Operation fault monitoring method and equipment of HVAC system based on data analysis

Abstract

The invention provides an operation fault monitoring method and equipment of an HVAC system based on data analysis, and relates to the technical field of data analysis. The method comprises the following steps: firstly, based on harmonic semantic information in equipment vibration time sequence data, carrying out semantic enhancement on the equipment vibration time sequence data to form an equipment vibration semantic vector; then, based on surge semantic information in the equipment sound time sequence data, performing semantic enhancement on the equipment sound time sequence data to form an equipment sound semantic vector; then, semantic mining is carried out on the equipment temperature time sequence data, and an equipment temperature semantic vector is formed; further, fusing the equipment vibration semantic vector, the equipment sound semantic vector and the equipment temperature semantic vector to form an equipment operation semantic vector; and finally, predicting and outputting operation fault data based on the equipment operation semantic vector. Based on the method, the problem of relatively low reliability of operation fault monitoring in the prior art can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and in particular, to a method and device for monitoring the operation faults of an HVAC system based on data analysis. Background Art

[0002] The HVAC system, which is a heating, ventilation, and air conditioning system, is a comprehensive facility used to regulate and control the temperature, humidity, and air quality inside a building and is applied in many scenarios. For example, it can be used as an environmental guarantee system for an intelligent low-temperature operation clean room or as an environmental guarantee system for an intelligent constant-temperature biological culture clean room. However, in the prior art, the research and development of the HVAC system mainly focus on the accuracy of control algorithms and energy-saving requirements. However, in the application scenarios as described above, since the HVAC system needs to provide continuous, stable, and effective control, it is particularly important to monitor the faults of the HVAC system. However, in the prior art, the faults are mainly detected manually at regular intervals or by simply comparing the operation parameters in one or more dimensions with empirical thresholds. In fact, both of these two methods are difficult to effectively monitor the operation faults, and thus it is also difficult to perform corresponding maintenance in a timely manner, resulting in the inability to achieve continuous, stable, and effective operation. That is to say, in the prior art, there is a problem that the reliability of operation fault monitoring is relatively low. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide a method and device for monitoring the operation faults of an HVAC system based on data analysis to improve the problem of relatively low reliability of operation fault monitoring existing in the prior art.

[0004] To achieve the above purpose, the embodiments of the present invention adopt the following technical solutions: A method for monitoring the operation faults of an HVAC system based on data analysis includes: Obtaining the device vibration time series data, device sound time series data, and device temperature time series data of a target device in a target HVAC system; Semantically enhancing the device vibration time series data based on the harmonic semantic information in the device vibration time series data to form a device vibration semantic vector; Semantically enhancing the device sound time series data based on the surge semantic information in the device sound time series data to form a device sound semantic vector; Performing semantic mining on the device temperature time series data to form a device temperature semantic vector; Fuse the device vibration semantic vector, the device sound semantic vector, and the device temperature semantic vector to form a device operation semantic vector; Predict and output the operation fault data of the target device based on the device operation semantic vector, where the operation fault data is used to characterize the probability of the target device having a fault.

[0005] In some preferred embodiments, in the above-mentioned operation fault monitoring method of the HVAC system based on data analysis, the step of fusing the device vibration semantic vector, the device sound semantic vector, and the device temperature semantic vector to form a device operation semantic vector includes: Perform a first fusion on the device temperature semantic vector and the device vibration semantic vector to form a first operation semantic vector, where the first fusion includes multiple fusion stages; Perform a second fusion on the device temperature semantic vector and the device sound semantic vector to form a second operation semantic vector, where the second fusion includes multiple fusion stages; Perform a convolutional fusion on the first operation semantic vector and the second operation semantic vector to form a device operation semantic vector.

[0006] In some preferred embodiments, in the above-mentioned operation fault monitoring method of the HVAC system based on data analysis, the step of performing a first fusion on the device temperature semantic vector and the device vibration semantic vector to form a first operation semantic vector includes: Deeply mine the vibration temperature fusion vector at the i-th time step to form the vibration temperature depth vector at the i-th time step, where when the i-th time step is the first time step among multiple time steps, the vibration temperature fusion vector at the i-th time step is the device vibration semantic vector; Perform semantic fusion based on the vibration temperature depth vector at the i-th time step and the device temperature semantic vector to form the fusion semantic vector at the i-th time step; Form a mapped semantic vector based on the mapping parameter distribution at the i-th time step and the fusion semantic vector at the i-th time step; Perform upsampling on the mapped semantic vector to form an upsampled semantic vector; Link the vibration temperature fusion vector at the i-th time step to the upsampled semantic vector to form the vibration temperature fusion vector at the i+1-th time step; Determine the first operation semantic vector based on the vibration temperature fusion vector at the last time step among the multiple time steps.

[0007] In some preferred embodiments, in the above-mentioned method for monitoring the operation faults of an HVAC system based on data analysis, the step of deeply mining the vibration-temperature fusion vector at the i-th time step to form the vibration-temperature deep vector at the i-th time step includes: Downsample the vibration-temperature fusion vector at the i-th time step to form a downsampled fusion vector; Segment the downsampled fusion vector to form multiple local fusion vectors, and serialize the multiple local fusion vectors to form a local fusion vector sequence; Perform sequence correlation mining on the local fusion vector sequence to form the vibration-temperature deep vector at the i-th time step, where the sequence correlation mining includes channel splicing and channel convolution.

[0008] In some preferred embodiments, in the above-mentioned method for monitoring the operation faults of an HVAC system based on data analysis, the step of downsampling the vibration-temperature fusion vector at the i-th time step to form a downsampled fusion vector includes: Perform pooling operations on the vibration-temperature fusion vector at the i-th time step at multiple scales to form pooling fusion vectors at multiple scales; Perform convolution compression on the vibration-temperature fusion vector at the i-th time step to form a convolution fusion vector; Perform channel splicing on the pooling fusion vectors at multiple scales and the convolution fusion vector, and perform channel convolution on the result of the channel splicing to form the downsampled fusion vector of the vibration-temperature fusion vector at the i-th time step.

[0009] In some preferred embodiments, in the above-mentioned method for monitoring the operation faults of an HVAC system based on data analysis, the step of performing semantic fusion based on the vibration-temperature deep vector at the i-th time step and the device temperature semantic vector to form the fusion semantic vector at the i-th time step includes: Perform semantic space transformation on the device temperature semantic vector to form a device temperature transformation vector, and perform gated adjustment on the device temperature transformation vector and the vibration-temperature fusion vector at the i-th time step to form a device temperature adjustment vector, where the gated parameter corresponding to the gated adjustment is mapped based on the vibration-temperature fusion vector at the i-th time step; Perform summation or mean calculation on the device temperature adjustment vector and the vibration-temperature deep vector to form a preliminary fusion semantic vector; Perform deep convolution and non-linear activation on the preliminary fusion semantic vector to form the fusion semantic vector at the i-th time step.

