Data identification method, ONU identification method, electronic equipment and program product

Abstract

The embodiment of the invention provides a data identification method, an ONU identification method, electronic equipment and a program product. The data identification method comprises the following steps: acquiring one-dimensional time sequence data; performing two-dimensional feature mapping on the one-dimensional time sequence data to obtain multi-channel two-dimensional data; the multi-channel two-dimensional data is identified according to a preset category set through the image classification model, a target category corresponding to the multi-channel two-dimensional data is obtained, and the preset category set comprises a normal category and at least one abnormal category. Through the scheme of the embodiment, the abnormal features of the abnormal data can be captured more accurately, and the recognition precision and efficiency are improved.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the field of communication network management and security technologies, and in particular, to a data recognition method, an ONU recognition method, an electronic device, and a program product. Background Art

[0002] One-dimensional time series data plays an important role in modern science and engineering fields and is widely used in multiple fields such as fault diagnosis and communication signal processing. Such data is usually recorded at equally spaced time points, reflecting the dynamic changes of the system over time. However, due to its simple structure, one-dimensional time series data has some inherent problems. First, it is difficult to directly capture complex time series relationships and periodic features with this data form, especially in low-quality data scenarios where information is lost due to a limiting amplifier or sampling frequency limitations. Second, it is difficult to intuitively display trends and potential patterns with one-dimensional time series data, and the feature extraction process is relatively complex, posing challenges to model training and analysis. Summary of the Invention

[0003] Embodiments of the present disclosure provide a data recognition method, an ONU recognition method, an electronic device, and a program product.

[0004] In a first aspect, embodiments of the present disclosure provide a data recognition method, which includes:

[0005] Collect one-dimensional time series data;

[0006] Perform two-dimensional feature mapping on the one-dimensional time series data to obtain multi-channel two-dimensional data;

[0007] Use an image classification model to recognize the multi-channel two-dimensional data according to a preset category set, and obtain the target category corresponding to the multi-channel two-dimensional data, where the preset category set includes a normal category and at least one abnormal category.

[0008] In a second aspect, embodiments of the present disclosure further provide an ONU recognition method, which includes:

[0009] Based on the data recognition method, obtain the target category corresponding to the one-dimensional time series data of the optical network unit ONU;

[0010] Identify the category of the ONU according to the target category.

[0011] In a third aspect, embodiments of the present disclosure further provide an electronic device, including:

[0012] One or more processors;

[0013] A memory on which one or more programs are stored. When the one or more programs are executed by the one or more processors, the one or more processors implement the data recognition method or the ONU recognition method.

[0014] One or more input / output (I / O) interfaces connected between the processor and the memory and configured to enable information interaction between the processor and the memory.

[0015] In a fourth aspect, an embodiment of the present disclosure further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the data recognition method or the ONU recognition method is implemented.

[0016] The embodiment solution of the present disclosure collects one-dimensional time-series data; performs two-dimensional feature mapping on the one-dimensional time-series data to obtain multi-channel two-dimensional data; and uses an image classification model to identify the multi-channel two-dimensional data according to a preset category set, obtaining the target category corresponding to the multi-channel two-dimensional data. The preset category set includes a normal category and at least one abnormal category. Through this embodiment solution, the abnormal features of abnormal data can be captured more accurately, improving the accuracy and efficiency of recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In the drawings of the embodiments of the present disclosure:

[0018] Figure 1 is a flowchart of the data recognition method provided by the embodiment of the present disclosure;

[0019] Figure 2 is a flowchart of the multi-channel two-dimensional eye diagram acquisition method provided by the embodiment of the present disclosure;

[0020] FIG. 3(a) is a schematic diagram of the coupled connection of a convolutional module and an SE module provided by the embodiment of the present disclosure;

[0021] FIG. 3(b) is a schematic diagram of the connection of a convolutional module provided by the related art;

[0022] FIG. 3(c) is a schematic diagram of the direct connection of a convolutional module and an SE module provided by the related art;

[0023] Figure 4 is a schematic diagram of an embodiment of a residual neural network classification model provided by the embodiment of the present disclosure;

[0024] Figure 5 is a flowchart of the ONU recognition method provided by the embodiment of the present disclosure;

[0025] Figure 6 is a schematic diagram of the ONU recognition method provided by the embodiment of the present disclosure;

[0026] Figure 7Schematic diagram of one-dimensional time-series data sampling provided by an embodiment of the present disclosure;

[0027] Figure 8 Schematic diagram of rogue ONU identification in the 1x sampling rate scenario provided by an embodiment of the present disclosure;

[0028] Figure 9 Block diagram of an electronic device provided by an embodiment of the present disclosure. Detailed implementation manners

[0029] To enable those skilled in the art to better understand the technical solutions of the present disclosure, the communication perception data processing method and computer-readable storage medium provided by the embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0030] In the following, the present disclosure will be described more fully with reference to the accompanying drawings. However, the illustrated embodiments may be embodied in different forms, and the present disclosure should not be construed as limited to the embodiments set forth below. On the contrary, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the present disclosure to those skilled in the art.

