Optical transmitter and receiver equipment fault prediction method based on deep learning

Abstract

The invention relates to an optical transmitter and receiver equipment fault prediction method based on deep learning. The method comprises the steps of receiving and preprocessing historical operation performance index data and historical environment state monitoring data of an optical transmitter and receiver; obtaining a to-be-tested operation performance index data sequence and a to-be-tested environment state monitoring data sequence from the historical operation performance index data and the historical environment state monitoring data of the optical transceiver through a time sliding window with a self-adaptive size; respectively inputting a to-be-detected operation performance index data sequence into the LSTM model and the first TCN model for feature extraction to obtain a global feature sequence and a local feature sequence, inputting a to-be-detected environment state monitoring data sequence into the second TCN model for feature extraction to obtain an external factor feature sequence, and obtaining a to-be-detected environment state monitoring data sequence; and carrying out weighted fusion on the global feature sequence, the local feature sequence and the external factor feature sequence by using a preset weighting factor to obtain a fused feature sequence, inputting the fused feature sequence into a Transformer encoder based on feature dimension attention and time dimension attention, and outputting a prediction result, so that the accuracy of the prediction result can be remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the field of optical terminal fault prediction, and in particular relates to an optical terminal equipment fault prediction method based on deep learning. Background Art

[0002] With the rapid development and widespread application of communication technology, optical terminals, as core equipment in modern communication networks, undertake important tasks such as sending, receiving and processing optical signals. The operating stability of optical terminal equipment directly affects the performance and service quality of the communication network. Traditional fault management methods, such as regular maintenance and post-fault repair, have problems such as lag, low efficiency and high reliance on manual experience, and are difficult to cope with the complexity and massiveness of equipment data. These problems not only reduce the reliability of the equipment, but may also cause communication interruptions and cause economic losses.

[0003] In the field of optical terminal fault prediction, traditional machine learning methods have difficulty in fully capturing the time series characteristics of equipment performance data. Although LSTM (Long Short-Term Memory) networks can effectively process long-term dependencies in time series data, they are limited in capturing local features. In contrast, TCN (Time Convolutional Network) is good at capturing local features, but may not be as good as LSTM when processing long-term dependencies. Therefore, using LSTM or TCN networks alone is difficult to meet the comprehensiveness and accuracy requirements of optical terminal fault prediction.

[0004] At the same time, the operating status of the optical transceiver is not only affected by the internal factors of the equipment, but also significantly affected by external environmental factors (such as temperature, humidity, electromagnetic interference, etc.). Traditional fault prediction methods often ignore these external environmental factors, resulting in inaccurate prediction results.

[0005] In addition, traditional fault prediction methods often use a fixed-size time sliding window to intercept data sequences. However, in practical applications, the operating data of optical transceivers may have different time scales and fluctuations, and a fixed-size time sliding window may not be able to adapt to such changes, thus affecting the accuracy of prediction.

[0006] Finally, the performance indicator data of optical terminal equipment is usually multi-dimensional and multi-variable. Existing methods are difficult to effectively process high-dimensional data, resulting in inaccurate prediction results. Summary of the invention

[0007] In view of the above problems, the present invention provides a method for predicting optical terminal equipment failure based on deep learning, which is characterized by comprising:

[0008] S1: Receive and pre-process the historical operating performance index data and historical environmental status monitoring data of the optical terminal;

[0009] S2: Obtain the to-be-tested operating performance indicator data sequence and the to-be-tested environmental status monitoring data sequence from the historical operating performance indicator data and the historical environmental status monitoring data of the optical terminal through a time sliding window of adaptive size;

[0010] S3: Input the data sequence of the operating performance indicator to be tested into the LSTM model and the first TCN model respectively for feature extraction to obtain a global feature sequence and a local feature sequence;

[0011] S4: Input the monitoring data sequence of the environmental state to be tested into the second TCN model for feature extraction to obtain the external factor feature sequence;

[0012] S5: using a preset weighting factor to weight the global feature sequence, the local feature sequence and the external factor feature sequence to obtain a fused feature sequence;

[0013] S6: Input the fused feature sequence into the Transformer encoder based on feature dimension attention and time dimension attention, and output the prediction result.