[0010] In some preferred embodiments, in the above-mentioned method for monitoring the operation faults of the HVAC system based on data analysis, the step of forming a mapping semantic vector according to the mapping parameter distribution at the i-th time step and the fused semantic vector at the i-th time step includes: Determine the mapping parameter distribution at the i-th time step. Wherein, when the i-th time step is the first time step among the multiple time steps, each parameter in the mapping parameter distribution at the i-th time step is 1; when the i-th time step is other than the first time step among the multiple time steps, the mapping parameter distribution at the i-th time step is determined according to the mapping parameter distribution at the previous time step; Perform a bitwise multiplication operation on the mapping parameter distribution at the i-th time step and the fused semantic vector at the i-th time step to form a mapping semantic vector.

[0011] In some preferred embodiments, in the above-mentioned method for monitoring the operation faults of the HVAC system based on data analysis, the step of enhancing the semantics of the equipment vibration time series data based on the harmonic semantic information in the equipment vibration time series data to form an equipment vibration semantic vector includes: Perform a time-frequency transformation on the equipment vibration time series data to form an equipment vibration spectrogram, and based on the equipment vibration spectrogram, determine the harmonic frequencies corresponding to the equipment vibration time series data; Filter the equipment vibration time series data based on the harmonic frequencies to form equipment vibration harmonic data, and perform a time-frequency transformation on the equipment vibration harmonic data to form a harmonic vibration spectrogram; Perform convolution mining on the equipment vibration spectrogram and the harmonic vibration spectrogram respectively to form an equipment vibration spectrum vector and a harmonic vibration spectrum vector; Based on the harmonic vibration spectrum vector, perform significant feature mining on the equipment vibration spectrum vector to form an equipment vibration semantic vector, where the significant feature mining includes at least one of gated adjustment and cross-attention processing, and the gated parameter corresponding to the gated adjustment is mapped based on the harmonic vibration spectrum vector.

[0012] In some preferred embodiments, in the above-mentioned method for monitoring the operation faults of the HVAC system based on data analysis, the step of enhancing the semantics of the equipment sound time series data based on the surge semantic information in the equipment sound time series data to form an equipment sound semantic vector includes: Filter the equipment sound time series data based on the surge frequency to form equipment sound filtered data, perform a time-frequency transformation on the equipment sound filtered data to form a harmonic sound spectrogram, and perform a time-frequency transformation on the equipment sound time series data to form an equipment sound spectrogram; Perform convolutional mining on the harmonic sound spectrogram and the device sound spectrogram respectively to form a harmonic sound spectral vector and a device sound spectral vector; Based on the harmonic sound spectral vector, perform significant feature mining on the device sound spectral vector to form a device sound semantic vector, where the significant feature mining includes at least one of gated adjustment and cross-attention processing, and the gating parameter corresponding to the gated adjustment is mapped based on the harmonic sound spectral vector.

[0013] An embodiment of the present invention also provides an electronic device, including a processor and a memory. The memory is used to store a computer program, and the processor is used to execute the computer program to implement the above-mentioned method for monitoring the operation faults of an HVAC system based on data analysis.

[0014] For the method and device for monitoring the operation faults of an HVAC system based on data analysis provided by the embodiments of the present invention, first, based on the harmonic semantic information in the device vibration time-series data, perform semantic enhancement on the device vibration time-series data to form a device vibration semantic vector; then, based on the surge semantic information in the device sound time-series data, perform semantic enhancement on the device sound time-series data to form a device sound semantic vector; then, perform semantic mining on the device temperature time-series data to form a device temperature semantic vector; further, fuse the device vibration semantic vector, the device sound semantic vector, and the device temperature semantic vector to form a device operation semantic vector; finally, predict and output operation fault data based on the device operation semantic vector. Based on the above method, in the first aspect, since the device time-series data in the three dimensions of vibration, sound, and temperature are semantically fused, the formed device operation semantic vector can comprehensively represent the operation process of the target device. In the second aspect, since the semantic information in the vibration dimension is enhanced based on the harmonic semantic information, and the harmonic has a good representation effect on the unbalanced situation inside the device (such as a compressor), therefore, when the device vibration semantic vector represents the global semantic information of the vibration, it can also importantly represent some internal anomalies. In the third aspect, since the semantic information in the sound dimension is semantically enhanced based on the surge semantic information, and the surge is a phenomenon of periodic fluctuations in pressure and flow rate caused by dynamic blockage of the gas flow path during the operation of the device (compressor), therefore, the surge plays an important role in the semantic representation of some abnormal situations. Therefore, when the device sound semantic vector represents the global semantic information of the sound, it can also importantly represent some anomalies of the device. Based on this, it is possible to comprehensively mine the potential semantic information and focus on the important semantic information during the operation process of the target device, thereby improving the reliability of fault prediction and further improving the relatively low reliability of operation fault monitoring existing in the prior art.

[0015] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following specifically presents preferred embodiments and, in conjunction with the accompanying drawings, provides a detailed description as follows. Description of the Drawings

[0016] Figure 1 It is a block diagram of an electronic device provided by an embodiment of the present invention.

[0017] Figure 2 It is a flowchart of each module included in an operation fault monitoring device of an HVAC system based on data analysis provided by an embodiment of the present invention.

[0018] Figure 3 It is a schematic diagram of each step included in an operation fault monitoring method of an HVAC system based on data analysis provided by an embodiment of the present invention.

[0019] Figure 4 It is a schematic diagram of semantic enhancement provided by an embodiment of the present invention.

[0020] Figure 5 It is a schematic diagram of the first fusion provided by an embodiment of the present invention.

[0021] Figure 6 It is a schematic diagram of sequence association mining provided by an embodiment of the present invention. Detailed Embodiments

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Usually, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0023] Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed present invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0024] As Figure 1 shown, an embodiment of the present invention provides an electronic device. Among them, the electronic device may include a memory and a processor.

[0025] Specifically, the memory and the processor are electrically connected directly or indirectly to achieve data transmission or interaction. For example, they can be electrically connected to each other through one or more communication buses or signal lines. The memory may store at least one software functional module (computer program) that can exist in the form of software or firmware. The processor can be used to execute the executable computer program stored in the memory, so as to implement the operation fault monitoring method of the HVAC system based on data analysis provided by the embodiments of the present invention (as described later). For specific content, reference can be made to the relevant descriptions later.