[0031] The accompanying drawings of the embodiments of the present disclosure are used to provide a further understanding of the embodiments of the present disclosure, and constitute a part of the specification, and are used to explain the present disclosure together with the detailed embodiments, and do not constitute a limitation to the present disclosure. By describing the detailed embodiments with reference to the accompanying drawings, the above and other features and advantages will become more obvious to those skilled in the art.

[0032] The present disclosure can be described with reference to the plan view and / or cross-sectional view by means of the ideal schematic diagram of the present disclosure. Therefore, the example illustrations can be modified according to the manufacturing technology and / or tolerances.

[0033] Without conflict, the embodiments of the present disclosure and the features in the embodiments can be combined with each other.

[0034] The terms used in the present disclosure are only for describing specific embodiments and are not intended to limit the present disclosure. As used in the present disclosure, the term "and / or" includes any and all combinations of one or more of the related listed items. As used in the present disclosure, the singular forms "a" and "the" are also intended to include the plural forms unless the context clearly indicates otherwise. As used in the present disclosure, the terms "comprising", "made of", specify the presence of the described features, wholes, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or their groups.

[0035] Unless otherwise defined, all terms (including technical and scientific terms) used in this disclosure shall have the same meaning as commonly understood by one of ordinary skill in the art. It will also be understood that terms such as those defined in common dictionaries shall be interpreted as having a meaning consistent with their meaning in the context of the relevant art and this disclosure, and shall not be interpreted as having an idealized or overly formal meaning unless expressly so defined in this disclosure.

[0036] To solve the problem of severe loss of effective features in certain sampling scenarios (e.g., including but not limited to low sampling rate scenarios and signal amplifier clipping sampling scenarios), which in turn leads to poor accuracy in identifying certain abnormal data (e.g., based on rogue ONUs) in one-dimensional time series data scenarios, the solution of the embodiment of this disclosure proposes a data recognition and prevention method based on two-dimensional feature mapping, which can be lightweight deployed on edge terminals and accurately identify abnormal data, thereby locating rogue ONUs.

[0037] The solution of the embodiment of this disclosure collects one-dimensional time series data; performs two-dimensional feature mapping on the one-dimensional time series data to obtain multi-channel two-dimensional data; and uses an image classification model to identify the multi-channel two-dimensional data according to a preset category set, obtaining the target category corresponding to the multi-channel two-dimensional data, where the preset category set includes a normal category and at least one abnormal category. Through this embodiment solution, the abnormal features of abnormal data can be captured more accurately, improving the accuracy and efficiency of recognition.

[0038] The solution of the embodiment of this disclosure can be applied to low sampling rate and strictly clipping sampling scenarios. For all bad data, the data of the embodiment of this application can be used for processing to obtain high-quality two-dimensional data, and bad data recognition can be achieved through image classification. The solution of the embodiment of this disclosure can be applied to device management and security protection in fiber access networks, which can include but not be limited to PON (Passive Optical Network), FTTH (Fiber To The Home), etc.

[0039] The following provides a detailed introduction to the solution of the embodiment of this disclosure.

[0040] The embodiment of this disclosure provides a data recognition method, as Figure 1 shown, which includes steps S11 - S13:

[0041] S11. Collect one-dimensional time series data.

[0042] In the embodiment of this disclosure, the one-dimensional time series data may include the collected data in one or more collection scenarios; the collection scenarios may include but not be limited to: low-quality scenarios, such as low sampling rate scenarios, clipping sampling scenarios (such as signal amplifier clipping sampling scenarios), etc.

[0043] In an embodiment of the present disclosure, the low sampling rate can be determined by a preset sampling rate threshold, and a sampling rate less than or equal to the sampling rate threshold can be determined as the low sampling rate.

[0044] In an embodiment of the present disclosure, the low sampling rate scenario may include, but is not limited to, a 1x sampling scenario.

[0045] S12. Perform two-dimensional feature mapping on the one-dimensional time series data to obtain multi-channel two-dimensional data.

[0046] In an embodiment of the present disclosure, the method may further include:

[0047] Divide the one-dimensional time series data into multiple pieces of time series data;

[0048] Perform two-dimensional feature mapping on the multiple pieces of time series data respectively to obtain multiple pieces of two-dimensional eye diagram data.

[0049] In an embodiment of the present disclosure, the multiple pieces may include, but are not limited to, 3 pieces.

[0050] In an embodiment of the present disclosure, for each piece of one-dimensional time series data, the corresponding two-dimensional data can be obtained according to the following scheme.

[0051] In an embodiment of the present disclosure, as Figure 2 shown, performing two-dimensional feature mapping on the one-dimensional time series data to obtain multi-channel two-dimensional data may include steps S21 - S23:

[0052] S21. Based on the acquisition scenario corresponding to the one-dimensional time series data, perform grouping processing on the one-dimensional time series data to obtain multiple groups of one-dimensional time series data to be processed.

[0053] In an embodiment of the present disclosure, based on the acquisition scenario corresponding to the one-dimensional time series data, performing grouping processing on the one-dimensional time series data to obtain multiple groups of one-dimensional time series data to be processed may include:

[0054] In response to the acquisition scenario being a low sampling rate scenario, through a preset interpolation algorithm, process the one-dimensional time series data respectively according to each preset magnification among multiple preset magnifications to obtain multiple groups of one-dimensional time series data to be processed corresponding one-to-one to the multiple preset magnifications.

