

[0069] S2013. Arrange the temperature condition slices in ascending order of the mean value of the temperature variables to obtain temperature condition slices based on the temperature condition axis.

[0070] In this embodiment, considering the continuous operation status of HVAC for several days, the operation status of a working day is defined as an operation batch I. Each operation batch contains two-dimensional data of time K and variable J. Arranging the normal operating condition data of different operation batches together forms a time-varying three-dimensional normal operation dataset of the HVAC system.

[0071] The data from different batches over several consecutive days exhibit time-varying dynamic characteristics; and within each day, the data from a single batch also exhibits a wide range of non-stationary characteristics over time.

[0072] To address the issue that the operating data of HVAC systems over multiple consecutive days exhibits time-varying dynamic characteristics in batches, and that the operating data for each day also shows a wide range of non-stationary operation over time, a condition-driven method based on temperature variables is adopted to reconstruct the three-dimensional time-varying normal operating data of HVAC systems from the time axis to the condition axis. This effectively handles the non-stationary operation characteristics of HVAC systems in actual operation, which involve frequent and dynamic switching of multiple operating conditions.

[0073] For example, the specific process is as follows:

[0074] First, obtain the time-varying three-dimensional normal operation dataset X(I×K×J) of the HVAC system. Then, perform a slicing operation perpendicular to the time axis in the time-varying three-dimensional normal operation dataset X(I×K×J) to obtain two-dimensional time slices X at different times. k (I×J), where k (k=1,2,…,K) represents the sampling time, I represents the batch, and J represents the variable.

[0075] Then, in the k-th two-dimensional time slice X k Calculate the mean value of the temperature variable in variable J of (I×J). And used as an indicator variable, where k = 1, 2, ..., K. To convert time slices indicated by time [1, 2, ..., k, ... K] into mean values ​​of temperature variables. Prepare slices for the indicated temperature conditions, where T k This represents the mean of the k-th temperature variable.

[0076] Finally, the temperature condition slices are divided according to the mean of the temperature variable. Arranging the values ​​in ascending order yields the mean temperature variable on the temperature condition axis. Slice X of temperature conditions that change sequentially from small to large m (I×J), where X m (I×J), m=1,2,…,K represents the m-th temperature condition slice.

[0077] S202. Perform canonical variable analysis on the temperature condition slices to obtain normal data under multiple operating conditions based on the temperature condition slices. Specific steps include:

[0078] S2021. Standardize the temperature condition slices to obtain standard temperature slices.

[0079] S2022. Perform canonical variable analysis on the standard temperature slices and obtain the correlation coefficients between adjacent standard temperature slices.

[0080] S2023. Based on the correlation coefficient and preset threshold, cluster the standard temperature slices to obtain normal data for multiple operating conditions based on temperature condition slices.

[0081] Given that existing clustering models generally fail to consider the correlation information and dynamic characteristics between variables in HVAC system data, leading to inaccurate segmentation of multiple operating conditions in HVAC systems, this embodiment employs canonical variable analysis for multi-condition segmentation, focusing on the mean temperature variable. Each temperature condition slice X is arranged in ascending order. m The canonical variable analysis algorithm is implemented on (I×J), m=1,2,…,K respectively to fully extract the data features of the dynamic process. The temperature conditions with high similarity are clustered into the same working condition in turn to obtain multiple working condition data with large temperature differences, so as to realize the accurate division of the fault detection model under multiple working conditions.

[0082] For example, the specific process is as follows:

[0083] (1) Standardize the temperature condition slices: Standardize the temperature condition slices X obtained by reconstructing 3D data based on temperature conditions. m (I×J) is standardized to obtain a slice under standard temperature conditions. The formula is as follows:

[0084]

[0085] in, This represents the m-th standard temperature condition slice. Slice X representing temperature conditions m The mean of D(X) m ) represents the temperature condition slice X m The variance.

[0086] (2) Modeling based on temperature condition slices: Perform canonical variable analysis on standardized temperature condition slices. The m-th temperature condition slice is represented as... The (m+1)th temperature condition slice is represented as

[0087]

[0088]

[0089] Where, x iJ Let J represent the variables in the i-th row, and R be the variable in the i-th row. IJ This represents a state space with I batches and J variables.

