Time series data fault diagnosis method and device, electronic equipment and storage medium

Abstract

The present disclosure provides a time series data fault diagnosis method and device, electronic equipment and storage medium, belonging to the technical field of fault diagnosis. The method comprises: acquiring time series data; determining an approximate optimal receptive field corresponding to the time series data according to frequency domain data of the time series data; adjusting a pre-constructed adaptive Inception network according to the approximate optimal receptive field to obtain an adjusted adaptive Inception network; inputting the time series data into the adjusted adaptive Inception network to obtain feature data of the time series data for prediction, classification or fault diagnosis. According to the present disclosure, the approximate optimal receptive field corresponding to the time series data is determined according to the frequency domain data of the time series data, without the need for a large number of experiments to select a suitable convolution kernel, thereby reducing the time cost.

Description

Technical Field

[0001] This disclosure relates to the field of fault diagnosis technology, and in particular to a method, apparatus, electronic device and storage medium for diagnosing time-series data faults. Background Technology

[0002] The data generated by electronic products is mostly time-series data. For fault diagnosis of electronic products, feature extraction of time series data is crucial, as the quality of the extracted sequence features has a significant impact on subsequent prediction, classification, and anomaly detection results. To extract better time-series features, the powerful feature extraction capabilities of convolutional neural networks are often utilized. However, selecting an appropriate convolutional kernel size often requires extensive experimentation, incurring significant time costs.

[0003] Public content

[0004] This disclosure provides a method, apparatus, electronic device, and storage medium for diagnosing time-series data faults, in order to solve the problem of spending a lot of time and resources on conducting numerous experiments to select a suitable convolution kernel size.

[0005] This disclosure provides a method for diagnosing time-series data faults, including:

[0006] Acquire time-series data, and determine the approximate optimal receptive field corresponding to the time-series data based on the frequency domain data of the time-series data;

[0007] The pre-constructed adaptive Inception network is adjusted based on the approximate optimal receptive field to obtain the adjusted adaptive Inception network.

[0008] The time-series data is input into the adjusted adaptive Inception network to obtain the feature data of the time-series data for prediction, classification or fault diagnosis.

[0009] As an example, the adaptive Inception network includes a first convolutional processing module, a pooling module, and a second convolutional processing module. The output features of the first convolutional processing module and the output features of the pooling module are concatenated and then output to the second convolutional processing module. The first convolutional processing module includes a one-dimensional convolutional layer and a first multi-channel convolutional network arranged sequentially. The second convolutional processing module includes a second multi-channel convolutional network.

[0010] As an example, the time-series data includes single-channel time-series data and multi-channel time-series data. The kernel size of the one-dimensional convolutional layer is determined based on the approximate optimal receptive field corresponding to the single-channel time-series data or the mode of the approximate optimal receptive field corresponding to the multi-channel time-series data. The kernel size of the first multi-channel convolutional network is determined based on the approximate optimal receptive field corresponding to the single-channel time-series data or the approximate optimal receptive field of each channel of the multi-channel time-series data.

[0011] As an example, if the time-series data is single-channel time-series data, the kernel sizes of the first multi-channel convolutional network are 1×1, r×1, 1.5r×1, and 2r×1, respectively, where r is the approximate optimal receptive field corresponding to the single-channel time-series data; if the time-series data is multi-channel time-series data, the kernel sizes of the first multi-channel convolutional network are 1×1, r ... i ×1, 1.5r i ×1 and 2r i ×1, r i This represents the approximate optimal receptive field for the i-th channel.

[0012] As one embodiment, the second multichannel convolutional network includes small-sized convolutional kernels.

[0013] As an example, the adaptive Inception network further includes a residual module, which connects the input of the adaptive Inception network to the output of the second multi-channel convolutional network.

[0014] As one embodiment, determining the approximate optimal receptive field corresponding to the time-series data based on the frequency domain data of the time-series data includes:

[0015] Based on the frequency domain data of the time series data, determine the frequency corresponding to the maximum frequency domain value;

[0016] Based on the period corresponding to the frequency, determine the approximate optimal receptive field corresponding to the time series data.

