Method and device for predicting poor quality of ONU (Optical Network Unit) and electronic equipment

Abstract

The embodiment of the invention provides an ONU poor quality prediction method and device and electronic equipment, and the method comprises the steps: collecting feature data which comprises the performance data and alarm data of an ONU; preprocessing the feature data to form a feature matrix; wherein the vector length of the performance data in the feature matrix is the same as the vector length of the alarm data; inputting the feature matrix into a deep learning model to obtain the poor quality probability of the ONU; wherein the deep learning model is used for predicting the poor quality probability of the ONU.

Description

Technical Field

[0001] The present application relates to the field of optical communication technology, and in particular to a method, device and electronic device for predicting quality difference of an optical network unit (ONU). Background Art

[0002] With the improvement of communication technology and the rapid development of network technology, home network has spread to most families. Facing the huge user group of home network, it is particularly important to ensure high-quality Internet access.

[0003] Failures in the Passive Optical Network (PON) access network often lead to a large number of ONU loss of optical signal (LOSi) alarms and ONU bit error and packet loss problems, affecting users' broadband service experience.

[0004] Existing ONU fault identification technology aims to identify faults after they occur to improve the efficiency of manual fault diagnosis. It cannot predict the quality difference of ONUs, and the maintenance cost of ONUs is high and the efficiency is low. Summary of the invention

[0005] The purpose of the embodiments of the present application is to provide a method, device and electronic device for predicting the quality difference of an ONU, which can solve the problem that the quality difference prediction of the ONU cannot be achieved.

[0006] To solve the above technical problems, the embodiments of the present application are implemented through the following aspects.

[0007] In a first aspect, an embodiment of the present application provides a method for predicting poor quality of an ONU, comprising: collecting feature data, the feature data comprising performance data and alarm data of the ONU; preprocessing the feature data to form a feature matrix; wherein the vector length of the performance data in the feature matrix is ​​the same as the vector length of the alarm data; inputting the feature matrix into a deep learning model to obtain the poor quality probability of the ONU; wherein the deep learning model is used to predict the poor quality probability of the ONU.

[0008] In a second aspect, an embodiment of the present application provides an ONU quality difference prediction device, comprising: an acquisition module, used to collect feature data, the feature data including performance data and alarm data of the ONU; a processing module, used to preprocess the feature data to form a feature matrix; wherein the vector length of the performance data in the feature matrix is ​​the same as the vector length of the alarm data; a prediction module, used to input the feature matrix into a deep learning model to obtain the quality difference probability of the ONU; wherein the deep learning model is used to predict the quality difference probability of the ONU.

[0009] In a third aspect, an embodiment of the present application provides an electronic device comprising: a memory, a processor, and computer executable instructions stored in the memory and executable on the processor, wherein the computer executable instructions, when executed by the processor, implement the steps of executing the method described in the first aspect.

[0010] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, which is used to store computer-executable instructions. When the computer-executable instructions are executed by a processor, the steps of the method described in the first aspect are implemented.

[0011] In the embodiment of the present application, the collected ONU performance data and alarm data are pre-processed to form a feature matrix, and then the poor quality probability of the ONU is predicted through a deep learning model, so that the poor quality ONU can be identified with a high accuracy, so that the rectification is completed through active operation and maintenance, and the Internet experience of users in the whole network is improved. The maintenance cost of the ONU is low and the efficiency is high; at the same time, the vector length of the performance data in the feature matrix is ​​the same as the vector length of the alarm data, so that the feature matrix is ​​more in line with the input of the deep learning model, which is conducive to improving the accuracy of the poor quality probability of the ONU predicted by the deep learning model. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0013] Figure 1 A schematic flow chart of a method for predicting quality difference of an ONU provided in an embodiment of the present application is shown;

[0014] Figure 2 A schematic diagram of a home network structure of an ONU quality difference prediction method provided in an embodiment of the present application;

[0015] Figure 3 A schematic diagram of a device for data collection and ONU quality difference output according to an embodiment of the present application;

[0016] Figure 4 A schematic diagram of a data processing process of an ONU quality difference prediction method provided in an embodiment of the present application;

[0017] Figure 5 A schematic diagram of data processing results of the ONU quality difference prediction method provided in an embodiment of the present application;

[0018] Figure 6A specific application diagram of the ONU quality difference prediction method provided in an embodiment of the present application;

[0019] Figure 7 A schematic diagram showing the structure of an ONU quality difference prediction device provided in an embodiment of the present application is shown;

[0020] Figure 8 A schematic diagram of the hardware structure of an electronic device for executing the ONU quality difference prediction method provided in an embodiment of the present application. DETAILED DESCRIPTION

[0021] In order to enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this application.