[0090] Standard temperature condition slices The HanKel matrices Y1 and Y2 can be represented as:

[0091]

[0092]

[0093] The dimension of the columns of the HanKel matrix is ​​P = I - 2J + 1.

[0094] Matrix Y m And matrix Y m+1 The covariance matrix can be calculated as follows:

[0095]

[0096]

[0097] Matrix Y m And matrix Y m+1 The cross-covariance matrix is:

[0098]

[0099] The purpose of canonical variable analysis is to find the variable a T y m+1 With variable b T y mThe correlation coefficient ρ between them m (a,b) The largest a, b, where ρ m (a,b) can be expressed as:

[0100]

[0101] Where a and b represent the coefficients of the linear combination of variables in the HanKel matrix.

[0102] definition The typical variable analysis problem can be described in mathematical terms as follows:

[0103]

[0104]

[0105] The above optimization problem is solved by performing SVD (Singular Value Decomposition) on the HanKel matrix H:

[0106]

[0107] Where U and V are orthogonal matrices, and Δ represents the descending order of the eigenvalues ​​of the HanKel matrix. By solving the optimization problem, the maximum correlation coefficient ρ between the two temperature condition slices is obtained. m (a,b).

[0108] (3) Multi-condition division of temperature condition slices.

[0109] The canonical variable analysis algorithm was used to model each temperature condition slice, and the correlation coefficient ρ between the m-th and (m+1)-th adjacent temperature condition slices was calculated. m Set a threshold α; if the correlation coefficient of adjacent temperature condition slices satisfies ρ... m If (a,b)≥α, it indicates that the two temperature condition slices have a high degree of similarity, and the two temperature condition slices are clustered into the same condition. If ρ m (a,b)<α indicates that the feature similarity between two adjacent temperature condition slices is low. The m-th temperature condition slice is taken as the last temperature condition slice of the previous operating condition, and the (m+1)-th temperature condition slice is taken as the first temperature condition slice of the next operating condition. Canonical variable analysis is used to accurately divide the Z sub-operating condition models, facilitating the setting of different statistical control limits according to different operating conditions and improving the accuracy of HVAC fault detection.

[0110] S203. Input the normal data under multiple operating conditions into the corresponding neural component analysis models for processing to obtain the statistical control limits for each operating condition. This includes:

[0111] S2031. Input the normal data under multiple working conditions into the corresponding neural component analysis model respectively to obtain the non - linear principal component features of the normal data under each working condition;

[0112] S2032. Based on the non - linear principal component features, construct the T 2 statistic and the SPE statistic of the non - linear latent features;

[0113] S2033. According to the T 2 statistic and the SPE statistic, construct the control limits of the T 2 statistic and the control limits of the SPE statistic through kernel density estimation.

[0114] Regarding the non - linear problems of the data of each sub - working condition in the HVAC system, in this embodiment, the neural component analysis method is used to perform non - linear modeling on the data characteristics of different divided working conditions. By imposing an orthogonal constraint on the auto - encoder model, mutually independent principal component features are extracted to further improve the fault detection effect. The neural component analysis model includes a non - linear encoder and a linear decoder. This module uses the neural component analysis model to extract the non - linear principal component features of the sub - working conditions respectively by training a feed - forward neural network with orthogonal constraints.

[0115] As Figure 4 shown, the neural component analysis model consists of an input layer (n input nodes), a hidden layer (p nodes, p < n) and an output layer (n input nodes). Input the Z working - condition data sets obtained in the multi - working - condition division based on canonical variate analysis into the neural component analysis model to obtain the corresponding non - linear principal component features. Map the data of the input layer to the hidden layer through a non - linear neural network (i.e., non - linear encoder), and use the output of the hidden layer as the input to decode it to the output layer through orthogonal transformation (i.e., decoder) to obtain the non - linear principal component features with removed correlation.

