Method, device, medium and product for detecting abnormality of multi-connected system

Abstract

The application belongs to the technical field of intelligent household appliances, and particularly relates to an abnormality detection method, device, medium and product of a multi-connected system. The method obtains operation parameters of the multi-connected system, and generates a mask vector corresponding to the operation parameters; based on the operation parameters and the mask vector, a dynamic feature vector of the multi-connected system is generated, the dynamic feature vector is input into a feature reconstruction model to obtain a reconstructed feature vector corresponding to the dynamic feature vector; based on the reconstructed feature vector, the dynamic feature vector and the mask vector, a reconstruction error of the operation parameters is determined, and in the case that the reconstruction error exceeds a preset error threshold, it is determined that the multi-connected system has an abnormality. The method solves the problems of non-uniformity of the operation data dimension and missing value interference of the multi-connected system through a dynamic mask mechanism, breaks through the dependence of traditional supervised learning on fixed dimension data and known fault types, and significantly improves the initiative and accuracy of operation and maintenance.

Description

Technical Field

[0001] This application belongs to the field of smart home appliance technology, specifically relating to an anomaly detection method, device, medium, and product for a multi-split air conditioning system. Background Technology

[0002] Multi-split air conditioning systems are highly efficient and flexible air conditioning control systems. Their core function is to achieve efficient regulation of temperature, humidity and air quality in different areas through the coordinated control of the outdoor unit and multiple indoor units.

[0003] As the application scenarios of multi-split air conditioning systems continue to expand, the complexity of the system also increases. The core components of the system include several key operating components such as compressors, condensers, expansion valves, and indoor unit fans, and the hardware configurations of different models of equipment vary. Among them, the inconsistency in sensor configuration is particularly prominent, with some models lacking temperature and pressure sensors.

[0004] However, the fault diagnosis algorithms of traditional multi-unit systems rely on fixed-dimensional data input, which cannot adapt to the data heterogeneity problem caused by different models of equipment. This leads to a significant decrease in the accuracy of fault diagnosis, making it difficult to meet the actual needs of system operation and maintenance. Summary of the Invention

[0005] This application provides an anomaly detection method, device, medium, and product for multi-split air conditioning systems, which solves the technical problem that traditional fault diagnosis algorithms for multi-split air conditioning systems rely on fixed-dimensional data input, cannot adapt to the data heterogeneity caused by different models of equipment, and thus lead to a significant decrease in the accuracy of fault diagnosis, making it difficult to meet the actual needs of system operation and maintenance.

[0006] Firstly, this application provides an anomaly detection method for a multi-unit air conditioning system, comprising:

[0007] Obtain the operating parameters of the multi-split air conditioning system and generate a mask vector corresponding to the operating parameters. The mask vector is used to represent the valid and invalid states of the operating parameters.

[0008] Based on the operating parameters and mask vector, a dynamic feature vector of the multi-unit system is generated. The dynamic feature vector is used to reflect the operating status of the multi-unit system under different feature dimensions.

[0009] The dynamic feature vector is input into the feature reconstruction model to obtain the reconstructed feature vector corresponding to the dynamic feature vector. The feature reconstruction model is used to learn the feature distribution law of the operating parameters of the multi-unit system in a healthy operating state.

[0010] Based on the reconstructed feature vector, dynamic feature vector, and mask vector, the reconstruction error of the operating parameters is determined. If the reconstruction error exceeds a preset error threshold, an anomaly is determined in the multi-unit system. The reconstruction error is used to quantify the degree of deviation between the dynamic feature vector and the reconstructed feature vector.

[0011] In one possible implementation, the running parameters include multi-dimensional feature values; generating a mask vector corresponding to the running parameters includes:

[0012] Determine a preset vector with the same dimension as the running parameters;

[0013] For any feature dimension, if the feature value is in a valid state, update the elements of the corresponding feature dimension of the preset vector to the first preset value;

[0014] If the feature value is invalid, update the elements of the feature dimension corresponding to the preset vector to the second preset value;

[0015] If all feature dimensions of the preset vector have been updated, the updated preset vector will be determined as the mask vector.

[0016] In one possible implementation, the feature reconstruction model includes an encoding branch and a decoding branch; the encoding branch is used to extract the latent space representation of the dynamic feature vector, and the decoding branch is used to reconstruct the features from the dynamic feature vector.

[0017] Inputting the dynamic feature vector and mask vector into the feature reconstruction model yields the reconstructed feature vector corresponding to the dynamic feature vector, including:

[0018] Input the dynamic feature vector into the encoding branch to obtain the latent vector after fusion encoding. The latent vector is a low-dimensional fusion feature representation vector generated after fusing multi-source high-dimensional features and performing dimensionality reduction encoding.

[0019] The latent vector is input into the decoding branch to obtain the reconstructed feature vector corresponding to the dynamic feature vector. The reconstructed feature vector has the same dimension as the dynamic feature vector.

[0020] In one possible implementation, the reconstruction error of the running parameters is determined based on the reconstructed feature vector, the dynamic feature vector, and the mask vector, including:

[0021] Determine the difference vector between the reconstructed feature vector and the dynamic feature vector;

[0022] The reconstruction error of the running parameters is determined based on the mask vector and the difference vector.

[0023] In one possible implementation, the method further includes:

[0024] Obtain the historical operating parameters of the multi-split air conditioning system, and the historical mask vector corresponding to the historical operating parameters;

[0025] Based on historical operating parameters, generate historical dynamic feature vectors and historical mask vectors for the multi-unit system;

[0026] A loss function based on historical mask vectors is constructed, and the autoencoder is pre-trained based on historical dynamic feature vectors with the optimization objective of minimizing the loss function, thus obtaining a feature reconstruction model.

[0027] In one possible implementation, the method further includes:

[0028] Based on the training data of the feature reconstruction model, a benchmark knowledge base for the multi-split air conditioning system is generated. The training data is used to represent a sample set of historical normal operating conditions of the multi-split air conditioning system under healthy operating conditions. The benchmark knowledge base includes the dynamic feature statistical benchmark and normal reconstruction error distribution of the multi-split air conditioning system under healthy operating conditions.

[0029] Among them, the dynamic feature statistical benchmark includes the benchmark mean and benchmark standard deviation for each feature dimension;

[0030] The normal reconstruction error distribution includes the set distribution of historical reconstruction errors corresponding to any historical dynamic feature vector in the training data.

[0031] In one possible implementation, after determining that an anomaly exists in the multi-split system, the method further includes:

[0032] For each feature dimension, based on the dynamic feature vector and the reconstructed feature vector, the reconstruction deviation of the corresponding feature dimension of the multi-connected system is determined. The reconstruction deviation is used to quantify the degree of deviation between the dynamic feature vector and the reconstructed feature vector under any feature dimension.