[0022] Figure 1 A flow chart of a method for predicting the quality difference of an ONU provided in an embodiment of the present application is shown. The method can be executed by an electronic device. In other words, the method can be executed by software or hardware installed in the electronic device. Figure 1 As shown, the method may include the following steps.

[0023] S10: Collect characteristic data, where the characteristic data includes performance data and alarm data of the ONU.

[0024] In this embodiment, the performance data can be used to characterize at least one of the following of the ONU: flow rate, bandwidth utilization, transmit and receive optical power, attenuation, temperature, voltage, current, positioning frame loss, etc.; the alarm data can be used to characterize at least one of the following of the ONU: signal loss, signal weakening, many optical path errors, power off, power failure, disconnection, etc.

[0025] Usually, the possibility of ONU failure is hidden in the medium- and long-term performance data and alarm data. For example, if the temperature rises abnormally for a long time, the possibility of ONU failure is higher. This embodiment can learn the relationship between ONU performance data, alarm data and failure through a deep learning model, predict the possibility of ONU failure and its probability through the performance data and alarm data of ONU in the medium and long term, and actively and intelligently identify the quality problems that may exist in ONU, so as to improve the active operation and maintenance capabilities.

[0026] S20: Preprocessing the characteristic data to form a characteristic matrix; wherein the vector length of the performance data in the characteristic matrix is ​​the same as the vector length of the alarm data.

[0027] Optionally, collecting feature data in S10 includes: collecting the performance data of x cycles, wherein the performance data is collected t times in each cycle; collecting the alarm data of x cycles, where x and t are positive integers; preprocessing the feature data in S20 includes: processing the collected performance data into a feature matrix; encoding each collected alarm data into a vector, and based on the vector length of the performance data in the feature matrix of the performance data, copying the vector of the alarm data so that the vector length of the performance data is the same as the vector length of the alarm data.

[0028] This embodiment uses the above-mentioned alarm data vector replication operation to make the vector length of the performance data the same as the vector length of the alarm data, which can retain the useful information in the alarm data and performance data as much as possible. At the same time, irregular alarm information is added to the feature matrix to make it consistent with the input of the deep learning model, which is beneficial to improving the accuracy of the ONU quality difference probability predicted by the deep learning model.

[0029] Optionally, each vector of the alarm data includes at least one of the following: the number of occurrences of the alarm data; the alarm duration, etc.

[0030] S30: Input the feature matrix into a deep learning model to obtain the poor quality probability of the ONU; wherein the deep learning model is used to predict the poor quality probability of the ONU.

[0031] In the embodiment of the present application, an ONU that is more likely to fail in the future is referred to as a poor-quality ONU.

[0032] Optionally, after this step inputs the feature matrix into a deep learning model to obtain the poor quality probability of the ONU, the method further includes: when the poor quality probability of the ONU is greater than a preset judgment threshold, outputting prompt information, wherein the prompt information is used to indicate the poor quality of the ONU and / or the poor quality probability of the ONU.

[0033] This embodiment can not only identify poor-quality ONUs, but also predict the probability of ONU failure. Subsequently, ONUs with a higher probability of failure can be pre-processed to prevent ONU failures, thereby ensuring the Internet quality of users.