[0116] Exemplarily, the specific process is as follows:

[0117] Assume that each working - condition data set has N data samples X i = [x i1 , x i2 , … x in , (i = 1, 2, …, N). Input the N data samples X i (i = 1, 2, …, N) of each working - condition data into the neural component analysis model respectively. The formula of the neural component analysis model is as follows:

[0118]

[0119] B T B = I P×P (14)

[0120] Where B represents a linear orthogonal transformation, containing three orthogonal bases, namely B = [b1, b2, b3], X i (i = 1, 2, ..., N) represents the i-th (i = 1, 2, ..., N) data sample of the working condition data, W represents the weights of the neural network, b represents the bias of the neural network, and let B = [b1, b2, ..., b p ],in, Then orthogonal constraint B T B = I means:

[0121]

[0122] Adding orthogonal constraint B to neural component analysis T B = I P×P This indicates that decoding latent features is an orthogonal reconstruction, which can significantly reduce the correlation between different variables. The score matrix for neural component analysis is constructed as follows:

[0123]

[0124] From equation (16), equation (13) in neural component analysis can be optimized to:

[0125]

[0126] The score matrix G contains key features for further fault detection analysis. Since the optimization problem in equation (17) is non-convex, the optimal W, b, and B are calculated iteratively. First, W and b are determined to obtain G, and then B is calculated through optimization as follows:

[0127]

[0128] Once B is obtained, W and b can be updated by solving the following optimization problem:

[0129]

[0130] The optimization problem in equation (19) can be solved by the backpropagation algorithm used in training the feedforward neural network. The solution to equation (18) is to determine an orthogonal matrix G to fit the input data matrix.

[0131] Construct T using the obtained features 2 Statistics and SPE statistics:

[0132] Feature g i T 2 The (Hotling) statistics are as follows:

[0133]

[0134] Among them, g i =g(X) i W,b) represents the i-th data sample X of the sub-condition in the normal operating condition dataset of the HVAC system. i The nonlinear principal component features of (i = 1, 2, ..., N) are given by ∑g. The relevant covariance matrix.

[0135] The SPE (Square Prediction Error) statistic for feature gi is as follows:

[0136] SPE i =||X i -g i B T || 2 (twenty one)

[0137] Based on normal operating condition data, since the prior distribution information of the nonlinear principal component features is unknown, kernel density estimation is used to construct T. 2 Control limits for the SPE statistic.

[0138] Let the density δ(·) be unknown. For nonlinear principal component features g1, g2, ... g N T 2 Statistic. T 2 The kernel density estimate of the statistic is:

[0139]

[0140] Where K(·) is a non-negative function with an integral of 1 and a mean of zero, h is the bandwidth parameter, and N is the number of samples in the sub-condition. Let SPE1, SPE2, ... SPE have unknown density η(·). N For X 11 ,X2,…X N The SPE statistic. The kernel density estimate of the SPE statistic is:

[0141]

[0142] Here, T 2 The kernel density estimate of the statistic is T. 2 The control limits for the SPE statistic are the kernel density estimates of the SPE statistic.

[0143] Furthermore, the real-time running data is input into the trained fault detection model for processing to obtain fault detection results.

[0144] Specifically, T is calculated based on real-time runtime data. 2The kernel density of the statistical measure and the SPE statistic is estimated and compared with the preset statistical control limits to obtain the fault detection results.

[0145] If T 2 The kernel density estimate of the statistic is greater than the pre-set T. 2 The control limits for the SPE statistic, or, if the kernel density estimate of the SPE statistic is greater than the preset control limits for the SPE statistic, then the HVAC system is malfunctioning.

[0146] For example, the statistics SPE for online test data test or like and This indicates that the test dataset X of the HVAC system test It's normal; otherwise, it indicates a malfunction.

[0147] To address the time-varying nature of operating conditions in HVAC systems, we found that although operating conditions vary uncertainly over time, the process characteristics are largely similar under the same conditions, potentially following a certain relationship. That is, the process characteristics of HVAC systems exhibit a certain regularity along the conditional direction. Therefore, this embodiment employs a condition-driven method to reconstruct the three-dimensional data of the HVAC system. Since temperature is the most direct manifestation of HVAC regulation, it is selected as the indicator variable for condition-driven processing. Time-slice data is converted into condition-slice data and arranged in ascending order of temperature values, thus reconstructing the three-dimensional time-varying data of the HVAC system from the time axis to the temperature axis.