[0026] In addition, the above software functional module can be each module included in the operation fault monitoring device of the HVAC system based on data analysis, such as Figure 2 shown in the figure. Among them, each module specifically includes: A timing data acquisition module, configured to acquire the device vibration timing data, device sound timing data, and device temperature timing data of a target device in a target HVAC system; A harmonic semantic enhancement module, configured to perform semantic enhancement on the device vibration timing data based on the harmonic semantic information in the device vibration timing data to form a device vibration semantic vector; A surge semantic enhancement module, configured to perform semantic enhancement on the device sound timing data based on the surge semantic information in the device sound timing data to form a device sound semantic vector; A semantic mining module, configured to perform semantic mining on the device temperature timing data to form a device temperature semantic vector; A semantic fusion module, configured to fuse the device vibration semantic vector, the device sound semantic vector, and the device temperature semantic vector to form a device operation semantic vector; A fault prediction module, configured to predict and output the operation fault data of the target device based on the device operation semantic vector, where the operation fault data is used to represent the probability that the target device has a fault.

[0027] Optionally, the memory may be, but is not limited to, a random access memory (RAM), a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), etc.

[0028] Optionally, the processor may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), a System on Chip (SoC), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0029] And, Figure 1 The structure shown is only illustrative, and the electronic device may also include more or fewer components than those shown in Figure 1 or have a different configuration from that shown in Figure 1 For example, it may include a communication unit for information interaction with other devices (such as various sensors).

[0030] Among them, in an alternative example, the electronic device may be a server with data processing capabilities.

[0031] In combination with Figure 3 , the embodiment of the present invention also provides an operation fault monitoring method for an HVAC system based on data analysis, which can be applied to the above-mentioned electronic device. Among them, the method steps defined by the process related to the operation fault monitoring method for the HVAC system based on data analysis can be implemented by the electronic device. The following will elaborate on the Figure 3 specific process shown in detail.

[0032] Step S110, obtain the device vibration time series data, device sound time series data, and device temperature time series data of the target device in the target HVAC system.

[0033] In the embodiment of the present invention, the electronic device can obtain the device vibration time series data, device sound time series data, and device temperature time series data of the target device in the target HVAC system. Exemplarily, the target device may be a compressor (which is responsible for driving the refrigerant cycle to ensure that heat can be effectively transferred and released, and plays a key role in the air conditioning system and is the core component for realizing the refrigerant cycle). Thus, data can be collected during the operation of the compressor through a vibration sensor, a sound sensor, and a temperature sensor provided on the compressor, so as to respectively form the corresponding device vibration time series data, device sound time series data, and device temperature time series data.

[0034] Step S120: Based on the harmonic semantic information in the device vibration time series data, perform semantic enhancement on the device vibration time series data to form a device vibration semantic vector.

[0035] In an embodiment of the present invention, after obtaining the device vibration time series data, the electronic device can perform semantic enhancement on the device vibration time series data based on the harmonic semantic information in the device vibration time series data to form a device vibration semantic vector. That is to say, the global semantic information of the device vibration can be semantically enhanced based on the harmonic semantic information, so that the device vibration semantic vector can not only represent the global semantic information of the device vibration, but also importantly represent the corresponding harmonic semantic information, taking into account the comprehensiveness and accuracy of semantic representation.

[0036] Step S130: Based on the surge semantic information in the device sound time series data, perform semantic enhancement on the device sound time series data to form a device sound semantic vector.

[0037] In an embodiment of the present invention, after obtaining the device sound time series data, the electronic device can perform semantic enhancement on the device sound time series data based on the surge semantic information in the device sound time series data to form a device sound semantic vector. That is to say, the global semantic information of the device sound can be semantically enhanced based on the surge semantic information, so that the device sound semantic vector can not only represent the global semantic information of the device sound, but also importantly represent the corresponding surge semantic information, taking into account the comprehensiveness and accuracy of semantic representation.

[0038] Step S140: Perform semantic mining on the device temperature time series data to form a device temperature semantic vector.

[0039] In an embodiment of the present invention, after obtaining the device temperature time series data, the electronic device can perform semantic mining on the device temperature time series data to form a device temperature semantic vector. That is to say, directly mine the global semantic information of the device temperature time series data.

[0040] Step S150: Fuse the device vibration semantic vector, the device sound semantic vector, and the device temperature semantic vector to form a device operation semantic vector.

[0041] In an embodiment of the present invention, after forming the device vibration semantic vector, the device sound semantic vector, and the device temperature semantic vector, the electronic device can fuse the device vibration semantic vector, the device sound semantic vector, and the device temperature semantic vector to form a device operation semantic vector. That is to say, the semantic information in three dimensions can be fused to obtain multi-dimensional semantic information during operation.

[0042] Step S160: Predict and output the operation fault data of the target device based on the operation semantic vector of the device.

[0043] In the embodiment of the present invention, after forming the operation semantic vector of the device, the electronic device can predict and output the operation fault data of the target device based on the operation semantic vector of the device. Among them, the operation fault data is used to characterize the probability of the target device having a fault, such as 0-1. It should be noted that in other embodiments, the operation data of the related devices of the target device can also be combined to adjust the operation semantic vector of the device, and then, based on the adjusted semantic vector, prediction and output are performed. For example, when the target device is a compressor, the related devices can be a condenser and an evaporator, etc. It is also possible to collect and perform semantic fusion on the data of the above three dimensions for each device to form the corresponding operation semantic vector of the device. Then, based on mechanisms such as cross-attention, the operation semantic vector corresponding to each related device and the operation semantic vector corresponding to the target device can be fused, and then prediction and output are performed based on the fused vector. It should be noted that the relevance of the related devices lies in: If the condenser is blocked or has poor heat dissipation, the refrigerant cannot effectively release heat, causing the compressor to overheat, and the overheating of the compressor may cause damage to internal components, such as bearings or motors; If the evaporator is blocked or cannot effectively dissipate heat, it will cause abnormal refrigerant pressure and too low suction pressure of the compressor, thus affecting the compressor efficiency and possibly shortening the compressor life.