[0033] Based on operating parameters and dynamic characteristic statistical benchmarks, the statistical deviations of corresponding characteristic dimensions of the multi-unit system are determined;

[0034] According to the preset weight ratio, the reconstruction deviation and statistical deviation are weighted to obtain the anomaly score of the corresponding feature dimension of the multi-unit system. The anomaly score is used to quantify the degree of anomaly of different feature dimensions of the multi-unit system.

[0035] Secondly, this application provides an anomaly detection device for a multi-unit air conditioning system, comprising:

[0036] The acquisition module is used to acquire the operating parameters of the multi-unit system.

[0037] The processing module is used to generate mask vectors corresponding to the running parameters. The mask vectors are used to represent the valid and invalid states of the running parameters.

[0038] The processing module is also used to generate dynamic feature vectors for the multi-unit system based on operating parameters and mask vectors. These dynamic feature vectors reflect the operating status of the multi-unit system under different feature dimensions.

[0039] The processing module is also used to input the dynamic feature vector into the feature reconstruction model to obtain the reconstructed feature vector corresponding to the dynamic feature vector. The feature reconstruction model is used to learn the feature distribution law of the operating parameters of the multi-unit system in a healthy operating state.

[0040] The processing module is also used to determine the reconstruction error of the operating parameters based on the reconstructed feature vector, dynamic feature vector and mask vector, and to determine that there is an anomaly in the multi-unit system when the reconstruction error exceeds the preset error threshold. The reconstruction error is used to quantify the degree of deviation between the dynamic feature vector and the reconstructed feature vector.

[0041] In one possible implementation, the processing module is also used to determine a preset vector with the same dimension as the running parameters.

[0042] The processing module is also used to update the elements of the corresponding feature dimension of the preset vector to the first preset value when the feature value is in a valid state for any feature dimension.

[0043] The processing module is also used to update the elements of the feature dimension corresponding to the preset vector to the second preset value when the feature value is invalid.

[0044] The processing module is also used to determine the updated preset vector as the mask vector when all feature dimensions of the preset vector have been updated.

[0045] In one possible implementation, the processing module is further configured to input the dynamic feature vector into the encoding branch to obtain the fused encoded latent vector, wherein the latent vector is a low-dimensional fused feature representation vector generated by fusing multi-source high-dimensional features and performing dimensionality reduction encoding.

[0046] The processing module is also used to input the latent vector into the decoding branch to obtain the reconstructed feature vector corresponding to the dynamic feature vector. The reconstructed feature vector has the same dimension as the dynamic feature vector.

[0047] In one possible implementation, the processing module is further configured to determine the difference vector between the reconstructed feature vector and the dynamic feature vector.

[0048] The processing module is also used to determine the reconstruction error of the running parameters based on the mask vector and the difference vector.

[0049] In one possible implementation, the acquisition module is further configured to acquire the historical operating parameters of the multi-unit system, as well as the historical mask vector corresponding to the historical operating parameters.

[0050] The processing module is also used to generate historical dynamic feature vectors and historical mask vectors for multi-unit systems based on historical operating parameters.

[0051] The processing module is also used to construct a loss function based on historical mask vectors, and to pre-train the autoencoder based on historical dynamic feature vectors with the optimization objective of minimizing the loss function, thereby obtaining a feature reconstruction model.

[0052] In one possible implementation, the processing module is further configured to generate a benchmark knowledge base for the multi-split air conditioning system based on the training data of the feature reconstruction model. The training data is used to represent a sample set of historical normal operating conditions of the multi-split air conditioning system under healthy operating conditions. The benchmark knowledge base includes dynamic feature statistical benchmarks and normal reconstruction error distributions of the multi-split air conditioning system under healthy operating conditions.

[0053] Among them, the dynamic feature statistical benchmark includes the benchmark mean and benchmark standard deviation for each feature dimension;

[0054] The normal reconstruction error distribution includes the set distribution of historical reconstruction errors corresponding to any historical dynamic feature vector in the training data.

[0055] In one possible implementation, the processing module is further configured to determine the reconstruction deviation of the corresponding feature dimension of the multi-connected system for each feature dimension, based on the dynamic feature vector and the reconstructed feature vector. The reconstruction deviation is used to quantify the degree of deviation between the dynamic feature vector and the reconstructed feature vector under any feature dimension.

[0056] The processing module is also used to determine the statistical deviation of the corresponding feature dimensions of the multi-unit system based on the operating parameters and dynamic feature statistical benchmarks.

[0057] The processing module is also used to weight the reconstruction deviation and statistical deviation according to a preset weight ratio to obtain the anomaly score of the corresponding feature dimension of the multi-unit system. The anomaly score is used to quantify the degree of anomaly of different feature dimensions of the multi-unit system.

[0058] Thirdly, this application provides an electronic device, comprising:

[0059] The processor, and the memory that is in communication with the processor;

[0060] The memory stores instructions that the computer executes;

[0061] The processor executes computer execution instructions stored in memory to implement the anomaly detection method for a multi-connected system as described in the first aspect and various possible implementations of the first aspect.

[0062] Fourthly, this application provides a computer storage medium storing computer execution instructions, which are executed by a processor to implement an anomaly detection method for a multi-connected system as described in the first aspect and various possible implementations of the first aspect.

[0063] Fifthly, this application provides a program product, including a computer program, which, when executed by a processor, implements the anomaly detection method for a multi-connected system as described in the first aspect and various possible implementations of the first aspect.

[0064] The anomaly detection method for multi-split air conditioning systems provided in this application involves: acquiring the operating parameters of the multi-split air conditioning system and determining a preset vector with the same dimension as the operating parameters; for any feature dimension, if the feature value is valid, updating the elements of the corresponding feature dimension of the preset vector to a first preset value; if the feature value is invalid, updating the elements of the corresponding feature dimension of the preset vector to a second preset value; if all elements of all feature dimensions in the preset vector have been updated, determining the updated preset vector as a mask vector; generating a dynamic feature vector of the multi-split air conditioning system based on the operating parameters and the mask vector; inputting the dynamic feature vector into the encoding branch to obtain a fused encoded latent vector, and then inputting the latent vector into the decoding branch to obtain a reconstructed feature vector corresponding to the dynamic feature vector; determining the difference vector between the reconstructed feature vector and the dynamic feature vector, and determining the reconstruction error of the operating parameters based on the mask vector and the difference vector; and determining that the multi-split air conditioning system has an anomaly if the reconstruction error exceeds a preset error threshold. This method addresses the issues of inconsistent data dimensions and missing value interference in multi-unit systems through a dynamic masking mechanism. It avoids model bias caused by fixed-dimensional data input in traditional methods, breaks through the dependence of traditional supervised learning on fixed-dimensional data and known fault types, and not only improves the model's generalization ability across different equipment models, but also significantly enhances the initiative and accuracy of operation and maintenance. Attached Figure Description

[0065] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0066] Figure 1 This is a flowchart illustrating the anomaly detection method for the multi-split air conditioning system provided in this application. Figure 1 ;

[0067] Figure 2 This is a flowchart illustrating the anomaly detection method for the multi-split air conditioning system provided in this application. Figure 2 ;

[0068] Figure 3 This is a schematic diagram of the anomaly detection device for the multi-unit air conditioning system provided in this application;

[0069] Figure 4 This is a schematic diagram of the structure of the electronic device provided in this application.