[0034] The method for predicting poor quality of an ONU provided in an embodiment of the present application pre-processes the collected ONU performance data and alarm data to form a feature matrix, and then predicts the poor quality probability of the ONU through a deep learning model, so that poor quality ONUs can be identified with a high accuracy rate, thereby actively completing rectification through operation and maintenance, improving the Internet experience of users across the entire network, and the maintenance cost of the ONU is low and the efficiency is high; at the same time, the vector length of the performance data in the feature matrix is ​​the same as the vector length of the alarm data, so that the feature matrix is ​​more in line with the input of the deep learning model, which is beneficial to improving the accuracy of the poor quality probability of the ONU predicted by the deep learning model.

[0035] Optionally, the characteristic data collected in this embodiment may be characteristic data within a longer period, such as data for N days, data for M weeks, etc. When the ONU has a poor quality problem, abnormal temperature rise may continue to occur, or the receiving / emitting power may continue to decrease. This trend will be reflected in the performance data of a longer period. This embodiment collects characteristic data within a longer period, and these characteristic data are more closely related to the poor quality of the ONU, so as to improve the accuracy of the poor quality probability of the ONU predicted by the deep learning model.

[0036] In addition, this embodiment uses more comprehensive feature data, and the possibility of ONU failure will also be reflected in more performance data, including flow rate, bandwidth utilization, transmit and receive optical power, attenuation, optical module temperature, etc. These feature data are more closely related to the ONU quality difference, which facilitates improving the accuracy of the ONU quality difference probability predicted by the deep learning model.

[0037] The ONU quality difference prediction method provided in the embodiment of the present application can at least achieve the following effects:

[0038] (1) High prediction accuracy: Practice has shown that it can accurately identify poor-quality ONUs when both the recall rate and accuracy are greater than 80%.

[0039] (2) The method is highly adaptable: Through in-depth analysis of performance data and alarm data over a period of time, a deep learning model for fault analysis is constructed using an AI algorithm. The model automatically learns the relationship between ONU performance, alarms, and faults in the medium and long term, reducing the trial and error of manual feature extraction.

[0040] Optionally, in each of the above embodiments, inputting the feature matrix into a deep learning model to obtain the poor quality probability of the ONU includes the following steps:

[0041] 1) Inputting the feature matrix into the input layer of the deep learning model, wherein the input layer is composed of a residual structure consisting of a one-dimensional causal dilated convolution.

[0042] 2) The feature matrix processed by the input layer passes through an intermediate layer, and the intermediate layer is used to encode, integrate and adjust the shape of the feature matrix.

[0043] 3) The feature matrix processed by the intermediate layer passes through an output layer, and the output layer is used to output the quality difference probability of the ONU.

[0044] This embodiment can use one-dimensional causal dilation convolution as the basic unit and design the feature extraction network of the model in combination with the residual structure. The one-dimensional causal dilation convolution only retains the connection from the front to the back in the convolution by blocking, and gradually expands the dilation coefficient in the hidden layer after the input, thereby increasing the perceptual range of the output features without increasing the number of network layers. The residual structure adds a pass-through layer to the network, and the input of the convolution is added to its output. What is learned is the residual of the output of the previous part of the network, rather than the entire output of the previous network, which reduces the information loss caused by convolution and improves the performance of the deep learning model.

[0045] Optionally, before collecting feature data in each of the above embodiments, the method further includes: obtaining the deep learning model through training based on the following loss function:

[0046] loss=-((1-ceil(y_true))*log(y0_pred)+y_true*log(y1_pred))

[0047] Among them, loss represents the loss function, ceil represents rounding up, y_true represents the quality difference label, and the quality difference label includes severe quality difference, moderate quality difference and no quality difference, y0_pred represents the probability that the ONU is predicted to have no quality difference, and y1_pred represents the probability that the ONU is predicted to have poor quality.

[0048] The deep learning model obtained by training based on the loss function includes: based on the loss function, performing cross entropy loss calculation on the label of the ONU and the output result of the deep learning model in training, thereby measuring the gap between the model prediction result and the actual result (label); using the stochastic gradient descent method to back-propagate the cross entropy loss to correct the parameters of the deep learning model in training, and obtaining the trained deep learning model after multiple iterations.

[0049] In this embodiment, the labels of the ONU include severe quality difference, moderate quality difference and no quality difference; similarly, Figure 1 In the illustrated embodiment, the poor quality probability of the ONU predicted by the deep learning model may also include severe poor quality, moderate poor quality, or no poor quality.