[0014] The present invention has at least the following beneficial effects

[0015] The present invention uses a time sliding window of adaptive size to intercept the data sequence. This adaptive method can dynamically adjust the window size according to the operating data characteristics and volatility of the optical terminal, so as to more accurately capture the time series features in the data and improve the accuracy of the prediction. This improvement enables the model to adapt to different data scenarios more flexibly, and improves the practicality and generalization ability of the prediction. By combining the LSTM network and the TCN network, the time series features of the optical terminal equipment are extracted from both the global and local levels. The LSTM network is responsible for capturing long-term dependencies, while the TCN network focuses on the extraction of local features. This combination method not only improves the model's processing ability for complex time series data, but also improves the accuracy of fault prediction. External environmental factors are incorporated into the fault prediction model, and the external environment status monitoring data is feature extracted through the TCN network to obtain the external factor feature sequence. This improvement enables the model to understand the operating status of the equipment more comprehensively, and further improves the accuracy and robustness of the prediction. The present invention introduces a Transformer encoder based on feature dimension attention and time dimension attention. This improved Transformer model can better capture the potential information and time series relationships in multi-dimensional time series data, and automatically assign weights to different time steps and feature dimensions through the attention mechanism, thereby further improving the accuracy and robustness of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0017] Figure 2 Schematic diagram of the structure of the cell unit in the LSTM model of the present invention;

[0018] Figure 3 Schematic diagram of the structure of the residual block of the TCN model of the present invention;

[0019] Figure 4 This is a schematic diagram of the structure of the Transformer encoder based on feature dimension attention and time dimension attention of the present invention. DETAILED DESCRIPTION

[0020] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0021] See also Figure 1 The present invention provides a method for predicting optical terminal equipment failure based on deep learning, comprising:

[0022] S1: Receive and pre-process the historical operating performance index data and historical environmental status monitoring data of the optical terminal;

[0023] Preferably, the operating performance indicator data of the optical terminal include: one or more of input optical power, output optical power, device temperature, board temperature, CPU utilization and memory utilization, wherein each operating performance indicator data includes three forms, namely, the instantaneous value at the current moment, and the maximum value and minimum value within the acquisition period; the environmental status monitoring data of the optical terminal include: one or more environmental indicators of ambient temperature, ambient humidity, ambient pressure and ambient wind speed.

[0024] In this embodiment, it is necessary to pre-process the historical operation performance index data and historical environmental status monitoring data of the optical terminal, including:

[0025] Clean the historical operating performance indicator data and historical environmental status monitoring data of the optical terminal, delete inconsistent data types: filter out abnormal data in the indicator data, and delete those data that are inconsistent with the preset format or type; fill in missing values ​​of the data: use reasonable methods to fill in missing data, such as mean, default value, or median. Delete duplicate data: and delete duplicate entries in the data set to ensure that each record is unique, avoid redundant data interfering with model training, and normalize each indicator.

[0026] In this embodiment, the operating performance index data of the optical terminal is usually collected in real time by the built-in sensors and monitoring system of the optical terminal, and transmitted to the central monitoring center or data warehouse through the network. When receiving this data, we need to ensure the integrity and accuracy of the data for subsequent analysis and decision-making.