[0017] This disclosure also provides a timing data fault diagnosis device, including:

[0018] The receptive field determination module is used to acquire time-series data and determine the approximate optimal receptive field corresponding to the time-series data based on the frequency domain data of the time-series data.

[0019] The network adjustment module is used to adjust the pre-constructed adaptive Inception network according to the approximate optimal receptive field to obtain the adjusted adaptive Inception network.

[0020] The feature extraction module is used to input the time series data into the adjusted adaptive Inception network to obtain the feature data of the time series data for prediction, classification or fault diagnosis.

[0021] This disclosure also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the timing data fault diagnosis method.

[0022] This disclosure also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the timing data fault diagnosis method.

[0023] This disclosure provides a method, apparatus, electronic device, and storage medium for fault diagnosis of time-series data. The method involves acquiring time-series data, determining an approximate optimal receptive field (ARF) based on the frequency domain data of the time-series data, adjusting a pre-constructed adaptive Inception network based on the ARF to obtain an adjusted adaptive Inception network, and inputting the time-series data into the adjusted adaptive Inception network to obtain feature data of the time-series data for prediction, classification, or fault diagnosis. This disclosure determines the ARF based on the frequency domain data of the time-series data, eliminating the need for extensive experiments to select suitable convolutional kernels and reducing time costs. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in this disclosure or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart illustrating the timing data fault diagnosis method provided in this embodiment of the disclosure;

[0026] Figure 2 This is a schematic diagram of the spectrum of multi-channel timing data provided in the embodiments of this disclosure;

[0027] Figure 3 This is a schematic diagram of the structure of the multi-channel adaptive Inception network provided in the embodiments of this disclosure;

[0028] Figure 4 This is a schematic diagram of the structure of the timing data fault diagnosis device provided in the embodiments of this disclosure;

[0029] Figure 5 This is a schematic diagram of the structure of the electronic device provided in this disclosure. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of this disclosure clearer, the technical solutions of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0031] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the embodiments of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0032] Figure 1 This is a flowchart illustrating the timing data fault diagnosis method provided in this embodiment of the disclosure, as shown below. Figure 1 As shown, this disclosure provides a method for diagnosing time-series data faults, including but not limited to the following steps:

[0033] Step S100: Acquire time-series data and determine the approximate optimal receptive field corresponding to the time-series data based on the frequency domain data of the time-series data. The time-series data includes single-channel time-series data and multi-channel time-series data. The time-series data can be acquired from electronic products that need to perform time series prediction, classification, and anomaly detection tasks. It can be collected by sensors or obtained after establishing a communication connection with the electronic product through a communication circuit. Before determining the approximate optimal receptive field corresponding to the time-series data based on the frequency domain data, the acquired time-series data can be preprocessed, such as data cleaning and normalization, to improve accuracy.

[0034] Step S200: Adjust the pre-constructed adaptive Inception network according to the approximate optimal receptive field to obtain the adjusted adaptive Inception network;

[0035] The adaptive Inception network consists of a multi-channel convolutional module and an Inception module. The parameters of the multi-channel convolutional module are adjusted according to an approximate optimal receptive field, allowing for matching with temporal data. The Inception module is a convolutional neural network (CNN) module, originally used in the ImageNet image classification challenge. As a core component of deep neural networks, the Inception module aims to address challenges such as redundant model parameters and heavy computational burden, improving feature extraction performance at different scales while reducing the number of parameters and computational complexity.

[0036] If the time series data is multi-channel, then the adaptive Inception network is called the Multichannel Adaptive Inception network, abbreviated as MCA-Inception. The MCA-Inception network can be used for feature extraction in time series data, offering both powerful feature extraction capabilities and the ability to adaptively match appropriate convolutional kernel sizes.

[0037] Step S300: Input the time series data into the adjusted adaptive Inception network to obtain the feature data of the time series data for prediction, classification or fault diagnosis.

[0038] It is understood that this disclosure determines the approximate optimal receptive field corresponding to the time series data based on the frequency domain data of the time series data, eliminating the need to conduct a large number of experiments to select a suitable convolution kernel, thus reducing time costs.