[0044] Based on the above method, in the first aspect, since the device time-series data in the three dimensions of vibration, sound, and temperature are semantically fused, the formed device operation semantic vector can comprehensively represent the operation process of the target device. In the second aspect, since the semantic information of the vibration dimension is enhanced based on harmonic semantic information, and harmonics has a good representational effect on the unbalanced conditions existing inside the device (such as a compressor), therefore, when the device vibration semantic vector represents the global semantic information of vibration, it can also importantly represent some internal abnormalities. In the third aspect, since the semantic information of the sound dimension is semantically enhanced based on surge semantic information, and surge is a phenomenon of periodic fluctuations in pressure and flow rate caused by dynamic blockage of the gas flow passage during the operation of the device (compressor), therefore, surge plays an important role in the semantic representation of some abnormal conditions. Thus, when the device sound semantic vector represents the global semantic information of sound, it can also importantly represent some abnormalities of the device. Based on this, it is possible to comprehensively mine the potential semantic information and focus on the important semantic information during the operation process of the target device, thereby improving the reliability of fault prediction and further improving the problem of relatively low reliability of operation fault monitoring existing in the prior art.

[0045] In the first part, regarding step S110, it should be noted that the specific method of obtaining the device vibration time-series data, device sound time-series data, and device temperature time-series data of the target device in the target HVAC system is not limited and can be selected according to actual needs.

[0046] For example, in an alternative implementation manner, the electronic device can communicate with the corresponding sensors in real time to obtain the corresponding time-series data, and then perform subsequent processing to achieve timely fault prediction. For another example, in another alternative implementation manner, the electronic device can also obtain the corresponding time-series data from the database and then perform subsequent processing, such as periodically predicting faults. In this way, after each sensor forms the corresponding time-series data, it can be first stored in the corresponding database.

[0047] In the second part, regarding step S120, it should be noted that the specific method of semantically enhancing the device vibration time-series data is not limited and can be selected according to actual needs.

[0048] For example, in an alternative embodiment, semantic mining in the time domain can be separately performed on the device vibration time series data and the harmonic components in the device vibration time series data (for example, it can be implemented by LSTM (Long Short-Term Memory) and GRU (Gated Recurrent Unit)), so as to obtain two corresponding semantic vectors. Then, the two semantic vectors can be concatenated, summed, averaged, or processed by attention, etc., so as to obtain the device vibration semantic vector.

[0049] For another example, in another alternative embodiment, in order to ensure that effective fusion of harmonic semantic information can be achieved during semantic enhancement, thereby improving the semantic representation ability of the formed device vibration semantic vector, the above step S120 may further include the following implementable content (in combination with Figure 4 shown): In the first step, time-frequency transformation (such as Fourier transform) can be performed on the device vibration time series data to form a device vibration spectrogram, and based on the device vibration spectrogram, the harmonic frequencies corresponding to the device vibration time series data can be determined. For example, the fundamental frequency can be determined first, and then, based on n times the fundamental frequency, each harmonic frequency can be obtained, such as 2 times the fundamental frequency, 3 times the fundamental frequency, etc.; In the second step, based on the harmonic frequencies, filtering can be performed on the device vibration time series data (for example, a corresponding filter can be designed based on the harmonic frequencies for filtering so that information of other frequencies is filtered out), forming device vibration harmonic data, and time-frequency transformation (such as Fourier transform) is performed on the device vibration harmonic data to form a harmonic vibration spectrogram; In the third step, convolutional mining can be separately performed on the device vibration spectrogram and the harmonic vibration spectrogram to form a device vibration spectral vector and a harmonic vibration spectral vector; In the fourth step, based on the harmonic vibration spectrum vector, significant feature mining can be performed on the device vibration spectrum vector to form a device vibration semantic vector. The significant feature mining includes at least one of gated adjustment and cross-attention processing. The gating parameter corresponding to the gated adjustment is mapped based on the harmonic vibration spectrum vector. For example, the harmonic vibration spectrum vector can be linearly transformed (e.g., implemented through a fully connected network layer), and then the result of the linear transformation is non-linearly activated to obtain the gating parameter. Then, the gating parameter and the device vibration spectrum vector can be bitwise multiplied, and the result of the bitwise multiplication and the device vibration spectrum vector can be averaged or added to obtain the device vibration semantic vector. Additionally, during cross-attention processing, the result of the cross-attention processing and the device vibration spectrum vector can be averaged or added to obtain the device vibration semantic vector. And during cross-attention processing, the query vector corresponds to the harmonic vibration spectrum vector, and the key vector and value vector correspond to the device vibration spectrum vector (the specific processing process can refer to relevant existing technologies). Additionally, the processing process of gated adjustment can refer to the following formula: Y = sigmiod(Ax1 + b) ⊙ x2 + x2; Where Y is the device vibration semantic vector, sigmiod is the function corresponding to non-linear activation (other functions can also be used according to requirements), A and b belong to the weight matrix and bias vector of the linear transformation, x1 is the harmonic vibration spectrum vector, x2 is the device vibration spectrum vector, and ⊙ represents bitwise multiplication.

[0050] For the third part, it should be noted that the specific method for semantic enhancement of the device sound time series data in step S130 is not limited and can be selected according to actual requirements.

[0051] For example, in an alternative implementation, semantic mining in the time domain can be performed on the device sound time series data and the surge component in the device sound time series data respectively (e.g., implemented through LSTM (Long Short-Term Memory Network) and GRU (Gated Recurrent Unit)), so as to obtain two corresponding semantic vectors. Then, the two semantic vectors can be concatenated, summed, averaged, or attention processed, etc., to obtain the device sound semantic vector.

[0052] For another example, in another alternative implementation, in order to ensure the effective fusion of surge semantic information during the semantic enhancement process, thereby improving the semantic representation ability of the formed device sound semantic vector, step S130 described above can include the following implementable content: In the first step, the sound time series data of the device can be filtered based on the surge frequency (which can be configured according to the surge frequency generally possessed by the target device. For example, when the target device is a compressor, it is generally between dozens of hertz and hundreds of hertz), to form filtered device sound data. Also, perform time-frequency transformation (such as Fourier transform, etc.) on the filtered device sound data to form a harmonic sound spectrogram, and perform time-frequency transformation (such as Fourier transform, etc.) on the sound time series data of the device to form a device sound spectrogram; In the second step, convolution mining can be respectively performed on the harmonic sound spectrogram and the device sound spectrogram to form a harmonic sound spectrum vector and a device sound spectrum vector; In the third step, based on the harmonic sound spectrum vector, significant feature mining can be performed on the device sound spectrum vector to form a device sound semantic vector, where the significant feature mining includes at least one of gated adjustment and cross-attention processing, and the gating parameter corresponding to the gated adjustment is mapped based on the harmonic sound spectrum vector, as described in the previous relevant description.

[0053] For the fourth part, regarding step S140, it should be noted that the specific method of semantic mining on the device temperature time series data is not limited either and can be selected according to actual needs.