[0070] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0071] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0072] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented, for example, in orders other than those illustrated or described herein.

[0073] In this application, the terms "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0074] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with relevant laws, regulations and standards, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0075] First, the terms used in this application will be explained.

[0076] Fully connected autoencoders are unsupervised neural network models built on fully connected layers. Their core objective is to compress input data into low-dimensional features (encoders), then reconstruct the original input through decoding, and learn the essential features of the data by minimizing the reconstruction error.

[0077] LeakyReLU activation function: An improved nonlinear activation function that modifies the linear unit. It preserves small gradients when the input is negative, solves the neuron inactivation problem of the traditional ReLU activation function, and ensures the continuity of gradients during backpropagation in deep neural networks.

[0078] Adam optimizer: Adaptive Moment Estimation is a first-order gradient optimization algorithm in the field of deep learning that adapts to the learning rate. It dynamically adjusts the update step size of each model parameter through adaptive calculation of first-order moment estimation (gradient mean) and second-order moment estimation (gradient variance), and is used to minimize the loss function and iteratively update network parameters during model training.

[0079] Multi-split air conditioning systems are highly efficient and flexible air conditioning control systems. Their core function is to achieve efficient regulation of temperature, humidity and air quality in different areas through the coordinated control of the outdoor unit and multiple indoor units.

[0080] As the application scenarios of multi-split air conditioning systems continue to expand, the complexity of the system also increases. The core components of the system include several key operating components such as compressors, condensers, expansion valves, and indoor unit fans, and the hardware configurations of different models of equipment vary. Among them, the inconsistency in sensor configuration is particularly prominent, with some models lacking temperature and pressure sensors.

[0081] Meanwhile, a large amount of operational data is generated during the operation of the multi-unit system. When a sensor is missing, the device firmware will fill the missing field with default values ​​(such as -999, 0xFFFF, etc.), resulting in a large amount of invalid information in the operational data and forming heterogeneous data characteristics.

[0082] However, the fault diagnosis algorithms of traditional multi-unit systems rely on fixed-dimensional data input, which cannot adapt to the data heterogeneity problem caused by different models of equipment. This leads to a significant decrease in the accuracy of fault diagnosis, making it difficult to meet the actual needs of system operation and maintenance.

[0083] Furthermore, traditional diagnostic methods cannot effectively identify early performance degradation (such as reduced compressor efficiency or pipe blockage) that occurs during long-term system operation, nor can they detect unknown fault types such as new component failures. This results in passive operation and maintenance work, delayed fault response, and an inability to guarantee the stable and reliable operation of the system.

[0084] To address the aforementioned issues, this application provides an anomaly detection method for multi-split air conditioning systems. By employing a missing data processing mechanism based on dynamic masking, it unifies the input data dimensions of different equipment models, effectively distinguishing between actual measured values ​​and default values ​​caused by sensor deficiencies, thus achieving standardized input data processing. Through an autoencoder model learning the normal operating modes of the multi-split system under mask weighting constraints, it constructs a generalized fault detection capability based solely on valid data. Combining masking information and statistical benchmarks, it achieves anomaly feature localization and severity assessment, forming a closed-loop intelligent diagnostic process. This method, through the collaborative design of a dynamic masking mechanism and an autoencoder model, solves the problems of inconsistent data dimensions, missing value interference, and insufficient model generalization ability in multi-split systems. It overcomes the dependence of traditional supervised learning on fixed-dimensional data and known fault types, achieving cross-model generalization, unified detection of unknown faults, and interpretable diagnosis, significantly improving the initiative and accuracy of operation and maintenance.

[0085] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0086] Figure 1 A flowchart illustrating the anomaly detection method for a multi-unit air conditioning system provided in this application embodiment. Figure 1 The executing entity in this embodiment can be, for example, the control center of a multi-unit system. Figure 1 As shown, the anomaly detection method for a multi-unit air conditioning system provided in this embodiment includes:

[0087] S101: Obtain the operating parameters of the multi-unit system and generate a mask vector corresponding to the operating parameters.

[0088] The mask vector is used to represent the valid and invalid states of the running parameters.

[0089] Understandably, operating parameters refer to the dynamic monitoring data collected in real time during the operation of a multi-split air conditioning system. These parameters are the direct basis for fault diagnosis and include, for example, compressor exhaust temperature, total system input current, and indoor unit coil temperature.

[0090] In this embodiment, after obtaining the operating parameters, the control center of the multi-unit system stores them in a one-dimensional vector storage format. This one-dimensional vector storage format is a standardized data storage format with a fixed order, fixed length, pure numerical type, and one-dimensional linear structure, specifically used to store the real-time operating parameters of the multi-unit system.

[0091] A mask vector is a vector whose dimensions perfectly match those of the runtime parameters stored in a one-dimensional vector storage format. The elements in this mask vector are used to mark the data validity status of the runtime parameter feature values. Therefore, by using fixed values ​​for different feature dimensions in the mask vector, the valid and invalid states of corresponding runtime parameter feature values ​​can be distinguished. Specifically, if the first... If the feature value corresponding to the first feature dimension is valid, it indicates that the data acquired by that feature dimension is a real sensor measurement value. In this case, the element of the mask vector corresponding to that feature dimension can be marked as 1; if the first feature dimension is valid, the second feature dimension is valid. If the feature value corresponding to a feature dimension is invalid, it means that the data obtained by that feature dimension represents the default value of the sensor's missing value. In this case, the element of the mask vector corresponding to that feature dimension can be recorded as 0.

[0092] For example, the operating parameters specifically include: "Thermodynamic cycle parameters: indoor unit coil temperature (supply / return water or refrigerant temperature), outdoor unit coil temperature (condensing / evaporating temperature), environmental and demand parameters (indoor temperature in each area, outdoor ambient temperature), control execution parameters (opening of indoor unit electronic expansion valve, opening of outdoor unit main valve), electrical and core performance parameters (total system input current, voltage, real-time power, compressor discharge temperature, compressor operating frequency)". The operating parameters acquired in each instance are collected in a fixed order and then concatenated into a one-dimensional numerical vector, thus obtaining the operating parameters of the multi-split air conditioning system in vector storage format, which can also be called the initial feature vector. And the corresponding vector dimension is d; if the mask vector obtained in this time is [1,0,1], it means that the temperature and current sensors are valid, but the pressure sensor is missing.

[0093] In this embodiment of the application, the operating parameters acquired by the multi-split system have a unique sampling timestamp, that is, the operating parameters represent the instantaneous operating status of the multi-split system at any sampling time.

[0094] S102: Generate dynamic feature vectors for the multi-unit system based on operating parameters and mask vectors.

[0095] Among them, the dynamic feature vector is used to reflect the operating status of the multi-unit system under different feature dimensions.