[0039] Based on the above embodiments, as an optional embodiment, determining the approximate optimal receptive field corresponding to the time-series data based on the frequency domain data of the time-series data includes:

[0040] Step S110: Determine the frequency corresponding to the maximum frequency domain value based on the frequency domain data of the time series data;

[0041] Step S120: Determine the approximate optimal receptive field corresponding to the time series data based on the period corresponding to the frequency.

[0042] In step S110, the frequency domain data of the time series data can be obtained by performing Fourier transform, fast Fourier transform, or wavelet transform on the time series data. This embodiment of the disclosure uses fast Fourier transform as an example for illustration.

[0043] Specifically, based on the Fast Fourier Transform (FFT), the amplitude of each channel of single-channel or multi-channel time-series data at different frequencies can be obtained. The amplitude represents the intensity of each frequency component in the signal spectrum. A frequency component with a larger amplitude indicates that a large amount of energy in the signal is concentrated at that frequency.

[0044] In step S120, the period with the largest contribution of each channel is calculated based on the frequency with the largest amplitude of each channel, and an appropriate convolution kernel size is adaptively selected to obtain the optimal receptive field, which greatly reduces the computational overhead required to try various sizes of convolution kernels.

[0045] Figure 2 This is a schematic diagram of the spectrum of multi-channel time-series data provided in the embodiments of this disclosure, such as... Figure 2 As shown, in this embodiment, the spectrum of the time series data is first obtained based on the Fast Fourier Transform (FFT), and the approximate optimal receptive field (RF) is determined based on the frequency corresponding to the maximum amplitude. For multi-channel time series, the frequencies that contribute the most to each channel and the corresponding optimal receptive fields can be obtained.

[0046] For periodic time-series data, the optimal receptive field should cover a complete period so that the network can capture key features within the periodic patterns. If the receptive field is too small, it may not cover the entire period, preventing the network from effectively learning periodic features. Conversely, if the receptive field is too large, it may contain information from multiple periods, resulting in less accurate learned features. Therefore, a receptive field that is 1-2 times the period corresponding to the maximum contributing frequency can be chosen as an approximate optimal receptive field.

[0047] The formula for calculating the receptive field is as follows:

[0048]

[0049] l k Let l represent the receptive field of the k-th layer. k-1 f represents the receptive field of the (k-1)th layer. k s represents the size of the k-th convolutional kernel. i This represents the step size of the i-th layer.

[0050] Understandably, this disclosure approximates the optimal receptive field by finding the frequency with the greatest contribution in each channel of the time-series data, thus greatly shortening the computation process of searching for the optimal convolutional kernel.

[0051] Based on the above embodiments, as an optional embodiment, the adaptive Inception network includes a first convolutional processing module, a pooling module, and a second convolutional processing module. The output features of the first convolutional processing module and the output features of the pooling module are concatenated and then output to the second convolutional processing module. The first convolutional processing module includes a one-dimensional convolutional layer and a first multi-channel convolutional network arranged sequentially. The second convolutional processing module includes a second multi-channel convolutional network. Optionally, the second multi-channel convolutional network includes small-sized convolutional kernels. Optionally, the pooling module may include a 3×1 max-pooling layer and a 1×1 one-dimensional convolutional layer.

[0052] Optionally, the kernel size of the one-dimensional convolutional layer is determined based on the approximate optimal receptive field corresponding to the single-channel temporal data or the mode of the approximate optimal receptive field corresponding to the multi-channel temporal data, and the kernel size of the first multi-channel convolutional network is determined based on the approximate optimal receptive field corresponding to the single-channel temporal data or the approximate optimal receptive field of each channel of the multi-channel temporal data.

[0053] Optionally, if the time-series data is single-channel time-series data, the kernel sizes of the first multi-channel convolutional network are 1×1, r×1, 1.5r×1, and 2r×1, respectively, where r is the approximate optimal receptive field corresponding to the single-channel time-series data; if the time-series data is multi-channel time-series data, the kernel sizes of the first multi-channel convolutional network are 1×1, r ... i ×1, 1.5r i ×1 and 2r i ×1, r i This represents the approximate optimal receptive field for the i-th channel.