[0054] For example, in an alternative implementation, semantic mining in the time domain can be performed on the device temperature time series data (for example, it can be achieved through LSTM (Long Short-Term Memory Network) and GRU (Gated Recurrent Unit)), so as to obtain the corresponding device temperature semantic vector. Or, self-attention processing can be performed on the result of semantic mining, thereby also achieving semantic enhancement, that is, through internal correlation mining, significant feature mining is realized, and thus a device temperature semantic vector is formed.

[0055] Also, for example, in another alternative implementation, time-frequency conversion can be performed on the device temperature time series data to form a corresponding device temperature spectrogram. Then, convolution mining can be performed on the device temperature spectrogram to form a corresponding device temperature semantic vector. Or, self-attention processing can be performed on the result of convolution mining, thereby also achieving semantic enhancement, that is, through internal correlation mining, significant feature mining is realized, and thus a device temperature semantic vector is formed.

[0056] For the fifth part, regarding step S150, it should be noted that the specific method of fusing the device vibration semantic vector, the device sound semantic vector, and the device temperature semantic vector is not limited either and can be selected accordingly according to actual needs.

[0057] For example, in an alternative embodiment, the device vibration semantic vector, the device sound semantic vector, and the device temperature semantic vector can be concatenated, and then, the concatenated result can be subjected to convolutional mining to form a device operation semantic vector.

[0058] For another example, in another alternative embodiment, to improve the fusion accuracy, step S150 above can further include step S151, step S152, and step S153, and the specific content of each step is described as follows.

[0059] Step S151: Perform a first fusion on the device temperature semantic vector and the device vibration semantic vector to form a first operation semantic vector.

[0060] In the embodiment of the present invention, the device temperature semantic vector and the device vibration semantic vector can be subjected to a first fusion to form a first operation semantic vector. Among them, the first fusion includes multiple fusion stages, so that sufficient and effective fusion can be achieved.

[0061] Step S152: Perform a second fusion on the device temperature semantic vector and the device sound semantic vector to form a second operation semantic vector.

[0062] In the embodiment of the present invention, a second fusion can be performed on the device temperature semantic vector and the device sound semantic vector to form a second operation semantic vector. Among them, the second fusion includes multiple fusion stages, so that sufficient and effective fusion can be achieved. It should be noted that the specific fusion methods of the second fusion and the first fusion can be the same.

[0063] Step S153: Perform a convolutional fusion on the first operation semantic vector and the second operation semantic vector to form a device operation semantic vector.

[0064] In the embodiment of the present invention, after the first operation semantic vector and the second operation semantic vector are formed, the first operation semantic vector and the second operation semantic vector can be subjected to convolutional fusion (that is, first concatenated and then subjected to convolutional mining) to form a device operation semantic vector.

[0065] That is to say, in the present invention, in one aspect, semantic enhancement is first performed internally in the vibration dimension based on harmonic semantic information, and then, based on the semantic information in the temperature dimension, external semantic enhancement is performed on the semantic information in the vibration dimension to obtain a first operating semantic vector. In another aspect, semantic enhancement is first performed internally in the sound dimension based on surge semantic information, and then, based on the semantic information in the temperature dimension, external semantic enhancement is performed on the semantic information in the sound dimension to obtain a second operating semantic vector. In a third aspect, the first operating semantic vector and the second operating semantic vector will be further fused to achieve high-precision and comprehensive capture of global semantic information.

[0066] Optionally, in the above step S151, the specific manner of performing the first fusion of the device temperature semantic vector and the device vibration semantic vector is not limited either. For example, in an alternative embodiment, in order to achieve sufficient fusion of semantic information in two dimensions in multiple stages, the above step S151 may further include step S151a, step S151b, step S151c, step S151d, step S151e, and step S151f. The specific content of each step is as follows (in combination with Figure 5 shown).

[0067] Step S151a: Deeply mine the vibration-temperature fusion vector at the i-th time step to form the vibration-temperature deep vector at the i-th time step.

[0068] In an embodiment of the present invention, the vibration-temperature fusion vector at the i-th time step can be deeply mined to form the vibration-temperature deep vector at the i-th time step. Among them, when the i-th time step is the first time step among multiple time steps, the vibration-temperature fusion vector at the i-th time step is the device vibration semantic vector. That is to say, the device vibration semantic vector can be first deeply mined to form the vibration-temperature deep vector at the first time step. In addition, i is an integer greater than or equal to 1.

[0069] Step S151b: Perform semantic fusion based on the vibration-temperature deep vector at the i-th time step and the device temperature semantic vector to form the fusion semantic vector at the i-th time step.

[0070] In an embodiment of the present invention, after forming the vibration-temperature deep vector at the i-th time step, semantic fusion can be performed based on the vibration-temperature deep vector at the i-th time step and the device temperature semantic vector to form the fusion semantic vector at the i-th time step. In this way, the fusion semantic vector at the i-th time step can synchronously represent the semantic information in the temperature and vibration dimensions, and in this way, the fusion of the semantic information in the temperature dimension can be achieved at each time step.

[0071] Step S151c: Form a mapped semantic vector based on the mapping parameter distribution at the i-th time step and the fused semantic vector at the i-th time step.

[0072] In an embodiment of the present invention, a mapped semantic vector is formed based on the mapping parameter distribution at the i-th time step and the fused semantic vector at the i-th time step. That is to say, through the mapping parameter distribution, the fused semantic vector at the corresponding time step can be further adjusted to make the representational ability of the formed mapped semantic vector better.

[0073] Step S151d: Upsample the mapped semantic vector to form an upsampled semantic vector.

[0074] In an embodiment of the present invention, considering that in the above-mentioned deep mining process, semantic vectors are generally compressed, it is possible to upsample (such as interpolation, transposed convolution, unpooling, etc.) the mapped semantic vector to form an upsampled semantic vector, so that the size of the semantic vector is restored to the original size, that is, the size of the device vibration semantic vector.

[0075] Step S151e: Link the vibration temperature fusion vector at the i-th time step to the upsampled semantic vector to form a vibration temperature fusion vector at the (i + 1)-th time step.

[0076] In an embodiment of the present invention, after forming the upsampled semantic vector, the vibration temperature fusion vector at the i-th time step can be linked to the upsampled semantic vector to form a vibration temperature fusion vector at the (i + 1)-th time step. For example, the vibration temperature fusion vector at the i-th time step and the upsampled semantic vector can be added or averaged to form a vibration temperature fusion vector at the (i + 1)-th time step, such as successively forming the vibration temperature fusion vectors at the second time step, the third time step, and the fourth time step, etc.