[0027] Preferably, preprocessing the historical environmental status monitoring data or the historical operating performance indicator data includes:

[0028] S11: Use the mean filling method to fill the missing values ​​in the historical environmental status monitoring data or historical operation performance index data:

[0029]

[0030] in, represents the mean value of the jth environmental indicator in the historical environmental status monitoring data or the jth performance indicator in the historical operation performance indicator data, n j Indicates the number of non-null values ​​of the jth environmental indicator or performance indicator;

[0031] S12: Use the LOG minimum-maximum normalization method to normalize the historical environmental status monitoring data or historical operating performance index data, and map all eigenvalues ​​to the range of [0,1]:

[0032]

[0033] Among them, x j,min is the minimum characteristic value of the jth environmental indicator in the historical environmental status monitoring data or the minimum characteristic value of the jth performance indicator in the historical operation performance indicator data; x j,max is the maximum eigenvalue of the jth environmental indicator in the historical environmental status monitoring data or the maximum eigenvalue of the jth performance indicator in the historical operation performance indicator data; x j,log_norm It is the normalized characteristic value of the j-th environmental indicator in the historical environmental status monitoring data or the normalized characteristic value of the j-th performance indicator in the historical operation performance indicator data.

[0034] S2: Obtain the to-be-tested operating performance indicator data sequence and the to-be-tested environmental status monitoring data sequence from the historical operating performance indicator data and the historical environmental status monitoring data of the optical terminal through a time sliding window of adaptive size;

[0035] Preferably, the step of obtaining the to-be-tested operating performance indicator data sequence and the to-be-tested environmental status monitoring data sequence through a time sliding window of adaptive size comprises:

[0036] S21: Construct an adaptive adjustment factor based on the historical operating performance index data of the optical terminal:

[0037] S22: Adaptively adjust the size of the time sliding window according to the adaptive adjustment factor:

[0038] W(d)=W(d-1)*α(d)

[0039] Among them, α(d) represents the adaptive adjustment factor at time d, W(d-1) represents the size of the time sliding window at time d-1; W(d) represents the size of the time sliding window at time d;

[0040] S23: using a time sliding window of adaptive size, respectively extracting the data closest to the current time t from the historical operation performance indicator data and the historical environmental status monitoring data as the operation performance indicator data sequence to be tested and the environmental status monitoring data sequence to be tested.

[0041] Preferably, the adaptive adjustment factor includes:

[0042] α(d)=max(α min ,1+tanh(ΔP(d)-Δp threshold ))

[0043]

[0044]

[0045]

[0046]

[0047] Where N represents the number of operating performance indicators, ΔP(d) represents the performance fluctuation amplitude of the optical terminal at time d; e represents the natural constant, α min Indicates the minimum threshold of the adjustment factor, such as 0.5, Δp threshold represents the mean of all performance fluctuations from the first time point to time d-1, x i ′(d) and x′(d-1) represent the normalized values ​​of the i-th performance indicator at time d and d-1, respectively; w i is the weight of the i-th performance indicator, Var(x i ) represents the i-th performance index x i The variance in the historical operating performance indicator data; Indicates the i-th performance indicator x in the historical running performance indicator data i The average value of; T represents the time length of the historical running performance indicator data.

[0048] In this embodiment, by adjusting the size of the time sliding window, the operation performance index data and environmental status monitoring data closest to the current moment can be intercepted and analyzed more flexibly to adapt to different data fluctuations and operation requirements. If the adaptive adjustment factor is large (greater than 1), it means that the data fluctuation is large or there are many unstable performance indicators. At this time, the size of the time sliding window should be appropriately increased to include more historical data for analysis. If the adaptive adjustment factor is small (less than 1), it means that the data fluctuation is small and there are many stable performance indicators. At this time, the size of the time sliding window can be reduced to analyze the recent data changes more focused.

[0049] During model training, the collected optical terminal operation performance indicator data is segmented through a time sliding window, the entire time series is divided into several time windows of adaptive size, and a suitable step size is set to ensure the overlap between each time window, thereby increasing the model's sensitivity to data changes.