[0055] In an embodiment of the present disclosure, for a low sampling rate scenario, the collected one-dimensional time series data can be respectively subjected to interpolation methods with different preset magnifications to obtain multiple groups of one-dimensional time series data to be processed with gradually finer waveform trends (for example, 3 groups of one-dimensional time series data to be processed).

[0056] In the embodiments of the present disclosure, the interpolation algorithms adopted may include but are not limited to: Fourier interpolation (such as fast Fourier transform FFT interpolation), cubic spline interpolation, Lagrange interpolation, Newton interpolation, radial basis function interpolation, etc.

[0057] In the embodiments of the present disclosure, the multiple preset magnification factors may include but are not limited to three preset magnification factors combined in an incrementally increasing manner, including but not limited to: {1, 3, 5}, {1, 2, 4}, {1, 2, 6}, etc. Here, the magnification factor values are required to be not less than 1 (a magnification factor value equal to a means that the amount of time-series data after interpolation is a times that before interpolation).

[0058] In the embodiments of the present disclosure, based on the acquisition scenario corresponding to the one-dimensional time-series data, grouping the one-dimensional time-series data to obtain multiple groups of one-dimensional time-series data to be processed may include:

[0059] In response to the acquisition scenario being a limited-amplitude sampling scenario, grouping the one-dimensional time-series data based on the relationship between at least one preset window mean threshold and the window mean of the one-dimensional time-series data to obtain multiple groups of one-dimensional time-series data to be processed.

[0060] In the embodiments of the present disclosure, for the limited-amplitude sampling scenario of a signal amplifier, the acquired one-dimensional time-series data can be divided according to the window mean threshold to obtain multiple groups of one-dimensional time-series data to be processed (for example, 3 groups of one-dimensional time-series data to be processed), and the window means in each group of one-dimensional time-series data to be processed are close.

[0061] In the embodiments of the present disclosure, the at least one window mean threshold may include but is not limited to the first window mean threshold;

[0062] Grouping the one-dimensional time-series data based on the relationship between at least one preset window mean threshold and the window mean of the one-dimensional time-series data to obtain multiple groups of one-dimensional time-series data to be processed may include:

[0063] Determining the one-dimensional time-series data with a window mean greater than the first window mean threshold as the first group of one-dimensional time-series data to be processed;

[0064] Determining the one-dimensional time-series data with a window mean less than the first window mean threshold as the second group of one-dimensional time-series data to be processed;

[0065] Taking the one-dimensional time-series data as the third group of one-dimensional time-series data to be processed;

[0066] The multiple groups of one-dimensional time-series data to be processed are composed of the first group of one-dimensional time-series data to be processed, the second group of one-dimensional time-series data to be processed, and the third group of one-dimensional time-series data to be processed.

[0067] In an embodiment of the present disclosure, for example, three groups of data can be obtained from one-dimensional time-series data, including data with a window mean greater than a set window mean threshold x, data with a window mean less than the set window mean threshold x, and the original data of the one-dimensional time-series data, thus constituting three groups of one-dimensional time-series data to be processed.

[0068] In an embodiment of the present disclosure, the method for selecting the window mean threshold x may include, but is not limited to: (peak mean + trough mean) / 2. Among them, the one-dimensional data collected can be pre-normalized, and the normalization operation is used to determine the approximate range of the window mean threshold x.

[0069] S22. Perform two-dimensional feature mapping on each group of one-dimensional time-series data to be processed to obtain multiple groups of single-channel two-dimensional data.

[0070] The methods for converting one-dimensional time-series data into two-dimensional data may include, but are not limited to:

[0071] 1. Eye Diagram: Suitable for signal processing and communication system analysis. By superimposing waveforms of multiple cycles, a two-dimensional image is formed to display signal jitter, distortion, and noise effects.

[0072] 2. Time Delay Embedding: Suitable for chaotic time-series analysis and pattern recognition. Through delay coordinate embedding, one-dimensional time-series data is mapped into a high-dimensional space, such as constructing a two-dimensional phase space trajectory.

[0073] 3. Resampling Windows: Suitable for deep learning and feature engineering. By sliding windows or resampling at fixed intervals, time series are converted into two-dimensional matrices.

[0074] 4. Short-Time Fourier Transform (STFT) or Wavelet Transform: Suitable for time-frequency analysis. By calculating the spectrum of a signal changing over time, one-dimensional time-series data is converted into a two-dimensional time-frequency diagram (spectrum diagram).

[0075] In an embodiment of the present disclosure, the following takes the eye diagram processing method as an example to illustrate the scheme of performing two-dimensional feature mapping on each group of one-dimensional time-series data to be processed to obtain multiple groups of single-channel two-dimensional data.

[0076] In an embodiment of the present disclosure, performing two-dimensional feature mapping on each group of one-dimensional time-series data to be processed to obtain multiple groups of single-channel two-dimensional data may include:

[0077] Determine the sliding window size based on the number of eyes in the eye diagram and the time-series sampling rate corresponding to each group of one-dimensional time-series data to be processed;

[0078] Split the one-dimensional time series data to be processed in each group based on the sliding window size to obtain the single-channel eye diagram data corresponding to the one-dimensional time series data to be processed in each group.