[0077] Step S151f: Determine a first running semantic vector based on the vibration temperature fusion vector at the last time step among the multiple time steps.

[0078] In an embodiment of the present invention, after completing the fusion of multiple stages, a first running semantic vector can be determined based on the vibration temperature fusion vector at the last time step among the multiple time steps. For example, the vibration temperature fusion vector at the last time step can be used as the first running semantic vector, or other mapping processes can also be performed.

[0079] Optionally, in the above step S151a, the specific method of deeply mining the vibration temperature fusion vector at the i-th time step is not limited. For example, in an alternative embodiment, in order to realize the mining of deep and high-level potential semantic information through deep mining, the above step S151a may further include step a1, step a2, and step a3, and the specific content of each step is described as follows.

[0080] Step a1, downsample the vibration temperature fusion vector at the i-th time step to form a downsampled fusion vector.

[0081] In the embodiment of the present invention, the vibration temperature fusion vector at the i-th time step can be downsampled to form a downsampled fusion vector. In this way, the data volume can be reduced through downsampling, the computational complexity can be reduced, the processing efficiency can be improved, and subsequent deep mining can be made more efficient.

[0082] Step a2, split the downsampled fusion vector to form multiple local fusion vectors, and serialize the multiple local fusion vectors to form a local fusion vector sequence.

[0083] In the embodiment of the present invention, after forming the downsampled fusion vector, the downsampled fusion vector can be split to form multiple local fusion vectors (exemplarily, sliding window splitting can be performed, and the step size and window size of the sliding window can be configured according to actual needs, or can also be determined during the learning process of corresponding samples and labels), and the multiple local fusion vectors are serialized to form a local fusion vector sequence. In this way, splitting the downsampled vector into multiple local fusion vectors helps to capture local features in the data, improve the sensitivity to local changes, and serializing the local vectors to form an ordered data sequence is convenient for sequence analysis and helps to identify and mine patterns and correlations in the sequence.

[0084] Step a3, perform sequence association mining on the local fusion vector sequence to form the vibration temperature depth vector at the i-th time step.

[0085] In the embodiment of the present invention, after forming the local fusion vector sequence, sequence association mining can be performed on the local fusion vector sequence to form the vibration temperature depth vector at the i-th time step. Among them, the sequence association mining includes channel splicing and channel convolution. In this way, sequence association mining is realized through channel splicing and channel convolution, which can further extract the deep features of the data, enhance the feature expression ability, and form a reliable vibration temperature depth vector. Regarding channel splicing and channel splicing, it should be noted that in combination with Figure 6, the size of each local fusion vector can be n*m, and the local fusion vector is a. Thus, by performing channel concatenation, the size of the concatenated vector formed can be a*n*m. Then, by performing channel convolution, a vibration temperature depth vector can be formed, and the size of this vibration temperature depth vector can be n*m, where the size of the convolution kernel for channel convolution is a*1*1. Additionally, in some other embodiments, after performing channel convolution, the result of channel convolution can be further processed, such as successively performing normalization processing, self-attention processing, residual connection with the result of channel convolution, normalization processing, etc., so as to form the vibration temperature depth vector at the i-th time step.

[0086] Optionally, in step a1 above, the specific manner of downsampling the vibration temperature fusion vector at the i-th time step is not limited. For example, in an alternative embodiment, in order to extract rich semantic information during the compression process, step a1 above can further include the following implementable content: First step, the vibration temperature fusion vector at the i-th time step can be subjected to pooling operations at multiple scales to form pooling fusion vectors at multiple scales. Thus, the capture of local semantic information at different scales can be achieved. Exemplarily, the sizes of the pooling fusion vectors at multiple scales can be 1 / 2, 1 / 4, 1 / 8, etc. of the vibration temperature fusion vector at the i-th time step; Second step, the vibration temperature fusion vector at the i-th time step can be subjected to convolution compression to form a convolution fusion vector. For example, convolution can be performed using a 3*3 convolution kernel with a stride equal to 2 and no padding at the edges; Third step, the pooling fusion vectors at multiple scales and the convolution fusion vector can be subjected to channel concatenation (it should be noted that since the sizes of the vectors may be different, upsampling to the largest size can be performed first and then channel concatenation can be achieved), and channel convolution is performed on the result of channel concatenation (such as the convolution kernel is b*1*1, where b is the cumulative number of the pooling fusion vectors at multiple scales and the convolution fusion vector) to form the downsampled fusion vector of the vibration temperature fusion vector at the i-th time step. Thus, the expression ability of the vector for non-linear semantic information is increased through the operations of channel concatenation and channel convolution.

[0087] Optionally, in step S151b above, the specific manner of semantic fusion based on the vibration temperature depth vector at the i-th time step and the device temperature semantic vector is not limited. In order to achieve reliable fusion of semantic information at different temperatures, step S151b above can further include the following implementable content: In the first step, the device temperature semantic vector can be subjected to semantic space transformation to form a device temperature transformation vector. Also, the device temperature transformation vector and the vibration temperature fusion vector at the i-th time step can be gated and adjusted to form a device temperature adjustment vector, where the gating parameter corresponding to this gating adjustment is mapped based on the vibration temperature fusion vector at the i-th time step (for the specific processing process, reference can be made to the relevant descriptions in the previous text). Additionally, since temperature and vibration actually belong to two different dimensions and the semantic spaces they are in are different, in order to achieve semantic fusion, that is, to ensure the reliability of the gating adjustment, it is necessary to first perform semantic space transformation on the device temperature semantic vector so as to transform it into a semantic space close to the vibration dimension. Specifically, the device temperature semantic vector can be transformed at least once (when performing multiple transformations, the object of the first transformation is the device temperature semantic vector, and the object of subsequent transformations is the output of the previous transformation), thereby forming a device temperature transformation vector, which can be specifically implemented through a corresponding transformation matrix and bias parameter, that is, multiplying the device temperature semantic vector and the transformation matrix in matrix multiplication, and then adding the result of the matrix multiplication and the bias parameter (adding bit by bit), thereby achieving the transformation. In this way, the device temperature semantic vector is transformed into a space more suitable for fusion, which not only improves the fusion efficiency but also enhances the feature expression ability. This transformation involves non-linear transformation, enabling better capture of the potential semantic relationship between device temperature and vibration; In the second step, the device temperature adjustment vector and the vibration temperature depth vector can be summed or averaged to form a preliminary fusion semantic vector; In the third step, the preliminary fusion semantic vector can be subjected to deep convolution and non-linear activation (that is, first perform deep convolution, and then perform non-linear activation on the result of the deep convolution) to form the fusion semantic vector at the i-th time step, where the deep convolution can be implemented through a cascade of multiple convolutional layers, that is, the input of the first convolutional layer is the preliminary fusion semantic vector, the input of the second and subsequent convolutional layers is the output of the previous convolutional layer, and the output of the last convolutional layer after non-linear activation can be used as the fusion semantic vector at the i-th time step.