[0148] To address the correlation and dynamic characteristics of data between temperature condition slices and to fully extract process features for multi-condition classification, canonical variable analysis (CVA) is performed on each temperature condition slice of the HVAC system. The CVA algorithm characterizes the similarity between two adjacent temperature condition slices by maximizing the correlation of linear combinations of feature variables arranged in ascending order of temperature mean. This clusters highly similar temperature condition slices into the same operating condition, achieving accurate multi-condition model partitioning. While the process features of temperature condition slices under the same operating condition are similar, those under different operating conditions differ. Establishing separate fault detection models for different operating conditions enables accurate fault detection in HVAC systems.

[0149] To address the significant nonlinear characteristics of the various operating condition data after the HVAC system is segmented, this invention employs neural component analysis (NCA) to establish nonlinear models for each segmented operating condition. By applying orthogonal constraints, independent nonlinear principal component features are extracted, eliminating redundant information between features. The NCA model comprises a nonlinear encoder and a linear decoder, designed to train a feedforward neural network with orthogonal constraints. On the normal dataset for each operating condition, a nonlinear neural network is first used as the encoder to extract nonlinear principal component features. Then, a linear orthogonal transformation is used to decode the nonlinear principal component features back to the original data space. Finally, the fault detection model is trained by minimizing the reconstruction error between the original and decoded data. Statistics are calculated based on the nonlinear principal component features extracted by the NCA model. Considering the unknown prior distribution information of the nonlinear principal component features, control limits are further calculated using kernel density estimation.

[0150] Example 2

[0151] This embodiment discloses a HVAC fault detection system based on conditional driving and neural component analysis, including:

[0152] The data acquisition module is used to acquire real-time operating data of the HVAC system.

[0153] The fault detection module is used to input real-time running data into the trained fault detection model for processing in order to obtain fault detection results.

[0154] The process of training the fault detection model includes:

[0155] Obtain time-varying three-dimensional normal operation data of HVAC system, and based on temperature condition drive, reconstruct the time-varying three-dimensional normal operation data from the time axis to the temperature condition axis to obtain temperature condition slices;

[0156] Perform canonical variable analysis on temperature condition slices to obtain normal data for multiple operating conditions based on temperature condition slices.

[0157] Normal data under multiple operating conditions are input into the neural component analysis model for processing to obtain the statistical control limits for each operating condition.

[0158] It should be noted that the data acquisition module and fault detection module described above correspond to the steps in Embodiment 1. The examples and application scenarios implemented by these modules and their corresponding steps are the same, but they are not limited to the content disclosed in Embodiment 1. It should also be noted that these modules, as part of the system, can be executed in a computer system, such as a set of computer-executable instructions.

[0159] Example 3

[0160] Embodiment 3 of the present invention provides an electronic device, including a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When the computer instructions are executed by the processor, they complete the steps of the above-mentioned HVAC fault detection method based on conditional driving and neural component analysis.

[0161] Example 4

[0162] Embodiment 4 of the present invention provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps of the above-described HVAC fault detection method based on conditional driving and neural component analysis.

[0163] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0164] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0165] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment, whereby a series of operational steps are performed to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0166] The descriptions of each embodiment in the above embodiments have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0167] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A HVAC fault detection method based on condition-driven and neural component analysis, characterized in that, The method comprises the following steps: acquiring real-time operation data of the HVAC; inputting the real-time operation data into a trained fault detection model for processing to obtain a fault detection result; wherein the process of training the fault detection model comprises: acquiring time-varying three-dimensional normal operation data of the HVAC, and reconstructing the time-varying three-dimensional normal operation data from a time axis to a temperature condition axis based on temperature condition driving to obtain temperature condition slices; performing typical variable analysis on the temperature condition slices to obtain multi-working-condition normal data divided based on the temperature condition slices; inputting the multi-working-condition normal data into neural component analysis models respectively for processing to obtain statistical control limits of each working condition; the process of reconstructing the time-varying three-dimensional normal operation data from the time axis to the temperature condition axis based on the temperature condition driving to obtain the temperature condition slices comprises: performing slice processing on the time-varying three-dimensional normal operation data perpendicular to the time axis to obtain two-dimensional time slices at different time instants; based on the two-dimensional time slices, calculating a temperature variable mean value in each two-dimensional time slice to obtain the temperature condition slices; arranging the temperature condition slices in an order of increasing temperature variable mean values to obtain the temperature condition slices based on the temperature condition axis.