[0079] In the embodiments of the present disclosure, based on the multiple groups of one-dimensional time series data processed in step S21, the data sampling rate (such as the time series sampling rate) can be multiplied by the number of eyes in the eye diagram (for example, the number of eyes value can be 4) to obtain the size of the sliding window. Using this sliding window, single-channel eye diagrams are respectively drawn for the multiple groups of one-dimensional time series data to achieve two-dimensional feature mapping and obtain multiple groups of single-channel two-dimensional data.

[0080] In the embodiments of the present disclosure, for example, the data sampling rate of the foregoing 3 groups of one-dimensional time series data can be multiplied by the number of eye diagrams to obtain the size of the sliding window. Using this sliding window, single-channel eye diagrams are respectively drawn for the 3 groups of one-dimensional time series data to obtain 3 groups of single-channel eye diagram data.

[0081] S23. Obtain multi-channel two-dimensional data according to the multiple groups of single-channel two-dimensional data.

[0082] In the embodiments of the present disclosure, the multiple groups of single-channel two-dimensional data obtained in step S22 can be merged to obtain multi-channel two-dimensional data.

[0083] In the embodiments of the present disclosure, for example, the above-mentioned 3 groups of single-channel eye diagrams can be merged to obtain 3-channel eye diagram data, such as 3-channel eye diagram data of 224×224 pixels.

[0084] S13. Identify the multi-channel two-dimensional data through an image classification model according to a preset category set to obtain the target category corresponding to the multi-channel two-dimensional data. The preset category set includes a normal category and at least one abnormal category.

[0085] In the embodiments of the present disclosure, the multi-channel two-dimensional data obtained in the above step S12, such as 3-channel two-dimensional data, can be input into a pre-trained image classification model to identify abnormal data in the multi-channel two-dimensional data through the image classification model.

[0086] In the embodiments of the present disclosure, when there are multiple copies of one-dimensional time series data, multi-channel two-dimensional data corresponding to each copy of one-dimensional time series data, such as 3-channel two-dimensional data, can be obtained and respectively input into a pre-trained image classification model to identify abnormal data in each group of multi-channel two-dimensional data.

[0087] In the embodiments of the present disclosure, in different application scenarios, the preset category set may include different categories.

[0088] The image classification model can adopt different models.

[0089] In an embodiment of the present disclosure, for example, the application scenario may include, but is not limited to, the field of rogue ONU identification. The preset category set may include the normal category of multi-channel two-dimensional data and other n - 1 (n > 1, n is a positive integer) groups of abnormal categories.

[0090] In an embodiment of the present disclosure, based on which category (normal category + other n - 1 groups of abnormal categories) the determined multi-channel two-dimensional data belongs to, it is determined whether there is a rogue ONU among the ONUs corresponding to these multi-channel two-dimensional data; if it is determined that there is a rogue ONU, it can be determined which ONU is the rogue ONU. For example, if it is identified that the multi-channel two-dimensional data corresponding to ONU2 is of the normal category and the multi-channel two-dimensional data corresponding to ONU31 is one of the n - 1 types of abnormal categories, it means that ONU31 is a rogue ONU.

[0091] In an embodiment of the present disclosure, for example, the application scenario may include, but is not limited to, the medical field, such as image-based disease identification in the medical field (such as electrocardiogram). The preset category set may include multiple categories such as normal, atrial fibrillation, atrial flutter, premature beats, atrioventricular block, myocardial infarction, ventricular hypertrophy, and arrhythmia.

[0092] In an embodiment of the present disclosure, the image classification model may include, but is not limited to, a residual neural network classification model.

[0093] In an embodiment of the present disclosure, the residual neural network classification model may include a hybrid loss function and a dual-scale residual connection method integrating a channel attention mechanism.

[0094] In an embodiment of the present disclosure, a large amount of multi-channel two-dimensional data can be obtained in advance through the data processing methods of the foregoing steps S11 and S12, and the large amount of multi-channel two-dimensional data can be divided into training data and test data. The pre-established residual neural network classification model architecture is trained with the training data, and the training results are verified based on the test data. Through multiple iterations of updating weights and losses, a high-precision residual neural network classification model for mining difficult samples (samples with a large error between the prediction and the true label during prediction) for abnormal data identification (for example, for rogue ONU identification and positioning) is obtained based on the feedback of the quantitative test results.

[0095] In an embodiment of the present disclosure, as shown in FIG. 3(a), the residual neural network classification model may include: a channel attention module and a convolution module; one channel attention module and one convolution module form a group;

[0096] The channel attention module and the convolution module in the same group perform residual connection in a dual connection form;

[0097] The convolutional module includes multiple convolutional sub - modules, and the channel attention module includes multiple channel attention sub - modules;

[0098] Each convolutional sub - module in each convolutional module can lead to any one of the channel attention sub - modules in the same group of channel attention modules.