[0054] As an example, the adaptive Inception network further includes a residual module, which connects the input of the adaptive Inception network to the output of the second multi-channel convolutional network.

[0055] For single-channel time-series data, assuming the approximate optimal receptive field obtained by FFT is r, a one-dimensional convolutional layer with a kernel size of r×1 is introduced in the initial stage to match the optimal receptive field and initially extract more refined features. Subsequently, these features are fed into a multi-channel one-dimensional convolution process (i.e., the first multi-channel convolutional network), where the convolution kernels of the one-dimensional convolution process adopt sizes of 1×1, r×1, 1.5r×1, and 2r×1, respectively.

[0056] For multi-channel time-series data, assume that the approximate optimal receptive field of each channel obtained by FFT is r. i r i The mode is r mThis refers to the approximate optimal receptive field where the frequency of occurrence in each channel is highest. In the initial stage, a convolutional kernel with size r is introduced. m A 1×1 one-dimensional convolutional layer is then used in the multi-channel one-dimensional convolution process (i.e., the first multi-channel convolutional network), followed by a 1×1 r layer. i ×1, 1.5r i ×1, 2r i ×1 one-dimensional convolution. The convolution result output from the multi-channel one-dimensional convolution process is concatenated with the input features after max pooling to form a new feature map. Subsequently, the feature map is subjected to multi-channel parallel one-dimensional convolution processing (i.e., a second multi-channel convolutional network) for secondary feature extraction. The convolution kernel size of the multi-channel parallel one-dimensional convolution is 1×1, 3×1, 5×1, 7×1. In this embodiment, a smaller convolution kernel is selected to better identify local patterns and features.

[0057] In a preferred embodiment, the adaptive Inception network of this disclosure also introduces a residual module at the input end, with the residual module branch passing through a 1×1 one-dimensional convolutional layer. After each concatenation operation, the first and second convolutional processing modules are backed by BN and ReLU layers. By setting BN and ReLU layers, the training process is accelerated, and the adaptive Inception network becomes more robust during training.

[0058] The first multi-channel convolutional network selects convolutional kernels of different sizes for each channel to obtain a larger receptive field. The outputs of the multi-scale convolutions are concatenated along the channel dimension, and the concatenated output can be represented as:

[0059]

[0060] in, The output representing the i-th channel can be expressed as:

[0061]

[0062] Where f0 represents r m ×1 convolution, f1 i Indicates the convolutional layer corresponding to the i-th channel; x pooling A pooled channel can be represented as:

[0063] x pooling =f2(Maxpooling(x) i ));

[0064] Here, f2 is a 1×1 convolution operation.

[0065] Understandably, this disclosure not only adaptively selects appropriate convolutional kernels to obtain an approximate optimal receptive field, significantly reducing the computational overhead in selecting suitable convolutional kernels, but also significantly improves the training efficiency of the model through the application of a multi-channel parallel convolution strategy. While maintaining the original advantages of Inception, it further enhances the performance and practicality in time series tasks. This disclosure meticulously optimizes traditional designs for time series tasks, adaptively selecting the convolutional kernel size and using Fast Fourier Transform to find the frequency with the greatest contribution in each channel of the time series to approximately determine the optimal receptive field. This greatly reduces the computational process of searching for the optimal convolutional kernel. Simultaneously, residual connections are employed to improve accuracy and network stability, solving problems such as gradient vanishing and training difficulties encountered during deep neural network training. It also improves the network's expressive and generalization abilities, exhibiting stronger feature extraction capabilities for time series data.

[0066] The timing data fault diagnosis device provided in this disclosure is described below. The timing data fault diagnosis device described below and the timing data fault diagnosis method described above can be referred to in correspondence.

[0067] Figure 4 This is a schematic diagram of the structure of the timing data fault diagnosis device provided in the embodiments of this disclosure, as shown below. Figure 4 As shown, this disclosure also provides a timing data fault diagnosis device, including:

[0068] The receptive field determination module 410 is used to acquire time-series data and determine the approximate optimal receptive field corresponding to the time-series data based on the frequency domain data of the time-series data.