[0050] For example, a time sliding window of size W(d) is defined. The size of the time sliding window at the initial moment and the window interval step are set by technical personnel in this field based on experience. Assuming that the historical time series of the optical terminal operation performance index data X = {x(1), x(2), ..., x(200)}, and the historical time series of the environmental status monitoring data Y = {y(1), y(2), ..., y(200)}, there are 200 time step units. The initial window size W(0)=100, the window interval step size S=10, the initial window sequence range of performance indicator data is {x(1), x(2),..., x(100)}, and the initial window sequence range of environmental status monitoring data is {y(1), y(2),..., y(100)}. If W(1)=50 is calculated based on the performance fluctuation amplitude, the adaptively generated performance indicator data window sequence is {x(11), x(12),..., x(60)}, and the environmental status monitoring data window sequence is {y(11), y(12),..., y(60)}, and so on. The operating performance indicator data and environmental status monitoring data are intercepted through a time sliding window of adaptive size to obtain a training sample set. Based on the training sample set and the expert's fault labeling, a cross entropy loss function is constructed to train the model.

[0051] In this embodiment, the present invention uses a time sliding window of adaptive size to intercept the data sequence. This adaptive method can dynamically adjust the window size according to the operating data characteristics and volatility of the optical terminal, so as to more accurately capture the time series characteristics in the data and improve the accuracy of the prediction. This improvement enables the model to adapt to different data scenarios more flexibly, and improves the practicality and generalization ability of the prediction.

[0052] S3: Input the data sequence of the operating performance indicator to be tested into the LSTM model and the first TCN model respectively for feature extraction to obtain a global feature sequence and a local feature sequence;

[0053] Preferably, the data sequence of the operating performance indicator to be tested is input into the LSTM model for feature extraction, and the global feature sequence obtained includes: the data sequence of the operating performance indicator to be tested is expressed as in, W(t) represents the number of time steps, that is, the width of the time sliding window; r represents the feature dimension of the data; the running performance indicator data sequence is input into the LSTM model to generate the hidden state h of each time step i ; The hidden states generated by the LSTM model at all time steps constitute the global feature sequence F G ={h t-W(t) ,h t-W(t)+1 ,...,h j ,...,h t}, h j represents the jth feature in the global feature sequence, t represents the current moment, W(t) Indicates the size of the sliding window at the current time t.

[0054] See also Figure 2 , the basic unit of the LSTM network includes three gating mechanisms: input gate, forget gate and output gate. The input gate is responsible for selecting which input information needs to be retained, and the output gate determines which information to extract from the memory. The implementation of both depends on the cooperation of the sigmoid layer and the tanh layer. The sigmoid layer compresses the data to the range of [0,1], indicating the retention ratio of information; the tanh layer maps the data to the range of [-1,1] to adjust the strength of the input or output information. The function of the forget gate is to control whether to retain the memory state h of the previous moment. t-1 , the proportion of retention is determined by the sigmoid layer. The specific formula is as follows:

[0055] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0056] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0057]

[0058]

[0059] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0060]

[0061] In the LSTM unit, f t and i t Represent the states of the forget gate and input gate respectively, W f and W i is the corresponding weight matrix. [h t-1 ,x t ] means concatenating two vectors into a new long vector, b f , b i They represent the bias terms of the forget gate and the input gate respectively, and σ is the sigmoid function. is the candidate state of the input. t and long-term memory c t-1 Combined together, a new unit state c is formed t .h t is the output at time t. This memory update mechanism enables LSTM to combine long-term memory c t-1 and the contribution of the current input c t , thereby generating a new memory state c t And output h t .

[0062] S4: Input the monitoring data sequence of the environmental state to be tested into the second TCN model for feature extraction to obtain the external factor feature sequence;

[0063] Preferably, the first TCN model and the second TCN model each include a plurality of residual blocks, which are sequentially connected in series.

[0064] Please refer to 3, Figure 3 The schematic diagram of the structure of the residual block in the TCN model, each residual block is connected by a residual connection. The TCN model in the embodiment of the present invention is mainly used to extract the local timing information of the operation data of the optical terminal equipment, especially to capture the long-distance dependencies and local patterns in the data through convolution operations. TCN avoids the gradient vanishing problem through causal convolution and dilated convolution, and can efficiently process long sequence data.