[0099] In the embodiments of the present disclosure, to improve the inference performance of the residual neural network classification model without increasing additional computational overhead, a new residual connection method is proposed. In many related residual network structures, as shown in Figure 3(b), the residual connection is usually carried out between convolutional modules composed of one or more convolutional layers, or as shown in Figure 3(c), between adjacent convolutional modules and SE modules. The solution of the embodiments of the present disclosure also introduces an SE module (i.e., the channel attention module) into the residual network, and regards an SE module {se1, se2, se3...} and a convolutional module {conv1, conv2, conv3...} as a group. On this basis, a dual connection method is constructed, so that the convolutional blocks conv in each group, such as conv1, conv2, conv3..., can lead to any one of the SE modules in the same group, such as se1, se2, se3..., thereby expanding the path of the residual connection without bringing huge computational overhead. For example, taking the convolutional sub - module conv1 as an example, through the new connection method of the solution of the embodiments of the present disclosure, the convolutional sub - module conv1 can reach any one of se1, se2, se3... without increasing computational consumption.

[0100] In the embodiments of the present disclosure, the recognition of multi - channel two - dimensional data by the image classification model may include:

[0101] During the process of constructing the residual connection based on multi - channel two - dimensional data by the residual neural network classification model, the data obtained after each convolutional module performs convolution is divided into two groups of data, and only one group of the two groups of data is subjected to the channel attention mechanism through the channel attention module, while the other group of data in the two groups of data remains unchanged.

[0102] In the embodiments of the present disclosure, as Figure 4As shown in the figure, it is a structural embodiment of the residual neural network classification model according to an embodiment of the present disclosure, including a first convolutional layer (Conv-1), a first batch normalization layer (BatchNorm-1), a first activation layer (ReLU-1), a second convolutional layer (Conv-2), a second batch normalization layer (BatchNorm-2), a second activation layer (ReLU-2), a dual-connected channel attention module and a convolutional module (for example, including a first convolutional module Conv_1, a second convolutional module Conv_2, a first SE module SE_1, and a second SE module SE_2), a third convolutional layer (Conv-3), a third batch normalization layer (BatchNorm-3), a third activation layer (ReLU-3), a pooling layer (Pooling), and a linear layer (Linear).

[0103] In the embodiment of the present disclosure, after the data x output by the second activation layer (ReLU-2) is input into the first convolutional module Conv_1 for convolution, the output data is split into two parts, divided into data x1 and x2. Among them, data x1 is directly output as y1, data x2 is input into the first SE module SE_1 to execute the channel attention mechanism and then output y2. Data y1 and data y2 are input into the second convolutional module Conv_2 together. After convolution, the output data is split into two parts, divided into data x3 and x4. Data x3 is directly output as y3, data x4 is input into the second SE module SE_2 to execute the channel attention mechanism and then output y4. Data y3 and data y4 are input into the third convolutional layer (Conv-3) together.

[0104] In the embodiment of the present disclosure, through the embodiment solution, the extraction of signal features at different scales can be realized, in order to better balance the local and global features of the data, so that the residual neural network classification model can fully pay attention to local details while also considering the global changes of the data.

[0105] In the embodiment of the present disclosure, the residual neural network classification model may include: a hybrid loss function;

[0106] The hybrid loss function may include, but is not limited to: the fusion focal loss Focal_Loss and the cross-entropy loss CE_Loss.

[0107] In the embodiment of the present disclosure, there is a phenomenon that pairwise conflicting data has high feature similarity. Therefore, a hybrid loss Total_Loss that combines the fusion focal loss Focal_Loss and the cross-entropy loss CE_Loss (CrossEntropy Loss) is introduced. Its overall architecture is as follows:

[0108]

[0109] In the embodiments of the present disclosure, by introducing a dual-scale residual connection method and a hybrid loss function that incorporate a channel attention mechanism into the residual neural network classification model, the residual neural network classification model can more deeply extract effective features from the input multi-channel two-dimensional eye diagram data. Among them, the channel attention module (referred to as the SE module) enhances the network's attention to important features by adaptively reweighting the feature channels, thereby improving the performance of the residual neural network classification model.

[0110] In the embodiments of the present disclosure, when multiple copies of the obtained multi-channel two-dimensional data are available, after the multi-channel two-dimensional data is identified by the image classification model according to a preset category set to obtain the target category corresponding to the multi-channel two-dimensional data, the method may further include:

[0111] Determine whether the one-dimensional time series data is normal or abnormal based on the multiple target categories identified corresponding to multiple copies of the multi-channel two-dimensional data. In the embodiments of the present disclosure, determining whether the one-dimensional time series data is normal or abnormal based on the multiple target categories identified corresponding to multiple copies of the multi-channel two-dimensional data may include:

[0112] Calculate the similarity of the multiple target categories;

[0113] In response to the similarity being lower than a preset similarity threshold, determine that the multiple target categories are normal and determine that the one-dimensional time series data is normal; or, in response to the similarity being higher than or equal to the similarity threshold, determine that the multiple target categories are abnormal and determine that the one-dimensional time series data is abnormal.

[0114] In the embodiments of the present disclosure, when it is determined that the input multi-channel two-dimensional data is abnormal, it may be determined that the corresponding one-dimensional time series data is abnormal, and when it is determined that the input multi-channel two-dimensional data is normal, it may be determined that the corresponding one-dimensional time series data is normal.

[0115] The solution of the embodiments of the present disclosure also provides an ONU identification method, as Figure 5 、 Figure 6 shown, which includes steps S31 - S32:

[0116] S31. Based on the above data identification method, obtain the target category corresponding to the one-dimensional time series data of the optical network unit ONU;

[0117] S32. Identify the category of the ONU according to the target category.