[0096] In this embodiment, the mask vector can determine the valid feature values ​​in the currently acquired operating parameters. During the process of generating the dynamic feature vector of the multi-unit system, the feature values ​​marked as invalid by the mask vector are processed in a targeted manner, the feature values ​​marked as valid are standardized, and the construction of the dynamic feature vector of the multi-unit system is realized. This ensures that the multi-unit systems with different sensor configurations have a unified input dimension, while retaining the distribution characteristics of valid data and eliminating the data heterogeneity problem caused by sensor absence.

[0097] In one possible implementation, the step of standardizing the running parameters based on the mask vector specifically includes: obtaining the mean and standard deviation of the valid data fields in the running parameters; performing Z-score normalization on the running parameters based on the mean and standard deviation for any valid data field; and setting the value of the default value filling field to zero to generate a dynamic feature vector.

[0098] Z-score normalization transforms data into a distribution with a mean of 0 and a standard deviation of 1 by subtracting the mean and dividing by the standard deviation. For example, the temperature value 25℃ is transformed into (25-20) / 5=1.

[0099] For example, Z-score normalization is performed on the valid data fields in the operating parameters, including: first, calculating the mean and standard deviation of the valid data fields (e.g., the mean of the temperature sensor is 20℃ and the standard deviation is 5℃); for each valid data field (feature with a mask of 1), normalization is performed using the Z-score normalization formula ((current value - mean) / standard deviation); for default value filler fields (features with a mask of 0), they are directly set to zero, thereby ensuring that multi-unit systems with different sensor configurations have a uniform input scale while preserving the distribution characteristics of valid data.

[0100] S103: Input the dynamic feature vector into the feature reconstruction model to obtain the reconstructed feature vector corresponding to the dynamic feature vector.

[0101] Among them, the feature reconstruction model is used to learn the feature distribution pattern of the operating parameters of the multi-unit system in a healthy operating state.

[0102] In this embodiment, the feature reconstruction model is a generative model based on unsupervised learning. This model uses the operating samples of the multi-unit system in a healthy operating state as training data to learn the joint probability distribution of dynamic feature vectors in a healthy operating state. The model extracts implicit features of the data through an encoding structure and then restores and generates reconstructed feature vectors through a decoding structure. Its training optimization objective is to minimize the error between the original dynamic feature vector and the reconstructed feature vector, so that the reconstruction result infinitely approximates the original input features, thereby accurately capturing the distribution law of operating parameters in a healthy operating state of the multi-unit system. This feature reconstruction model can be, for example, trained based on a fully connected autoencoder.

[0103] The dynamic feature vector is input into the feature reconstruction model (such as a learning model trained based on a fully connected autoencoder). This model extracts implicit features through the encoder and reconstructs the dynamic feature vector through the decoder to generate the reconstructed feature vector.

[0104] For example, the dynamic feature vector corresponding to the current sampling time. The input encoder extracts a low-dimensional latent vector (z) through a fully connected layer and the LeakyReLU activation function, and outputs the mean (z). ) and variance The latent vector (z) is then input into the decoder, and the dynamic features are reconstructed through a reverse fully connected layer to obtain the reconstructed feature vector at the current sampling time. ).

[0105] S104: Based on the reconstructed feature vector, dynamic feature vector, and mask vector, determine the reconstruction error of the operating parameters, and if the reconstruction error exceeds the preset error threshold, determine that there is an anomaly in the multi-unit system.

[0106] The reconstruction error is used to quantify the degree of deviation between the dynamic feature vector and the reconstructed feature vector.

[0107] Understandably, the preset error threshold is the critical value set by the feature reconstruction model to determine the reconstruction error. It is a quantitative standard to distinguish whether the operating state of the multi-unit system is normal or abnormal. When the real-time reconstruction error output by the model exceeds this value, the operating state of the multi-unit system can be determined to be abnormal. This preset error threshold can be the maximum historical reconstruction error that the multi-unit system has under normal operating conditions (healthy operating state) during the feature reconstruction model training process, which also represents the maximum reconstruction error that the model may produce under healthy conditions.

[0108] Reconstruction error is a quantitative indicator calculated by comparing the difference between the dynamic feature vector and the reconstructed feature vector. It is used to characterize the degree of deviation between the current operating state and the healthy baseline.

[0109] For example, the reconfiguration error of the multi-unit system is calculated using the following formula 1, as detailed below:

[0110]

[0111] in, The reconstruction error at the current sampling time (in scoring form); The first dynamic feature vector Each feature dimension The nth element in the mask vector corresponding to the current sampling time One element, This represents the dynamic feature vector at the current sampling time. This is the reconstructed feature vector output by the feature reconstruction model at the current sampling time.

[0112] If the preset error threshold is ,and Greater than This indicates that the operating status of the multi-unit system at the current sampling time deviates from the characteristic distribution of normal operating conditions, and there is a possibility of abnormal operation or equipment failure.

[0113] In one possible implementation, a preset error threshold is used. Typically, this feature is used to reconstruct the abnormal scores of all normal samples in the training set during model training. of quantiles (e.g.) Right now ;like Greater than If so, the current status of the multi-unit system is determined to be abnormal.

[0114] In this embodiment, the operating parameters and equipment parameters acquired by the multi-split system have unique sampling timestamps. Therefore, in the process of determining the reconstruction error of the multi-split system, the data used for the reconstruction error calculation at that time is time-sensitive and can only represent whether there is an anomaly in the instantaneous operating state of the multi-split system at the corresponding sampling time. In addition, the anomaly detection processes at different sampling times are independent of each other.

[0115] The anomaly detection method for multi-split air conditioning systems provided in this embodiment obtains the operating parameters of the multi-split air conditioning system and generates a mask vector corresponding to the operating parameters. Based on the operating parameters and the mask vector, a dynamic feature vector of the multi-split air conditioning system is generated. The dynamic feature vector is input into a feature reconstruction model to obtain a reconstructed feature vector corresponding to the dynamic feature vector. Based on the reconstructed feature vector, the dynamic feature vector, and the mask vector, the reconstruction error of the operating parameters is determined. If the reconstruction error exceeds a preset error threshold, an anomaly is determined to exist in the multi-split air conditioning system. This method solves the problems of inconsistent data dimensions and missing value interference in the operating data of multi-split air conditioning systems through a dynamic mask mechanism. It avoids the model bias caused by fixed-dimensional data input in traditional methods and breaks through the dependence of traditional supervised learning on fixed-dimensional data and known fault types, significantly improving the initiative and accuracy of operation and maintenance.

[0116] Figure 2 A flowchart illustrating the anomaly detection method for a multi-unit air conditioning system provided in this application embodiment. Figure 2 .like Figure 2 As shown, in this embodiment... Figure 1 Based on the embodiments, the anomaly detection method for multi-split air conditioning systems is described in detail. The anomaly detection method for multi-split air conditioning systems shown in this embodiment includes:

[0117] S201: Obtain the operating parameters of the multi-unit system.