[0065] S5: using a preset weighting factor to weight the global feature sequence, the local feature sequence and the external factor feature sequence to obtain a fused feature sequence;

[0066] Preferably, the fusion signature sequence includes:

[0067] F 内部 =αF LSTM +βF TCN

[0068] F Fusion =γF 内部 +δF 外部

[0069] Among them, F Fusion represents the fusion feature sequence, F TCN represents the local feature sequence, F LSTM represents the global feature sequence, α, β, γ and δ represent weighting factors, and α+β=1, γ+δ=1.

[0070] S6: Input the fused feature sequence into the Transformer encoder based on feature dimension attention and time dimension attention, and output the prediction result.

[0071] Preferably, the Transformer encoder based on feature dimension attention and time dimension attention includes: multiple Transformer coding layers, and the multi-head attention layer of each Transformer coding layer includes: a first multi-head attention module and a second multi-head attention module; the first multi-head attention module performs a multi-head attention operation on the input feature sequence to obtain an output intermediate feature sequence; the intermediate feature sequence is transposed and then input into the second multi-head attention module for multi-head attention operation, and the output result of the second multi-head attention module is transposed to obtain the output feature sequence of the multi-head attention layer. Among them, the first multi-head attention module and the second multi-head attention module perform feature extraction on the feature sequence of the input multi-head attention layer in the time dimension and feature dimension respectively. The structure of the Transformer encoder, such as Figure 4 shown.

[0072] In summary, the present invention uses a time sliding window of adaptive size to intercept data sequences. This adaptive method can dynamically adjust the window size according to the operating data characteristics and volatility of the optical terminal, so as to more accurately capture the time series features in the data and improve the accuracy of the prediction. This improvement enables the model to adapt to different data scenarios more flexibly, and improves the practicality and generalization ability of the prediction. By combining the LSTM network and the TCN network, the time series features of the optical terminal equipment are extracted from both the global and local levels. The LSTM network is responsible for capturing long-term dependencies, while the TCN network focuses on the extraction of local features. This combination method not only improves the model's processing ability for complex time series data, but also improves the accuracy of fault prediction. External environmental factors are incorporated into the fault prediction model, and the external environment status monitoring data is feature extracted through the TCN network to obtain the external factor feature sequence. This improvement enables the model to understand the operating status of the equipment more comprehensively, and further improves the accuracy and robustness of the prediction. The present invention introduces a Transformer encoder based on feature dimension attention and time dimension attention. This improved Transformer model can better capture the potential information and time series relationships in multi-dimensional time series data, and automatically assign weights to different time steps and feature dimensions through the attention mechanism, thereby further improving the accuracy and robustness of the prediction.

[0073] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A method for predicting optical terminal equipment failure based on deep learning, characterized in that: include: S1: Receive and pre-process the historical operating performance index data and historical environmental status monitoring data of the optical terminal; S2: Obtain the to-be-tested operating performance indicator data sequence and the to-be-tested environmental status monitoring data sequence from the historical operating performance indicator data and the historical environmental status monitoring data of the optical terminal through a time sliding window of adaptive size; S3: Input the data sequence of the operating performance indicator to be tested into the LSTM model and the first TCN model respectively for feature extraction to obtain a global feature sequence and a local feature sequence; S4: Input the monitoring data sequence of the environmental state to be tested into the second TCN model for feature extraction to obtain the external factor feature sequence; S5: using a preset weighting factor to weight the global feature sequence, the local feature sequence and the external factor feature sequence to obtain a fused feature sequence; S6: Input the fused feature sequence into the Transformer encoder based on feature dimension attention and time dimension attention, and output the prediction result.

2. According to the method for predicting optical terminal equipment failure based on deep learning according to claim 1, it is characterized in that: The operating performance indicator data of the optical terminal include: one or more of input optical power, output optical power, device temperature, board temperature, CPU utilization and memory utilization, among which each operating performance indicator data includes three forms, namely, the instantaneous value at the current moment, and the maximum and minimum values ​​within the collection period; the environmental status monitoring data of the optical terminal include: one or more environmental indicators of ambient temperature, ambient humidity, ambient pressure and ambient wind speed.