[0118] In a fiber optic access network, such as PON, FTTH, etc., the ONU (Optical Network Unit) is a key device that connects to the OLT (Optical Line Terminal) and transmits optical signals to the end users. However, with the expansion of the network scale and the diversification of user requirements, there may occasionally be rogue ONU devices in the network that are unauthorized or have abnormal behaviors. Such rogue ONUs are usually caused by illegal access, configuration errors, or device failures, etc., which can lead to abnormal occupation of network resources, degradation of service quality, and even serious security hazards. Therefore, how to efficiently detect and identify these rogue ONU devices has become an important challenge for ensuring the security and stable operation of the fiber optic access network.

[0119] Common rogue ONU detection methods mainly rely on traditional means such as comparing device identification information, manually verifying configuration parameters, and applying preset user rules. However, with the increasing complexity of the network environment, these methods face many challenges such as low detection efficiency, high false alarm rate, and poor adaptability to ONU devices of different models from multiple manufacturers. The current method of relying on manual step-by-step comparison and positioning has been difficult to meet the requirements for rapid and accurate detection of rogue ONUs in a complex network environment. Therefore, there is an urgent need for a more intelligent and automated detection method to improve the accuracy and response speed of rogue ONU positioning, while reducing the burden of manual operation and maintenance.

[0120] Currently, some methods based on one-dimensional data analysis have been applied to the detection of rogue ONUs. These methods analyze the time series of device signals to identify possible abnormal behaviors. However, one-dimensional data analysis has limitations in accurately capturing the abnormal behavior characteristics of complex signals, especially in dealing with low sampling rate scenarios (one-fold sampling rate) and strictly limited amplitude scenarios (quadruple limited amplitude sampling of signal amplifiers). In these scenarios, these methods often struggle to comprehensively reflect the global state and dynamic changes of the signal.

[0121] In the embodiments of the present disclosure, the one-dimensional time series data may include but are not limited to:

[0122] The normal light emission data of each ONU among multiple ONUs and the dual-ONU conflict data between any one ONU and each of the other ONUs among the multiple ONUs.

[0123] In the embodiments of the present disclosure, for the low sampling rate scenario and the limited amplitude sampling scenario, the normal light emission data of a single ONU and the dual-ONU conflict data can be collected respectively to obtain a rogue ONU random light emission time series database. Specifically, as Figure 6As shown, taking n ONUs as an example (for example, ONU_1, ONU_2, ONU_3, ……, ONU_n in the ONU group), the description is as follows. First, the normal light-emitting data of each single ONU among the n ONUs within a certain time sequence can be collected. Secondly, the pairwise conflict data between this ONU and the other n - 1 ONUs can be collected. Finally, for each ONU, sampling data containing n data can be obtained.

[0124] In the embodiment of the present disclosure, the conflict data is the collected data obtained in a real network environment by triggering specific conflict events (such as, including but not limited to: sequentially setting the random signal of other ONUs to long light emission).

[0125] In the embodiment of the present disclosure, identifying the category of the ONU according to the target category may include:

[0126] In response to the target category being the normal category in the preset category set, determining that the ONU is a normal ONU;

[0127] In response to the target category being the abnormal category in the preset category set, determining that the ONU is an abnormal ONU.

[0128] In the embodiment of the present disclosure, based on the target category corresponding to the one-dimensional time sequence data of the ONU, the corresponding ONU of this one-dimensional time sequence data can be determined, so as to identify and locate the rogue ONU.

[0129] In the embodiment of the present disclosure, several rogue ONU location methods based on one-dimensional data are introduced as a comparison group to evaluate the reliability of the results. These location methods include: SVM (Support Vector Machine), LR (Logistic Regression), and ROCKET (Random Concolution KernelTransform, a time series classification method).

[0130] In the embodiment of the present disclosure, as shown in Table 1, the quantitative evaluation results of different methods are summarized. The higher accuracy rate shows that the two-dimensional feature mapping strategy proposed in the embodiment of the present disclosure can better overcome the problem of difficult classification of one-dimensional time sequence data in a low-quality scenario. The method proposed in the embodiment of the present disclosure uses a targeted two-dimensional feature mapping strategy and an attention mechanism to greatly affect the processing result without increasing additional computational overhead.

[0131] Table 1

[0132]

[0133] In the embodiment of the present disclosure, a detailed embodiment of the solution of the embodiment of the present disclosure is given below.

[0134] For example, a classifier (i.e., a residual neural network classification model) can be configured for each of the n ONUs in advance. When the ONU_2 among the n ONUs exhibits the phenomenon of random backlight (such as emitting light not in accordance with the control requirements), ONU_2 can be regarded as a rogue ONU. The one-dimensional time-series data sampling process in the low sampling rate (such as 1-fold sampling rate) and amplifier-limited sampling scenario (four-fold sampling rate) can be as Figure 7 shown.