[0118] Step S201 is similar to step S101 above, and will not be repeated here.

[0119] S202: Determine a preset vector with the same dimension as the running parameters.

[0120] S203: For any feature dimension, if the feature value is in a valid state, update the elements of the feature dimension corresponding to the preset vector to the first preset value.

[0121] S204: When the feature value is invalid, update the elements of the feature dimension corresponding to the preset vector to the second preset value.

[0122] S205: If all feature dimensions of the preset vector have been updated, the updated preset vector will be determined as the mask vector.

[0123] The first preset value is used to identify the value of the valid data field, for example, it can be 1; the second preset value is used to identify the value of the default value fill field, for example, it can be 0.

[0124] Understandably, the preset vector is the initial carrier of the mask vector. This preset vector is the initialization base vector, and its dimension is the same as that of the running parameters. By assigning / updating the elements in the preset vector dimension by dimension, the mask vector corresponding to the running parameters is finally generated.

[0125] For example, at the current sampling time, the multi-unit system uses the currently acquired initial feature vector. Generate a binary mask vector of the same dimension. , Each element in the array has a value of 0 or 1; specifically, if The first in If each characteristic value is a true sensor measurement, then The value at position i is 1; if The first in Each feature value represents a default value indicating a missing sensor value. The value at position i is 0.

[0126] In one possible implementation, the logic for identifying default values ​​is based on a preset invalid value list; this preset invalid value list is used to identify a list of default values ​​and outlier ranges in the operating parameters, such as a default value containing [-999, 0xFFFF] and an outlier range where the outdoor temperature exceeds [-50, 60].

[0127] Understandably, the validity of the feature values ​​of each feature dimension in the running parameters is determined by a preset invalid value list. Specifically, the preset invalid value list includes default values ​​(such as -999) and abnormal value ranges (such as outdoor temperatures exceeding [-50, 60]). For each feature dimension, if its value range exceeds the preset invalid value list (such as a temperature value of -999 or 1000), it is determined to be a default value field, and the corresponding position of the mask vector is updated to the second preset value (such as 0). Otherwise, it is determined to be a valid data field, and the corresponding position of the mask vector is updated to the first preset value (such as 1).

[0128] S206: Generate dynamic feature vectors for multi-unit systems based on operating parameters and mask vectors.

[0129] Step S206 is similar to step S102 above, and will not be described again here.

[0130] S207: Input the dynamic feature vector into the encoding branch to obtain the fused encoded latent vector.

[0131] S208: Input the latent vector into the decoding branch to obtain the reconstructed feature vector corresponding to the dynamic feature vector.

[0132] In this embodiment, the feature reconstruction model includes an encoding branch and a decoding branch; the encoding branch is used to extract the latent space representation of the dynamic feature vector, and the decoding branch is used to reconstruct the features of the dynamic feature vector; the latent vector is a low-dimensional fused feature representation vector generated by fusing multi-source high-dimensional features and performing dimensionality reduction encoding, and the reconstructed feature vector has the same dimension as the dynamic feature vector.

[0133] Specifically, the encoding branch is a neural network module used to map the input data into a low-dimensional latent space representation. For example, it can be a multi-layer network composed of fully connected layers and LeakyReLU activation functions. The latent vector is a low-dimensional feature vector output by the encoding branch, used to represent the core features of the input data, such as a vector containing key information such as compressor status and condensing pressure. The decoding branch is a neural network module used to reconstruct the latent space representation into a dynamic feature vector. For example, it can be a multi-layer network composed of fully connected layers and LeakyReLU activation functions.

[0134] The dynamic feature vector is compressed into a latent vector through the encoding branch to extract the core features of the multi-unit system's operating state at the corresponding sampling time. The latent vector is reconstructed through the decoding branch to obtain the corresponding reconstructed feature vector. The reconstructed feature vector and the input dynamic feature vector are completely identical in dimensional structure, but there are essential differences in numerical values, data distribution, and information content. The reconstructed feature vector is the fitted data generated by the model based on the learned healthy operating mode of the multi-unit system, rather than the original collected real operating parameters.

[0135] For example, a feature reconstruction model trained based on a fully connected autoencoder compresses features into a low-dimensional latent space through an encoding branch, and then reconstructs the original data through a decoding branch; wherein, the input to the encoding branch is a standardized dynamic feature vector. (Dimension d), this encoding branch is the encoder in a fully connected autoencoder. The encoder network consists of an input layer, multiple hidden layers, and an output layer. The input layer receives a d-dimensional vector; the hidden layers are multi-layer networks consisting of three fully connected layers and a LeakyReLU activation function. The first layer is a fully connected layer with an output dimension of 128, followed by the LeakyReLU activation function to accelerate model training; the second layer is a fully connected layer with an output dimension of 64, followed by the LeakyReLU activation function; the third layer is a fully connected layer with an output dimension of 32, followed by the LeakyReLU activation function. The formula for LeakyReLU is: ,in It is the input of the function (i.e., the output value of the previous layer of neurons). It is a preset fixed slope coefficient that takes values ​​in the (0,1) interval, usually set to a small value, such as 0.01; the output layer takes the 32-dimensional features output by the above hidden layer as input, and outputs the mean vector through two parallel, fully connected sub-layers without activation functions. And the dimension is (e.g., 16), and the log-variance vector The dimension is also Then, through reparameterization from the distribution Mid-sampling yields latent variables (dimension) ), which is also known as implicit vector.

[0136] This decoding branch is also the decoder in the fully connected autoencoder, which uses the sampled latent variables As input, its dimension is d; its network structure is as follows: the input layer receives the hidden vector. , dimension The first hidden layer is a fully connected layer, and the input... The first layer is a fully connected layer with a 32-dimensional input and a 64-dimensional output, followed by a LeakyReLU activation function. The second layer is also a fully connected layer with a 64-dimensional input and a 128-dimensional output, followed by a LeakyReLU activation function. The third layer is a fully connected layer with a 128-dimensional input and a d-dimensional output, using a linear activation function to generate a reconstructed dynamic feature vector. .

[0137] In one possible implementation, the historical operating parameters of the multi-unit system and the historical mask vector corresponding to the historical operating parameters are obtained; based on the historical operating parameters, the historical dynamic feature vector and the historical mask vector of the multi-unit system are generated; a loss function based on the historical mask vector is constructed, and based on the historical dynamic feature vector, the autoencoder is pre-trained with minimizing the loss function as the optimization objective to obtain the feature reconstruction model.

[0138] Understandably, the loss function is the reconstruction loss, used to measure the difference between the input vector and the reconstructed vector of the feature reconstruction model; and the essence of model training is to continuously adjust the network parameters through backpropagation to minimize the output value of the loss function, so that the model's prediction results are as close as possible to the true values; therefore, the training process of the feature reconstruction model is to minimize the loss function as the optimization objective.