[0069] The network adjustment module 420 is used to adjust the pre-constructed adaptive Inception network according to the near-optimal receptive field to obtain the adjusted adaptive Inception network.

[0070] The feature extraction module 430 is used to input the time series data into the adjusted adaptive Inception network to obtain the feature data of the time series data for prediction, classification or fault diagnosis.

[0071] As an example, the adaptive Inception network includes a first convolutional processing module, a pooling module, and a second convolutional processing module. The output features of the first convolutional processing module and the output features of the pooling module are concatenated and then output to the second convolutional processing module. The first convolutional processing module includes a one-dimensional convolutional layer and a first multi-channel convolutional network arranged sequentially. The second convolutional processing module includes a second multi-channel convolutional network.

[0072] As an example, the time-series data includes single-channel time-series data and multi-channel time-series data. The kernel size of the one-dimensional convolutional layer is determined based on the approximate optimal receptive field corresponding to the single-channel time-series data or the mode of the approximate optimal receptive field corresponding to the multi-channel time-series data. The kernel size of the first multi-channel convolutional network is determined based on the approximate optimal receptive field corresponding to the single-channel time-series data or the approximate optimal receptive field of each channel of the multi-channel time-series data.

[0073] As an example, if the time-series data is single-channel time-series data, the kernel sizes of the first multi-channel convolutional network are 1×1, r×1, 1.5r×1, and 2r×1, respectively, where r is the approximate optimal receptive field corresponding to the single-channel time-series data; if the time-series data is multi-channel time-series data, the kernel sizes of the first multi-channel convolutional network are 1×1, r ... i ×1, 1.5r i ×1 and 2r i ×1, r i This represents the approximate optimal receptive field for the i-th channel.

[0074] As one embodiment, the second multichannel convolutional network includes small-sized convolutional kernels.

[0075] As an example, the adaptive Inception network further includes a residual module, which connects the input of the adaptive Inception network to the output of the second multi-channel convolutional network.

[0076] As one embodiment, the receptive field determination module 410 is further configured to:

[0077] Based on the frequency domain data of the time series data, determine the frequency corresponding to the maximum frequency domain value;

[0078] Based on the period corresponding to the frequency, determine the approximate optimal receptive field corresponding to the time series data.

[0079] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5 As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other via the communication bus 540. The processor 510 can call logical instructions in the memory 530 to execute a timing data fault diagnosis method, which includes:

[0080] Acquire time-series data, and determine the approximate optimal receptive field corresponding to the time-series data based on the frequency domain data of the time-series data;

[0081] The pre-constructed adaptive Inception network is adjusted based on the approximate optimal receptive field to obtain the adjusted adaptive Inception network.

[0082] The time-series data is input into the adjusted adaptive Inception network to obtain the feature data of the time-series data for prediction, classification or fault diagnosis.

[0083] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this disclosure, in essence, or the part that contributes to the prior art, or a part of the 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 to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this disclosure. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0084] On the other hand, this disclosure also provides a computer program product, the computer program product including a computer program that can be stored on a non-transitory computer-readable storage medium, and when the computer program is executed by a processor, the computer is able to execute a timing data fault diagnosis method, the method including:

[0085] Acquire time-series data, and determine the approximate optimal receptive field corresponding to the time-series data based on the frequency domain data of the time-series data;

[0086] The pre-constructed adaptive Inception network is adjusted based on the approximate optimal receptive field to obtain the adjusted adaptive Inception network.

[0087] The time-series data is input into the adjusted adaptive Inception network to obtain the feature data of the time-series data for prediction, classification or fault diagnosis.

[0088] In another aspect, this disclosure also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform a timing data fault diagnosis method, the method comprising:

[0089] Acquire time-series data, and determine the approximate optimal receptive field corresponding to the time-series data based on the frequency domain data of the time-series data;

[0090] The pre-constructed adaptive Inception network is adjusted based on the approximate optimal receptive field to obtain the adjusted adaptive Inception network.

[0091] The time-series data is input into the adjusted adaptive Inception network to obtain the feature data of the time-series data for prediction, classification or fault diagnosis.