[0135] The identification of rogue ONUs and result verification in the 1-fold sampling rate scenario are as Figure 8 shown. For the one-dimensional time-series data collected from ONU_2, it can be divided into three parts. For each part of the one-dimensional time-series data, two-dimensional data mapping is respectively performed based on the two-dimensional feature mapping method for low sampling rates to obtain corresponding multi-channel two-dimensional eye diagram data. For example, eye Figure 1 , eye Figure 2 , and eye diagram 3. Respectively input eye Figure 1 , eye Figure 2 , and eye diagram 3 into the residual neural network classification model based on the attention mechanism for abnormal data identification (for realizing the identification of rogue ONUs), and obtain the identification results respectively. For example, result 1, result 2, and result 3. Perform rule matching on result 1, result 2, and result 3. For example, detect the result similarity of result 1, result 2, and result 3. When the similarity of result 1, result 2, and result 3 is lower than the similarity threshold, it can be determined that the result similarity is low, and it is determined that the multi-channel two-dimensional eye diagram data is normal, so the corresponding ONU_2 is normal; when the similarity of result 1, result 2, and result 3 is higher than or equal to the similarity threshold, it can be determined that the result similarity is high, and it is determined that the multi-channel two-dimensional eye diagram data is abnormal, so the corresponding ONU_2 is abnormal, thereby locating the rogue ONU as ONU_2.

[0136] The embodiments of the present disclosure also provide an electronic device 100, as Figure 9 shown, including:

[0137] One or more processors 101;

[0138] A memory 102, on which one or more programs are stored. When the one or more programs are executed by the one or more processors, the one or more processors 101 implement the data recognition method, or the ONU recognition method;

[0139] One or more input / output I / O interfaces 103, connected between the processor 101 and the memory 102, configured to implement the information interaction between the processor 101 and the memory 102.

[0140] Embodiments of the present disclosure also provide a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the data recognition method, or the ONU recognition method, is implemented.

[0141] Embodiments of the present disclosure also provide a computer program product, which includes a computer program. When the computer program is executed by a processor, the data recognition method, or the ONU recognition method, is implemented.

[0142] In the embodiments of the present disclosure, any of the above-described embodiments of the data recognition method is applicable to the embodiments of the electronic device, storage medium, and program product, and will not be elaborated herein one by one.

[0143] Those of ordinary skill in the art can understand that all or some of the functional modules / units disclosed above can be implemented as software, firmware, hardware, and their appropriate combinations.

[0144] In the hardware implementation, the division between the functional modules / units mentioned above does not necessarily correspond to the division of physical components; for example, one physical component can have multiple functions, or one function or step can be executed by several physical components in cooperation.

[0145] Some physical components or all physical components can be implemented as software executed by a processor, such as a central processing unit (CPU), a digital signal processor, or a microprocessor, or can be implemented as hardware, or can be implemented as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include a computer storage medium (or non-transitory medium) and a communication medium (or transitory medium). As is well known to those of ordinary skill in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules, or other data. Computer storage media include, but are not limited to, random access memory (RAM, more specifically SDRAM, DDR, etc.), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory (FLASH), or other disk memories; compact disc read-only memory (CD-ROM), digital versatile disc (DVD), or other optical disc memories; magnetic cartridges, tapes, magnetic disk storage, or other magnetic memories; and any other medium that can be used to store the desired information and can be accessed by a computer. In addition, as is well known to those of ordinary skill in the art, a communication medium generally contains computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transmission mechanism, and can include any information delivery medium.

[0146] The present disclosure has disclosed exemplary embodiments, and although specific terms are employed, they are used only and should be interpreted only as general illustrative meanings and not for the purpose of limitation. In some instances, it will be apparent to those skilled in the art that, unless otherwise expressly stated, features, characteristics, and / or elements described in connection with a particular embodiment may be used alone or in combination with features, characteristics, and / or elements described in connection with other embodiments. Accordingly, those skilled in the art will understand that various forms and details of changes may be made without departing from the scope of the present disclosure as set forth by the appended claims.

Claims

1. A data recognition method, comprising: Collecting one-dimensional time series data; Performing two-dimensional feature mapping on the one-dimensional time series data to obtain multi-channel two-dimensional data; Identifying the multi-channel two-dimensional data according to a preset category set through an image classification model to obtain a target category corresponding to the multi-channel two-dimensional data, where the preset category set includes a normal category and at least one abnormal category.

2. The data recognition method according to claim 1, wherein The performing two-dimensional feature mapping on the one-dimensional time series data to obtain multi-channel two-dimensional data includes: Based on the acquisition scenario corresponding to the one-dimensional time series data, performing grouping processing on the one-dimensional time series data to obtain multiple groups of one-dimensional time series data to be processed; Performing two-dimensional feature mapping on each group of the one-dimensional time series data to be processed respectively to obtain multiple groups of single-channel two-dimensional data; According to the multiple groups of single-channel two-dimensional data, obtaining the multi-channel two-dimensional data.

3. The data recognition method according to claim 2, wherein, The performing two-dimensional feature mapping on each group of the one-dimensional time series data to be processed respectively to obtain multiple groups of single-channel two-dimensional data includes: Determining a sliding window size based on the number of eyes in an eye diagram and the time series sampling rate corresponding to each group of the one-dimensional time series data to be processed; Based on the sliding window size, splitting each group of the one-dimensional time series data to be processed to obtain single-channel eye diagram data corresponding to each group of the one-dimensional time series data to be processed.