[0139] In this embodiment, for the historical operating data of the multi-unit system under normal operating conditions (healthy operating state), the corresponding historical operating parameters and historical mask vectors are determined, and a sample set (including historical operating parameters and historical mask vectors) that can be used for model training is generated. By standardizing the historical operating parameters in the sample set, the differences in the dimensions of different parameters are eliminated, ensuring the uniformity and comparability of the model input data. Then, the feature distribution of normal data is learned through a fully connected autoencoder, so that the reconstruction error can be used to determine whether the multi-unit system deviates from the healthy operating state.

[0140] For example, the sample set is used as the training set for the model. The fully connected autoencoder is pre-trained to obtain a feature reconstruction model; that is, the feature reconstruction model uses the fully connected autoencoder to learn the latent space distribution of normal data and complete the feature reconstruction. Represents the first in the sample set Each sampling time, For the sample set The historical dynamic feature vector corresponding to each sampling time point For the sample set The historical mask vector corresponding to each sampling moment.

[0141] In one possible implementation, the training objective of the feature reconstruction model is to minimize the loss function. The loss function As shown in Formula 2 below:

[0142]

[0143] in, The reconstruction loss of the model is used to measure the difference between the model output and the true target. For fully connected self-encoders The reconstructed output; This indicates element-wise multiplication. This indicates the calculation of the square of the L2 norm (i.e., the sum of the squares of the elements). It is the first The number of 1s in a sample mask vector, i.e., the number of effective features, is epsilon, which is a very small positive number to prevent division by zero errors.

[0144] The physical meaning of this loss function is that the feature reconstruction model only needs to work hard to reconstruct features with a mask of 1 (real existence). For features with a mask of 0 (missing sensor), no matter what their output value is, they do not contribute to the loss gradient, so that the model can adaptively learn effective data patterns under different device configurations and achieve cross-model generalization of the model.

[0145] In one possible implementation, a benchmark knowledge base for the multi-unit system is generated based on the training data of the feature reconstruction model.

[0146] The training data represents a sample set of historical normal operating conditions of the multi-split air conditioning system under healthy operating conditions. The benchmark knowledge base includes the dynamic characteristic statistical benchmark and normal reconfiguration error distribution of the multi-split air conditioning system under healthy operating conditions.

[0147] The dynamic feature statistical benchmark includes the benchmark mean and benchmark standard deviation for each feature dimension; the normal reconstruction error distribution includes the set distribution of historical reconstruction errors corresponding to any historical dynamic feature vector in the training data.

[0148] Understandably, the training process of the feature reconstruction model uses a massive amount of validated normal operating condition samples, i.e., the training set. (After training is completed, save the model parameters obtained in this training and establish a benchmark knowledge base corresponding to the multi-unit system to store the dynamic feature statistical benchmark and normal reconstruction error distribution under the healthy operation state of the multi-unit system.)

[0149] In this embodiment, the dynamic feature statistical benchmark is used when standardizing each valid dynamic feature. and Its dynamic feature statistical benchmark is used to standardize operating parameters, providing a unified scale for data normalization in online detection, ensuring that real-time data and training data are in the same metric space, and serving as the basis for all subsequent quantitative comparisons; the normal reconstruction error distribution records all training samples The values ​​constitute the set of abnormal scores for all normal samples. Its function is to characterize the intrinsic fluctuation level of a multi-unit system under healthy conditions and to determine an adaptive and robust dynamic threshold for anomaly detection, so as to effectively distinguish between normal fluctuations and real anomalies and improve the reliability of detection.

[0150] For example, the Adam optimizer is used to train the model until convergence; after the model training is complete, the model parameters are saved; simultaneously, based on the entire training set... For each feature Calculate the mean of its valid data and standard deviation The data is stored in the knowledge base as a normal benchmark for subsequent anomaly localization.

[0151] S209: Determine the difference vector between the reconstructed feature vector and the dynamic feature vector.

[0152] S210: Determine the reconstruction error of the running parameters based on the mask vector and the difference vector.

[0153] S211: If the reconstruction error exceeds the preset error threshold, it is determined that there is an anomaly in the multi-unit system.

[0154] The difference vector refers to the element-wise difference between the dynamic feature vector and the reconstructed feature vector, which is used to characterize the degree to which the parameters deviate from the healthy baseline; the reconstruction error is used to quantify the degree of abnormality of each dynamic feature.

[0155] By calculating the difference vector between the reconstructed feature vector and the dynamic feature vector, and combining it with the mask vector to calculate the reconstruction error of the effective feature value, and when the reconstruction error exceeds the preset error threshold, it is determined that there is an anomaly in the multi-unit system, thereby realizing the localization of the abnormal dynamic features and further achieving the accurate localization of the core dynamic features of the anomaly.

[0156] In one possible implementation, after determining that there is an anomaly in the multi-split air conditioning system, for each feature dimension, based on the dynamic feature vector and the reconstructed feature vector, the reconstruction deviation of the corresponding feature dimension of the multi-split air conditioning system is determined; based on the operating parameters and the dynamic feature statistical benchmark, the statistical deviation of the corresponding feature dimension of the multi-split air conditioning system is determined; and the reconstruction deviation and the statistical deviation are weighted according to a preset weight ratio to obtain the anomaly score of the corresponding feature dimension of the multi-split air conditioning system.

[0157] Among them, reconstruction bias is used to quantify the degree of deviation between dynamic feature vector and reconstructed feature vector under any feature dimension, statistical bias refers to the degree of deviation between the original data of dynamic feature of any feature dimension and historical statistical benchmark, and anomaly score is used to quantify the degree of anomaly of different feature dimensions of multi-unit system.

[0158] For example, the reconstruction deviation is calculated using the following formula 3:

[0159]

[0160] in, For the first The absolute value of the reconstruction bias corresponding to each feature dimension At the current sampling time, the first Standardized runtime data for each feature dimension; At the current sampling time, the first Reconstructed data based on each feature dimension.

[0161] Statistical deviation is calculated using the following formula 4:

[0162]

[0163] in, For the first Statistical bias of each feature dimension At the current sampling time, the first The original running parameters for each feature dimension were not standardized. and The dynamic feature statistical benchmark is retrieved from the knowledge base to reflect the degree to which the original value of the feature deviates from the historical normal value benchmark.

[0164] according to and The pre-defined weighting ratio is used to calculate the anomaly score of the current multi-split system using the following formula 5:

[0165]

[0166] in, For the first A comprehensive anomaly score for the runtime parameters corresponding to each feature dimension. This is the weighting coefficient (corresponding to the preset weight ratio). The maximum value of the reconstruction bias across all feature dimensions. This is the statistical bias cutoff value.

[0167] In one possible implementation, a comprehensive anomaly score is given for each feature dimension determined in this instance. The features are sorted in descending order; the features ranked first are the core suspicious parameters that caused this anomaly.