2. The HVAC fault detection method based on condition-driven and neural component analysis as claimed in claim 1, wherein, the process of performing typical variable analysis on the temperature condition slices to obtain the multi-working-condition normal data divided based on the temperature condition slices comprises: performing standardization processing on the temperature condition slices to obtain standard temperature slices; performing typical variable analysis on the standard temperature slices to obtain correlation coefficients between adjacent standard temperature slices; based on the correlation coefficients and a preset threshold, clustering the standard temperature slices to obtain the multi-working-condition normal data divided based on the temperature condition slices.

3. The HVAC fault detection method based on condition-driven and neural component analysis as claimed in claim 1, wherein, the process of inputting the multi-working-condition normal data into neural component analysis models respectively for processing comprises: inputting the multi-working-condition normal data into corresponding neural component analysis models respectively to obtain nonlinear principal component features of the normal data of each working condition; based on the nonlinear principal component features, constructing a Hotelling statistic of nonlinear latent features and a squared prediction error statistic; based on the Hotelling statistic and the squared prediction error statistic, constructing a Hotelling statistic control limit and a squared prediction error statistic control limit through kernel density estimation.

4. The HVAC fault detection method based on condition-driven and neural component analysis as claimed in claim 1, wherein, the process of inputting the real-time operation data into the trained fault detection model for processing comprises: based on the real-time operation data, calculating kernel density estimations of the Hotelling statistic and the squared prediction error statistic, and comparing the kernel density estimations with preset statistical control limits to obtain the fault detection result.

5. The HVAC fault detection method based on condition-driven and neural component analysis as claimed in claim 4, wherein, if the kernel density estimation of the Hotelling statistic is greater than the preset Hotelling statistic control limit, or if the kernel density estimation of the squared prediction error statistic is greater than the preset squared prediction error statistic control limit, the HVAC has a fault.

6. The HVAC fault detection method based on condition-driven and neural component analysis as claimed in claim 1, wherein, the fault detection model comprises a nonlinear encoder and a decoder, the nonlinear encoder is configured to extract nonlinear principal component features according to working condition data; the decoder is configured to decode the nonlinear principal component features to original data space; the fault detection model is trained by minimizing reconstruction errors between the working condition data and the decoded data.

7. A HVAC fault detection system based on condition-driven and neural component analysis, characterized in that, The method comprises the following steps: a data acquisition module is configured to acquire real-time operation data of the HVAC; The fault detection module is configured to input real-time operation data into the trained fault detection model for processing to obtain a fault detection result. The process of training the fault detection model comprises: obtaining time-varying three-dimensional normal operation data of the HVAC, and reconstructing the time-varying three-dimensional normal operation data from a time axis to a temperature condition axis based on temperature condition driving to obtain temperature condition slices; performing typical variable analysis on the temperature condition slices to obtain multi-working-condition normal data divided based on the temperature condition slices; inputting the multi-working-condition normal data into a neural component analysis model for processing to obtain statistical quantity control limits of each working condition; the process of reconstructing the time-varying three-dimensional normal operation data from the time axis to the temperature condition axis based on the temperature condition driving to obtain the temperature condition slices comprises: performing slice processing on the time-varying three-dimensional normal operation data perpendicular to the time axis to obtain two-dimensional time slices at different time instants; based on the two-dimensional time slices, calculating a temperature variable mean value in each two-dimensional time slice to obtain the temperature condition slices; arranging the temperature condition slices in an order of increasing temperature variable mean values to obtain the temperature condition slices based on the temperature condition axis.

8. An electronic device, comprising: A computer program product comprising a memory and a processor and computer instructions stored on the memory and running on the processor, when the computer instructions are run by the processor, the steps of any one of claims 1-6 are completed.

9. A computer-readable storage medium, characterized in that, A computer program product for storing computer instructions, when the computer instructions are executed by the processor, the steps of any one of claims 1-6 are completed.

Images