[0050] To explain in detail the ONU quality difference prediction method provided in the embodiment of the present application, the following will be described in combination with specific application scenarios and specific embodiments.

[0051] The embodiments of the present application can be applied in home network user products, mainly to assist operators in finding poor quality ONUs, so the application products are mainly home network products of operators. The technical background of the embodiments of the present application is to search for poor quality ONUs through performance data and alarm data, so it is necessary to search for fault scenarios, especially the fault may involve two aspects of data, on the one hand, the data arrives regularly, and on the other hand, the data is random and irregular, and the embodiments of the present application can be applied, such as faulty equipment identification, faulty OTN identification, etc.

[0052] The embodiments of the present application can be used for fixed-line home broadband services for users, mainly targeting the passive optical network (PON) network structure. PON is (in the optical distribution network) that does not contain any electronic devices or electronic power supplies; a passive optical network includes an optical line terminal (OLT) installed in a central control station, and a group of supporting optical network units (ONUs) installed in user locations. The optical distribution network (ODN) between the OLT and the ONU contains optical fibers and passive splitters or couplers. Figure 2 As shown, the method proposed in the embodiment of the present application mainly uses the performance data and alarm data collected by the OLT to determine the probability of poor quality of the ONU.

[0053] OLT: usually includes control board, DC power board, fan unit, service frame, and other upstream boards. One side of OLT is connected to the switch at the coverage level through a cable and converted into an optical signal; on the other side, it is connected to the optical splitter on the customer terminal through an optical fiber. It can be used to control, manage, and measure the distance to customer terminal devices such as ONU. OLT is one of the devices on the service node of the access network.

[0054] ODU: Optical Distribution Network. ODN is a FTTH (Fiber to the Home) optical cable network based on PON equipment. Its function is to provide an optical transmission channel between OLT and ONU. From the functional point of view, ODN can be divided into four parts from the central office to the user end: feeder cable subsystem, distribution cable subsystem, drop cable subsystem and optical fiber terminal subsystem.

[0055] ONU: Optical Network Unit, ONU is divided into active optical network unit and passive optical network unit. Generally, the equipment equipped with optical receiver, upstream optical transmitter, multiple bridge amplifier network monitoring is called optical node. ONU provides data, IPTV (interactive network television), voice (using IAD, Integrated Access Device) and other services. Common home ONU is generally called optical modem.

[0056] Terminal: Internet access device, such as mobile phones, computers, etc.

[0057] The method provided in the embodiment of the present application can be used for Figure 2 In the OLT module where the DPI blade is located, the DPI blade is built into the OLT to monitor Internet applications and perceive the user's real business traffic, so that it can be used to collect feature data.

[0058] Figure 3 It is a specific quality difference judgment device, and the entire architecture is briefly described below.

[0059] The original data collection device includes a performance data collection device and an alarm data collection device, which is usually deployed in the operator's bearer equipment. The equipment will mirror the data passing through the bearer equipment, obtain the user's IP packet header information, and then calculate information including upstream and downstream peak flow rates, upstream and downstream average flow rates, etc.

[0060] Data storage devices are mainly used to store massive amounts of network indicator information. Currently, open source distributed file systems such as FastDFS, Ceph, and HDFS are used. The file system is integrated through a self-developed platform to complete the persistence of massive data. At present, the data traffic of operators' bearer equipment can reach hundreds of Gb per second. It is difficult and unnecessary to achieve real-time analysis. A timed collection strategy can be used, such as a granularity of 5 minutes or hours.

[0061] The data query device needs to use the computing power of multiple cluster servers to query and calculate massive amounts of data. This is done through self-developed query modules and the integration of distributed computing frameworks such as Hadoop 3.0 and Spark. The distributed framework uses the computing power of multiple clusters to quickly complete data query and calculation. At the same time, it divides the business with high traffic into multiple cycles and uses a multi-task timing scheduling algorithm to output the results.