[0168] For example, based on the list of abnormal features obtained in descending order, the key abnormal features are located. According to its statistical bias The severity level is determined based on the specific level of abnormality identified in that instance. In the case of determining the corresponding number The anomaly level of the feature in this dimension is level 1. At this point, the feature value exceeds the normal range by 2-3 standard deviations, meaning there is a slight deviation between the output of the feature reconstruction model and the input dynamic features, indicating a minor anomaly. An alert can be issued for this dimension's anomaly to remind users that the multi-connector system may have early degradation or minor misalignment issues. In the case of determining the corresponding number The anomaly level of the feature dimension is level 2. At this point, there is a significant deviation between the output of the feature reconstruction model and the input dynamic features, indicating a significant anomaly. An alert can be issued for this dimension's anomaly to remind the user of a clear performance failure or component degradation. In the case of determining the corresponding number The anomaly level of the dimension's features is level 3. At this point, there is a serious deviation between the output of the feature reconstruction model and the input dynamic features, that is, there is a serious anomaly. For the anomaly of this dimension, an emergency alarm can be issued first to indicate that the hardware of the multi-unit system is damaged or the system is about to / has already stopped.

[0169] In one possible implementation, a diagnostic report is generated based on the anomaly confidence level, key anomaly characteristics, anomaly level, and maintenance guidance recommendations.

[0170] Understandably, the anomaly confidence level refers to the degree of exceeding a relative threshold; for example, it could be... The main abnormal features refer to those described in the following order: The sorted list, each item containing the feature name, current value, normal range, and statistical deviation. and reconstruction deviation Anomaly level refers to the level (e.g., level 1, level 2, level 3) and description corresponding to each dynamic feature; maintenance guidance suggestions refer to preliminary maintenance suggestions mapped from the knowledge base based on the combination and level of anomaly features (e.g., "Focus on checking the compressor's intake and exhaust ports and current"); diagnostic report refers to the final output structured report.

[0171] The anomaly detection method for multi-unit systems provided in this application addresses data heterogeneity through dynamic masking, achieves unsupervised general anomaly perception through autoencoder, and provides accurate and interpretable fault location and assessment through multi-dimensional analysis, forming a complete intelligent diagnostic closed loop.

[0172] The anomaly detection method for multi-split air conditioning systems provided in this embodiment obtains the operating parameters of the multi-split air conditioning system and determines a preset vector with the same dimension as the operating parameters. For any feature dimension, if the feature value is valid, the elements of the feature dimension corresponding to the preset vector are updated to a first preset value; if the feature value is invalid, the elements of the feature dimension corresponding to the preset vector are updated to a second preset value. If all elements of the feature dimensions in the preset vector have been updated, the updated preset vector is determined as a mask vector. Based on the operating parameters and the mask vector, a dynamic feature vector of the multi-split air conditioning system is generated. The dynamic feature vector is input into the encoding branch to obtain a fused encoded latent vector, and then the latent vector is input into the decoding branch to obtain a reconstructed feature vector corresponding to the dynamic feature vector. The difference vector between the reconstructed feature vector and the dynamic feature vector is determined, and based on the mask vector and the difference vector, the reconstruction error of the operating parameters is determined. If the reconstruction error exceeds a preset error threshold, an anomaly is determined in the multi-split air conditioning system. This method addresses the issues of inconsistent data dimensions and missing value interference in multi-unit systems through a dynamic masking mechanism. It avoids model bias caused by fixed-dimensional data input in traditional methods, breaks through the dependence of traditional supervised learning on fixed-dimensional data and known fault types, and not only improves the model's generalization ability across different equipment models, but also significantly enhances the initiative and accuracy of operation and maintenance.

[0173] Figure 3 This is a schematic diagram of the anomaly detection device for the multi-split air conditioning system provided in this application. Figure 3 As shown, this application provides an anomaly detection device for a multi-split air conditioning system. The anomaly detection device 300 for the multi-split air conditioning system includes:

[0174] The acquisition module 301 is used to acquire the operating parameters of the multi-unit system.

[0175] The processing module 302 is used to generate a mask vector corresponding to the running parameters. The mask vector is used to represent the valid and invalid states of the running parameters.

[0176] The processing module 302 is also used to generate dynamic feature vectors of the multi-unit system based on the operating parameters and the mask vector. The dynamic feature vectors are used to reflect the operating status of the multi-unit system under different feature dimensions.

[0177] The processing module 302 is also used to input the dynamic feature vector into the feature reconstruction model to obtain the reconstructed feature vector corresponding to the dynamic feature vector. The feature reconstruction model is used to learn the feature distribution law of the operating parameters of the multi-unit system in a healthy operating state.

[0178] The processing module 302 is also used to determine the reconstruction error of the operating parameters based on the reconstructed feature vector, the dynamic feature vector and the mask vector, and to determine that there is an anomaly in the multi-unit system when the reconstruction error exceeds a preset error threshold. The reconstruction error is used to quantify the degree of deviation between the dynamic feature vector and the reconstructed feature vector.

[0179] In one possible implementation, the processing module 302 is further configured to determine a preset vector of the same dimension as the running parameters.

[0180] The processing module 302 is also used to update the elements of the feature dimension corresponding to the preset vector to the first preset value when the feature value is in a valid state for any feature dimension.

[0181] The processing module 302 is also used to update the elements of the feature dimension corresponding to the preset vector to the second preset value when the feature value is invalid.

[0182] The processing module 302 is also used to determine the updated preset vector as the mask vector when all feature dimensions of the preset vector have been updated.

[0183] In one possible implementation, the processing module 302 is further configured to input the dynamic feature vector into the encoding branch to obtain the fused encoded latent vector, wherein the latent vector is a low-dimensional fused feature representation vector generated after fusing and dimensionality-reducing encoding of multi-source high-dimensional features.

[0184] The processing module 302 is also used to input the latent vector into the decoding branch to obtain the reconstructed feature vector corresponding to the dynamic feature vector, and the reconstructed feature vector has the same dimension as the dynamic feature vector.

[0185] In one possible implementation, the processing module 302 is further configured to determine the difference vector between the reconstructed feature vector and the dynamic feature vector.

[0186] The processing module 302 is also used to determine the reconstruction error of the running parameters based on the mask vector and the difference vector.

[0187] In one possible implementation, the acquisition module 301 is further configured to acquire the historical operating parameters of the multi-unit system and the historical mask vector corresponding to the historical operating parameters.

[0188] The processing module 302 is also used to generate historical dynamic feature vectors and historical mask vectors of the multi-unit system based on historical operating parameters.

[0189] The processing module 302 is also used to construct a loss function based on historical mask vectors, and to pre-train the autoencoder based on historical dynamic feature vectors with minimizing the loss function as the optimization objective, so as to obtain a feature reconstruction model.

[0190] In one possible implementation, the processing module 302 is further configured to generate a benchmark knowledge base for the multi-unit system based on the training data of the feature reconstruction model. The training data is used to represent a sample set of historical normal operating conditions of the multi-unit system under healthy operating conditions. The benchmark knowledge base includes dynamic feature statistical benchmarks and normal reconstruction error distribution of the multi-unit system under healthy operating conditions.