3. According to the method for predicting optical terminal equipment failure based on deep learning according to claim 2, it is characterized in that: The method of obtaining the to-be-tested operation performance indicator data sequence and the to-be-tested environment status monitoring data sequence through the time sliding window of adaptive size comprises: S21: Construct an adaptive adjustment factor based on the historical operating performance index data of the optical terminal: S22: Adaptively adjust the size of the time sliding window according to the adaptive adjustment factor: W(d)=W(d-1)*α(d) Among them, α(d) represents the adaptive adjustment factor at time d, W(d-1) represents the size of the time sliding window at time d-1; W(d) represents the size of the time sliding window at time d; S23: using a time sliding window of adaptive size, respectively extracting the data closest to the current time t from the historical operation performance indicator data and the historical environmental status monitoring data as the operation performance indicator data sequence to be tested and the environmental status monitoring data sequence to be tested.

4. According to the method for predicting optical terminal equipment failure based on deep learning in claim 3, it is characterized in that: The adaptive adjustment factor includes: α(d)=max(α min ,1+tanh(ΔP(d)-Δp threshold )) Where N represents the number of operating performance indicators, ΔP(d) represents the performance fluctuation amplitude of the optical terminal at time d; e represents the natural constant, α min represents the lowest threshold of the regulation factor, Δp threshold represents the mean of all performance fluctuations from the first time point to time d-1, x i ′(d) and x′(d-1) represent the normalized values ​​of the i-th performance indicator at time d and d-1, respectively; w i is the weight of the i-th performance indicator, Var(x i ) represents the i-th performance index x i The variance in the historical operating performance indicator data; Indicates the i-th performance indicator x in the historical running performance indicator data i The average value of; T represents the time length of the historical running performance indicator data.

5. According to the method for predicting optical terminal equipment failure based on deep learning in claim 3, it is characterized in that: The data sequence of the operating performance indicator to be tested is input into the LSTM model for feature extraction, and the global feature sequence obtained includes: The data sequence of the operating performance indicator to be tested is expressed as in, W(t) represents the number of time steps, that is, the width of the time sliding window; r represents the feature dimension of the data; the running performance indicator data sequence is input into the LSTM model to generate the hidden state h of each time step i ; The hidden states generated by the LSTM model at all time steps constitute the global feature sequence F G ={h t-W(t) ,h t-W(t)+1 ,...,h j ,...,h t }, h j represents the jth feature in the global feature sequence, t represents the current moment, W(t) Indicates the size of the sliding window at the current time t.

6. According to the method for predicting optical terminal equipment failure based on deep learning in claim 3, it is characterized in that: The first TCN model and the second TCN model both include multiple residual blocks, which are multiple residual blocks connected in series.

7. The method for predicting optical terminal equipment failure based on deep learning according to claim 1 is characterized in that: The fusion signature sequence comprises: F 内部 =αF LSTM +βF TCN F Fusion =γF 内部 +δF 外部 Among them, F Fusion represents the fusion feature sequence, F TCN represents the local feature sequence, F LSTM represents the global feature sequence, α, β, γ and δ represent weighting factors, and α+β=11, γ+δ=1.

8. According to the method of predicting optical terminal equipment failure based on deep learning in claim 1, it is characterized in that: The Transformer encoder based on feature dimension attention and time dimension attention includes: multiple Transformer encoding layers, and the multi-head attention layer of each Transformer encoding layer includes: a first multi-head attention module and a second multi-head attention module; the first multi-head attention module performs a multi-head attention operation on the input feature sequence to obtain an output intermediate feature sequence; the intermediate feature sequence is transposed and then input into the second multi-head attention module for multi-head attention operation, and the output result of the second multi-head attention module is transposed to obtain the output feature sequence of the multi-head attention layer.

Images