[0088] Optionally, in step S151c above, the specific manner of depending on the mapping parameter distribution at the i-th time step and the fusion semantic vector at the i-th time step is not limited. For example, in an alternative implementation manner, in order to capture important semantic information through corresponding mapping, step S151c above can further include the following implementable content: In the first step, the mapping parameter distribution at the \(i\)-th time step can be determined. When the \(i\)-th time step is the first time step among the multiple time steps, all parameters in the mapping parameter distribution at the \(i\)-th time step are 1 (in this way, uniform mapping can be achieved, ensuring that all fused semantic vectors are treated equally in the initial stage). When the \(i\)-th time step is other than the first time step among the multiple time steps, the mapping parameter distribution at the \(i\)-th time step is determined based on the mapping parameter distribution at the previous time step. In this way, this dynamic adjustment enables the semantic vectors at the current time step to be gradually optimized according to the previous state information; In the second step, a bitwise multiplication operation can be performed on the mapping parameter distribution at the \(i\)-th time step and the fused semantic vector at the \(i\)-th time step to form a mapped semantic vector.

[0089] Among them, it can be selected that in the above steps, the specific method for determining the mapping parameter distribution at the current time step based on the mapping parameter distribution at the previous time step can be: First, a difference calculation (absolute difference) can be performed on the vibration temperature fusion vector at the previous time step (i.e., the vibration temperature fusion vector at the \((i - 1)\)-th time step, where \(i\) is greater than or equal to 3) and the vibration temperature fusion vector at the two previous time steps (i.e., the vibration temperature fusion vector at the \((i - 2)\)-th time step) to obtain a corresponding difference vector. Then, a non-linear activation (mapped to the interval 0 - 1) can be performed on this difference vector to obtain an activated difference vector. Then, the activated difference vector can be downsampled to obtain a corresponding downsampled difference vector (the size can be the same as the size of the mapping parameter distribution). Then, a bitwise multiplication operation can be performed on the downsampled difference vector and the mapping parameter distribution at the previous time step to form the mapping parameter distribution at the current time step.

[0090] For the sixth part, it should be noted that for step S160, the specific method for predicting and outputting the operation fault data of the target device based on the device operation semantic vector is not limited and can be selected according to actual needs.

[0091] For example, in an alternative implementation, a fully connected process can be performed on the device operation semantic vector to form a corresponding fully connected vector. The size of this fully connected vector can be 1 * 1. Then, an identity mapping or a linear mapping can be performed on this fully connected vector to obtain a probability value between 0 and 1, which is used as the operation fault data.

[0092] In summary, the operation fault monitoring method and device for the HVAC system based on data analysis provided by the present invention are as follows. First, based on the harmonic semantic information in the device vibration time series data, semantic enhancement is performed on the device vibration time series data to form a device vibration semantic vector. Then, based on the surge semantic information in the device sound time series data, semantic enhancement is performed on the device sound time series data to form a device sound semantic vector. After that, semantic mining is performed on the device temperature time series data to form a device temperature semantic vector. Further, the device vibration semantic vector, the device sound semantic vector, and the device temperature semantic vector are fused to form a device operation semantic vector. Finally, operation fault data is predicted and output based on the device operation semantic vector. Based on the above method, in the first aspect, since the device time series data in the three dimensions of vibration, sound, and temperature are semantically fused, the formed device operation semantic vector can comprehensively represent the operation process of the target device. In the second aspect, since the semantic information in the vibration dimension is enhanced based on the harmonic semantic information, and the harmonic has a good representation effect on the unbalanced situation inside the device (such as a compressor), therefore, when the device vibration semantic vector represents the global semantic information of the vibration, it can also importantly represent some internal anomalies. In the third aspect, since the semantic information in the sound dimension is semantically enhanced based on the surge semantic information, and the surge is a phenomenon of periodic fluctuations in pressure and flow rate caused by dynamic blockage of the gas flow path during the operation of the device (compressor), therefore, the surge plays an important role in the semantic representation of some abnormal situations. Therefore, when the device sound semantic vector represents the global semantic information of the sound, it can also importantly represent some anomalies of the device. Based on this, it is possible to comprehensively mine the potential semantic information and focus on the important semantic information during the operation process of the target device, thereby improving the reliability of fault prediction and further improving the problem of relatively low reliability of operation fault monitoring existing in the prior art.

[0093] In several embodiments provided by the embodiments of the present invention, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device and method embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architectures, functions, and operations of devices, methods, and computer program products according to multiple embodiments of the present invention. In this regard, each block in the flowchart or block diagram may represent a module, a program segment, or a part of code, and the part of the module, program segment, or code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than that marked in the accompanying drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, as well as the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0094] In addition, each functional module in various embodiments of the present invention may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0095] If the described functions are implemented in the form of software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, an electronic device, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes. It should be noted that in this article, the term "including", "comprising", or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article, or device including a series of elements not only includes those elements, but also includes other elements that are not explicitly listed, or also includes elements inherent to such a process, method, article, or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of another identical element in the process, method, article, or device including the said element.

[0096] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for monitoring operation faults of an HVAC system based on data analysis, characterized in that, Including: Obtain the device vibration time series data, device sound time series data, and device temperature time series data of the target device in the target HVAC system; Based on the harmonic semantic information in the device vibration time series data, perform semantic enhancement on the device vibration time series data to form a device vibration semantic vector; Based on the surge semantic information in the device sound time series data, perform semantic enhancement on the device sound time series data to form a device sound semantic vector; Perform semantic mining on the device temperature time series data to form a device temperature semantic vector; Fuse the device vibration semantic vector, the device sound semantic vector, and the device temperature semantic vector to form a device operation semantic vector; Predict and output the operation fault data of the target device based on the device operation semantic vector, where the operation fault data is used to characterize the probability of the target device having a fault.

2. The method for monitoring the operation failure of an HVAC system based on data analysis according to claim 1, characterized in that, The step of fusing the device vibration semantic vector, the device sound semantic vector, and the device temperature semantic vector to form a device operation semantic vector includes: Perform a first fusion on the device temperature semantic vector and the device vibration semantic vector to form a first operation semantic vector, where the first fusion includes multiple fusion stages; Perform a second fusion on the device temperature semantic vector and the device sound semantic vector to form a second operation semantic vector, where the second fusion includes multiple fusion stages; Perform a convolutional fusion on the first operation semantic vector and the second operation semantic vector to form a device operation semantic vector.