[0092] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0093] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0094] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this disclosure, and are not intended to limit them. Although this disclosure 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 of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this disclosure.

Claims

1. A method for diagnosing time-series data faults, characterized in that, include: Acquire time-series data, and determine the approximate optimal receptive field corresponding to the time-series data based on the frequency domain data of the time-series data; The pre-constructed adaptive Inception network is adjusted based on the approximate optimal receptive field to obtain the adjusted adaptive Inception network. The time series data is input into the adjusted adaptive Inception network to obtain the feature data of the time series data for prediction, classification or fault diagnosis. The adaptive Inception network includes a first convolutional processing module, a pooling module, and a second convolutional processing module. The output features of the first convolutional processing module and the output features of the pooling module are concatenated and then output to the second convolutional processing module. The first convolutional processing module includes a one-dimensional convolutional layer and a first multi-channel convolutional network arranged in sequence. The second convolutional processing module includes a second multi-channel convolutional network. The time-series data includes single-channel time-series data and multi-channel time-series data. The kernel size of the one-dimensional convolutional layer is determined based on the approximate optimal receptive field corresponding to the single-channel time-series data or the mode of the approximate optimal receptive field corresponding to the multi-channel time-series data. The kernel size of the first multi-channel convolutional network is determined based on the approximate optimal receptive field corresponding to the single-channel time-series data or the approximate optimal receptive field of each channel of the multi-channel time-series data.

2. The time-series data fault diagnosis method according to claim 1, characterized in that, If the time-series data is the single-channel time-series data, the kernel size of the first multi-channel convolutional network is 1 × 1, r × 1, 1.5 r × 1 and 2 r × 1, r The approximate optimal receptive field corresponds to the single-channel temporal data; if the temporal data is multi-channel temporal data, the kernel size of the first multi-channel convolutional network is 1 × 1, × 1, 1.5 × 1 and 2 × 1, For the first i The approximate optimal receptive field of each channel.

3. The time-series data fault diagnosis method according to claim 1, characterized in that, The second multichannel convolutional network includes small-sized convolutional kernels.

4. The time-series data fault diagnosis method according to claim 1, characterized in that, The adaptive Inception network further includes a residual module, which connects the input of the adaptive Inception network to the output of the second multi-channel convolutional network.

5. The time-series data fault diagnosis method according to claim 1, characterized in that, Determining the approximate optimal receptive field corresponding to the time-series data based on the frequency domain data of the time-series data includes: Based on the frequency domain data of the time series data, determine the frequency corresponding to the maximum frequency domain value; Based on the period corresponding to the frequency, determine the approximate optimal receptive field corresponding to the time series data.

6. A timing data fault diagnosis device, characterized in that, include: The receptive field determination module is used to acquire time-series data and determine the approximate optimal receptive field corresponding to the time-series data based on the frequency domain data of the time-series data. The network adjustment module is used to adjust the pre-constructed adaptive Inception network according to the approximate optimal receptive field to obtain the adjusted adaptive Inception network. The feature extraction module is used to input the time series data into the adjusted adaptive Inception network to obtain the feature data of the time series data for prediction, classification or fault diagnosis; The adaptive Inception network includes a first convolutional processing module, a pooling module, and a second convolutional processing module. The output features of the first convolutional processing module and the output features of the pooling module are concatenated and then output to the second convolutional processing module. The first convolutional processing module includes a one-dimensional convolutional layer and a first multi-channel convolutional network arranged in sequence. The second convolutional processing module includes a second multi-channel convolutional network. The time-series data includes single-channel time-series data and multi-channel time-series data. The kernel size of the one-dimensional convolutional layer is determined based on the approximate optimal receptive field corresponding to the single-channel time-series data or the mode of the approximate optimal receptive field corresponding to the multi-channel time-series data. The kernel size of the first multi-channel convolutional network is determined based on the approximate optimal receptive field corresponding to the single-channel time-series data or the approximate optimal receptive field of each channel of the multi-channel time-series data.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the timing data fault diagnosis method as described in any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the timing data fault diagnosis method as described in any one of claims 1 to 5.

Images