[0191] Among them, the dynamic feature statistical benchmark includes the benchmark mean and benchmark standard deviation for each feature dimension;

[0192] The normal reconstruction error distribution includes the set distribution of historical reconstruction errors corresponding to any historical dynamic feature vector in the training data.

[0193] In one possible implementation, the processing module 302 is further configured to determine the reconstruction deviation of the corresponding feature dimension of the multi-connected system for each feature dimension, based on the dynamic feature vector and the reconstructed feature vector. The reconstruction deviation is used to quantify the degree of deviation between the dynamic feature vector and the reconstructed feature vector under any feature dimension.

[0194] The processing module 302 is also used to determine the statistical deviation of the corresponding feature dimension of the multi-unit system based on the operating parameters and dynamic feature statistical benchmarks.

[0195] The processing module 302 is also used to perform weighted processing on the reconstruction deviation and statistical deviation according to the preset weight ratio to obtain the anomaly score of the corresponding feature dimension of the multi-unit system. The anomaly score is used to quantify the degree of anomaly of different feature dimensions of the multi-unit system.

[0196] Figure 4 A schematic diagram of the structure of the electronic device provided in this application. Figure 4 As shown, this application provides an electronic device 400, which includes a receiver 401, a transmitter 402, a processor 403, and a memory 404.

[0197] Receiver 401 is used to receive instructions and data;

[0198] Transmitter 402 is used to send commands and data;

[0199] Memory 404 is used to store instructions executed by the computer;

[0200] Processor 403 is used to execute computer execution instructions stored in memory 404 to implement the various steps performed by the anomaly detection method for the multi-unit system in the above embodiments. For details, please refer to the relevant descriptions in the foregoing embodiments of the anomaly detection method for the multi-unit system.

[0201] Optionally, the memory 404 can be either standalone or integrated with the processor 403.

[0202] When the memory 404 is set up independently, the electronic device also includes a bus for connecting the memory 404 and the processor 403.

[0203] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement an anomaly detection method for a multi-connected system as described above in the electronic device.

[0204] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the anomaly detection method for a multi-unit system according to any of the foregoing embodiments.

[0205] It will be understood by those skilled in the art that all or some of the steps, systems, or apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software may be distributed on a computer-readable medium, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0206] The technical solutions of this application have been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it is readily understood by those skilled in the art that the scope of protection of this application is obviously not limited to these specific embodiments. The above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for detecting anomalies in a multi-split air conditioning system, characterized in that, include: The operating parameters of the multi-unit system are obtained, and a mask vector corresponding to the operating parameters is generated. The mask vector is used to characterize the valid and invalid states of the operating parameters. Based on the operating parameters and the mask vector, a dynamic feature vector of the multi-unit system is generated. The dynamic feature vector is used to reflect the operating status of the multi-unit system under different feature dimensions. The dynamic feature vector is input into the feature reconstruction model to obtain the reconstructed feature vector corresponding to the dynamic feature vector. The feature reconstruction model is used to learn the feature distribution law of the operating parameters of the multi-unit system in a healthy operating state. Based on the reconstructed feature vector, the dynamic feature vector, and the mask vector, the reconstruction error of the operating parameters is determined, and if the reconstruction error exceeds a preset error threshold, it is determined that the multi-unit system has an anomaly. The reconstruction error is used to quantify the degree of deviation between the dynamic feature vector and the reconstructed feature vector.

2. The method according to claim 1, characterized in that, The operating parameters include multi-dimensional feature values; generating the mask vector corresponding to the operating parameters includes: Determine a preset vector with the same dimension as the operating parameters; For any feature dimension, if the feature value is in a valid state, the elements of the feature dimension corresponding to the preset vector are updated to the first preset value; If the feature value is invalid, the elements of the feature dimension corresponding to the preset vector are updated to the second preset value; If all feature dimensions of the preset vector have been updated, the updated preset vector is determined as the mask vector.

3. The method according to claim 2, characterized in that, The feature reconstruction model includes an encoding branch and a decoding branch; the encoding branch is used to extract the latent space representation of the dynamic feature vector, and the decoding branch is used to reconstruct the features of the dynamic feature vector. The step of inputting the dynamic feature vector and the mask vector into the feature reconstruction model to obtain the reconstructed feature vector corresponding to the dynamic feature vector includes: The dynamic feature vector is input into the encoding branch to obtain the fused encoding latent vector, which is a low-dimensional fused feature representation vector generated by fusing multi-source high-dimensional features and performing dimensionality reduction encoding. The latent vector is input into the decoding branch to obtain the reconstructed feature vector corresponding to the dynamic feature vector. The reconstructed feature vector has the same dimension as the dynamic feature vector.

4. The method according to claim 3, characterized in that, The step of determining the reconstruction error of the running parameters based on the reconstructed feature vector, the dynamic feature vector, and the mask vector includes: Determine the difference vector between the reconstructed feature vector and the dynamic feature vector; Based on the mask vector and the difference vector, the reconstruction error of the running parameters is determined.

5. The method according to claim 4, characterized in that, The method further includes: Obtain the historical operating parameters of the multi-unit system, and the historical mask vector corresponding to the historical operating parameters; Based on the historical operating parameters, the historical dynamic feature vector and historical mask vector of the multi-unit system are generated; A loss function based on the historical mask vector is constructed, and the autoencoder is pre-trained based on the historical dynamic feature vector with the goal of minimizing the loss function, to obtain the feature reconstruction model.

6. The method according to claim 5, characterized in that, The method further includes: Based on the training data of the feature reconstruction model, a benchmark knowledge base for the multi-split air conditioning system is generated. The training data is used to represent a sample set of historical normal operating conditions of the multi-split air conditioning system under healthy operating conditions. The benchmark knowledge base includes the dynamic feature statistical benchmark and normal reconstruction error distribution of the multi-split air conditioning system under healthy operating conditions. The dynamic feature statistical benchmark includes the benchmark mean and benchmark standard deviation for each feature dimension. The normal reconstruction error distribution includes the set distribution of historical reconstruction errors corresponding to any historical dynamic feature vector in the training data.

7. The method according to claim 6, characterized in that, After determining that the multi-split air conditioning system has an anomaly, the method further includes: For each feature dimension, based on the dynamic feature vector and the reconstructed feature vector, the reconstruction deviation of the corresponding feature dimension of the multi-connector system is determined. The reconstruction deviation is used to quantify the degree of deviation between the dynamic feature vector and the reconstructed feature vector under any feature dimension. Based on the operating parameters and the dynamic feature statistical benchmark, the statistical deviation of the corresponding feature dimension of the multi-unit system is determined; The reconstruction deviation and the statistical deviation are weighted according to a preset weight ratio to obtain the anomaly score of the corresponding feature dimension of the multi-unit system. The anomaly score is used to quantify the degree of anomaly of different feature dimensions of the multi-unit system.

8. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the method as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-7.

10. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method described in any one of claims 1-7.

Images