4. The data recognition method according to claim 3, wherein The obtaining the multi-channel two-dimensional data according to the multiple groups of single-channel two-dimensional data includes: Merging the multiple groups of single-channel eye diagrams to obtain multi-channel eye diagram data.

5. The data recognition method according to claim 2, wherein, The performing grouping processing on the one-dimensional time series data based on the acquisition scenario corresponding to the one-dimensional time series data to obtain multiple groups of one-dimensional time series data to be processed includes: In response to the acquisition scenario being a low sampling rate scenario, processing the one-dimensional time series data respectively according to each preset magnification among multiple preset magnifications through a preset interpolation algorithm to obtain multiple groups of the one-dimensional time series data to be processed corresponding one-to-one to the multiple preset magnifications.

6. The data recognition method according to claim 2, wherein The performing grouping processing on the one-dimensional time series data based on the acquisition scenario corresponding to the one-dimensional time series data to obtain multiple groups of one-dimensional time series data to be processed includes: In response to the acquisition scenario being a clipping sampling scenario, grouping the one-dimensional time series data based on the relationship between at least one preset window mean threshold and the window mean of the one-dimensional time series data to obtain multiple groups of one-dimensional time series data to be processed.

7. The data recognition method according to claim 6, wherein, The at least one window mean threshold includes a first window mean threshold; The grouping the one-dimensional time series data based on the relationship between at least one preset window mean threshold and the window mean of the one-dimensional time series data to obtain multiple groups of one-dimensional time series data to be processed includes: Determining the one-dimensional time series data with a window mean greater than the first window mean threshold as the first group of one-dimensional time series data to be processed; Determining the one-dimensional time series data with a window mean less than the first window mean threshold as the second group of one-dimensional time series data to be processed; Taking the one-dimensional time series data as the third group of one-dimensional time series data to be processed; The multiple groups of one-dimensional time series data to be processed are composed of the first group of one-dimensional time series data to be processed, the second group of one-dimensional time series data to be processed, and the third group of one-dimensional time series data to be processed.

8. The data recognition method according to claim 1, wherein The image classification model includes: a residual neural network classification model; the residual neural network classification model includes: a channel attention module and a convolution module; one channel attention module and one convolution module form a group; The channel attention module and the convolution module in the same group are residually connected in a dual connection form; The convolution module includes a plurality of convolution sub-modules, and the channel attention module includes a plurality of channel attention sub-modules; Each convolution sub-module in each convolution module can lead to any one of the channel attention sub-modules in the channel attention module of the same group.

9. The data recognition method according to claim 8, wherein, The recognition of the multi-channel two-dimensional data by the image classification model includes: During the process of constructing the residual connection based on the multi-channel two-dimensional data by the residual neural network classification model, the data obtained after each convolution module performs convolution is divided into two groups of data, and only one group of the two groups of data is passed through the channel attention module to perform the channel attention mechanism, while keeping the other group of the two groups of data unchanged.

10. The data recognition method according to claim 1, wherein, The two-dimensional feature mapping of the one-dimensional time series data to obtain multi-channel two-dimensional data further includes: Dividing the one-dimensional time series data into multiple pieces of time series data; Performing two-dimensional feature mapping on the multiple pieces of time series data respectively to obtain multiple pieces of the multi-channel two-dimensional data; After the multi-channel two-dimensional data is recognized by the image classification model according to a preset category set to obtain the target category corresponding to the multi-channel two-dimensional data, the method further includes: Determining whether the one-dimensional time series data is normal according to the multiple target categories identified corresponding to the multiple pieces of the multi-channel two-dimensional data.

11. The data identification method according to claim 10, wherein, The determining whether the one-dimensional time series data is normal according to the multiple target categories identified corresponding to the multiple pieces of the multi-channel two-dimensional data includes: Calculating the similarity of the multiple target categories; In response to the similarity being lower than a preset similarity threshold, determining that the one-dimensional time series data is normal; or, in response to the similarity being higher than or equal to the similarity threshold, determining that the one-dimensional time series data is abnormal.

12. An ONU recognition method, which includes: Based on the data recognition method according to any one of claims 1-11, obtaining the target category corresponding to the one-dimensional time series data of the optical network unit ONU; Recognizing the category of the ONU according to the target category.

13. The ONU identification method according to claim 12, wherein, The one-dimensional time series data includes: The normal light emission data of each ONU among multiple ONUs and the dual ONU conflict data between any one ONU and each other ONU among the multiple ONUs.

14. The ONU identification method according to claim 12, wherein, The recognizing the category of the ONU according to the target category includes: In response to the target category being a normal category in the preset category set, determining that the ONU is a normal ONU; In response to the target category being an abnormal category in the preset category set, determining that the ONU is an abnormal ONU.

15. An electronic device, including: One or more processors; A memory having one or more programs stored thereon, which when executed by the one or more processors cause the one or more processors to implement the data recognition method according to any one of claims 1-11, or the ONU recognition method according to any one of claims 12-14; One or more input / output I / O interfaces, connected between the processor and the memory, configured to implement information interaction between the processor and the memory.

16. A computer program product comprising a computer program which, when executed by a processor, implements the data recognition method according to any one of claims 1-11, or the ONU recognition method according to any one of claims 12-14.

Images