3. The method for monitoring the operation faults of the HVAC system based on data analysis according to claim 2, characterized in that, The step of performing a first fusion on the device temperature semantic vector and the device vibration semantic vector to form a first operation semantic vector includes: Deeply mine the vibration-temperature fusion vector at the i-th time step to form the vibration-temperature deep vector at the i-th time step, where when the i-th time step is the first time step among multiple time steps, the vibration-temperature fusion vector at the i-th time step is the device vibration semantic vector; Perform semantic fusion based on the vibration-temperature deep vector at the i-th time step and the device temperature semantic vector to form the fusion semantic vector at the i-th time step; Form a mapped semantic vector based on the mapping parameter distribution at the i-th time step and the fusion semantic vector at the i-th time step; Perform upsampling on the mapped semantic vector to form an upsampled semantic vector; Link the vibration-temperature fusion vector at the i-th time step to the upsampled semantic vector to form the vibration-temperature fusion vector at the (i + 1)-th time step; Determine the first operation semantic vector based on the vibration-temperature fusion vector at the last time step among the multiple time steps.

4. The method for monitoring the operation faults of the HVAC system based on data analysis according to claim 3, wherein, The step of deeply mining the vibration-temperature fusion vector at the i-th time step to form the vibration-temperature deep vector at the i-th time step includes: Perform downsampling on the vibration-temperature fusion vector at the i-th time step to form a downsampled fusion vector; Segment the downsampled fusion vector to form multiple local fusion vectors, and serialize the multiple local fusion vectors to form a local fusion vector sequence; Perform sequence correlation mining on the local fusion vector sequence to form the vibration temperature depth vector at the \(i\)-th time step, where the sequence correlation mining includes channel splicing and channel convolution.

5. The method for monitoring the operation faults of the HVAC system based on data analysis according to claim 4, wherein, The step of downsampling the vibration temperature fusion vector at the \(i\)-th time step to form a downsampled fusion vector includes: Performing pooling operations at multiple scales on the vibration temperature fusion vector at the \(i\)-th time step to form pooling fusion vectors at multiple scales; Performing convolution compression on the vibration temperature fusion vector at the \(i\)-th time step to form a convolution fusion vector; Performing channel splicing on the pooling fusion vectors at multiple scales and the convolution fusion vector, and performing channel convolution on the result of the channel splicing to form the downsampled fusion vector of the vibration temperature fusion vector at the \(i\)-th time step.

6. The method for monitoring the operation failure of an HVAC system based on data analysis according to claim 3, wherein The step of performing semantic fusion based on the vibration temperature depth vector at the \(i\)-th time step and the device temperature semantic vector to form the fusion semantic vector at the \(i\)-th time step includes: Performing semantic space transformation on the device temperature semantic vector to form a device temperature transformation vector, and performing gated adjustment on the device temperature transformation vector and the vibration temperature fusion vector at the \(i\)-th time step to form a device temperature adjustment vector, where the gated parameter corresponding to the gated adjustment is mapped based on the vibration temperature fusion vector at the \(i\)-th time step; Performing summation or mean calculation on the device temperature adjustment vector and the vibration temperature depth vector to form a preliminary fusion semantic vector; Performing depth convolution and non-linear activation on the preliminary fusion semantic vector to form the fusion semantic vector at the \(i\)-th time step.

7. The method for monitoring the operation faults of the HVAC system based on data analysis according to claim 3, wherein, The step of forming a mapped semantic vector based on the mapping parameter distribution at the \(i\)-th time step and the fusion semantic vector at the \(i\)-th time step includes: Determining the mapping parameter distribution at the \(i\)-th time step, where when the \(i\)-th time step is the first time step among the multiple time steps, all parameters in the mapping parameter distribution at the \(i\)-th time step are 1, and when the \(i\)-th time step is other than the first time step among the multiple time steps, the mapping parameter distribution at the \(i\)-th time step is determined based on the mapping parameter distribution of the previous time step; Performing a bitwise multiplication operation on the mapping parameter distribution at the \(i\)-th time step and the fusion semantic vector at the \(i\)-th time step to form a mapped semantic vector.

8. The operation fault monitoring method of the HVAC system based on data analysis according to any one of claims 1-7, characterized in that, The step of performing semantic enhancement on the device vibration time series data based on the harmonic semantic information in the device vibration time series data to form a device vibration semantic vector includes: Performing time-frequency transformation on the device vibration time series data to form a device vibration spectrogram, and determining the harmonic frequency corresponding to the device vibration time series data based on the device vibration spectrogram; Filtering the device vibration time series data based on the harmonic frequency to form device vibration harmonic data, and performing time-frequency transformation on the device vibration harmonic data to form a harmonic vibration spectrogram; Perform convolution mining on the device vibration spectrogram and the harmonic vibration spectrogram respectively to form a device vibration spectrum vector and a harmonic vibration spectrum vector; Based on the harmonic vibration spectrum vector, perform significant feature mining on the device vibration spectrum vector to form a device vibration semantic vector, where the significant feature mining includes at least one of gated adjustment and cross-attention processing, and the gated parameter corresponding to the gated adjustment is mapped based on the harmonic vibration spectrum vector.

9. The operation fault monitoring method of the HVAC system based on data analysis according to any one of claims 1-7, characterized in that, The step of performing semantic enhancement on the device sound time series data based on the surge semantic information in the device sound time series data to form a device sound semantic vector includes: Filter the device sound time series data based on the surge frequency to form device sound filtered data, perform time-frequency transformation on the device sound filtered data to form a harmonic sound spectrogram, and perform time-frequency transformation on the device sound time series data to form a device sound spectrogram; Perform convolution mining on the harmonic sound spectrogram and the device sound spectrogram respectively to form a harmonic sound spectrum vector and a device sound spectrum vector; Based on the harmonic sound spectrum vector, perform significant feature mining on the device sound spectrum vector to form a device sound semantic vector, where the significant feature mining includes at least one of gated adjustment and cross-attention processing, and the gated parameter corresponding to the gated adjustment is mapped based on the harmonic sound spectrum vector.

10. An electronic device, characterized in that, Comprising a processor and a memory, the memory is used for storing a computer program, and the processor is used for executing the computer program to implement the operation fault monitoring method of the HVAC system based on data analysis according to any one of claims 1-9.

Images