[0062] The poor quality judgment device is the core content of the embodiment of this application. Through the self-developed software module, the original network indicators are queried, and the poor quality ONUs of the entire network are obtained by analyzing the indicator data, and the reasons for the poor quality and rectification suggestions are given. The final result is presented to the operator in the form of a report, and the ONU rectification can be completed according to the report, thereby improving the overall network quality and enhancing the user's Internet experience.

[0063] The ONU quality difference prediction technology proposed in the embodiment of the present application is divided into two stages, the training stage and the verification stage. The training stage mainly updates the parameters of the deep learning model through data; the verification stage directly outputs the probability of ONU quality difference after deployment. The two stages share the same data processing steps and the same deep learning model. Next, the data processing steps and deep learning models are first explained, and then the training stage, transfer learning stage and verification stage are explained.

[0064] Data processing, such as Figures 3 to 5 shown.

[0065] Each ONU sends performance data t times a day, and generates t*x performance data in a period of x days. The number of alarms and the alarm codes involved for each ONU are random. For the performance and alarms of an ONU, the processing steps are as follows:

[0066] Step 1: Keep the key indicators of the performance data and process the performance data of each ONU into matrices of the same shape.

[0067] Step 2: Anomalies and missing values ​​of performance data are filled in the same way.

[0068] Step 3: Collect the alarm data of ONU within the period x days (if any). Due to the wide variety of alarms, only the more frequently occurring and important alarm codes are retained for encoding. Encode the alarm of each ONU into a vector, including important information such as the number of occurrences of the alarm code and the duration of the alarm code.

[0069] Step 4: Repeat the warning vector t*x times to expand the dimension, which is the same length as the performance data, so that it can be easily fed into the deep learning model at the same time. This method retains useful information as much as possible, reduces the size of the data, has a regular arrangement of features, and also adds irregular warning information to the matrix to make it suitable for the input of the deep learning model.

[0070] The core invention of step 3 and step 4 is to process irregular alarm data into a regular matrix, which is suitable for the input of deep learning models and has certain substitutability, such as changing the order of dimensions or the encoding method of alarm codes.

[0071] Deep Learning Models

[0072] The deep learning model integrates the alarm matrix (step 4 of data processing) and the performance matrix (step 2 of data processing) and outputs the poor quality user (i.e., ONU). This method enables the model to perceive comprehensive information about the ONU, which can enhance the accuracy and generalization of the model. The feature extraction network of the model is designed with one-dimensional causal dilation convolution as the basic unit and combined with the residual structure. The one-dimensional causal dilation convolution only retains the connection from the front to the back in the convolution by blocking, and gradually expands the dilation coefficient in the hidden layer after the input, thereby increasing the perception range of the output feature without increasing the number of network layers. The residual structure adds a pass-through layer to the network, and the input of the convolution is added to its output. What is learned is the residual of the output of the previous part of the network, rather than the entire output of the previous network, which reduces the information loss caused by the convolution and improves the performance of the model.

[0073] Step 1: Convolutional coding, perform three module convolutions on the input fusion data matrix, each module consists of a residual structure composed of three layers of one-dimensional causal dilated convolutions.

[0074] Step 2: Further encode and integrate the convolutionally encoded features through a 2-layer neural network.

[0075] Step 3: Reshape the features through convolution, flattening, and fully connected layers (processing operations for deep learning) to make them suitable for the output layer.

[0076] Step 4: Through the softmax output layer, the result of step 3 is converted into poor quality probability, and the poor quality probability range is 0 to 1.

[0077] All steps of the deep learning model involve network layers. The core invention is to use neural networks for automatic feature extraction, which has certain substitutability, such as using neural networks with more or fewer layers, or other types of neural networks to extract features, such as LSTM, RNN or transformer.

[0078] Training steps

[0079] The training step can be performed after accumulating data for a period of time on site and obtaining the training data set; or the model at other sites can be migrated to the current site. The amount of data required for the latter can be appropriately reduced, and the migrated model can be used as the model initialization parameter for subsequent training. The former requires a large amount of data and can be trained directly.

[0080] Step 1: Dataset preparation,Combined with the fault severity during on-site maintenance, the faults are subdivided into severe quality problems and moderate quality problems, and different labels are assigned, that is, the labels include severe quality problems, moderate quality problems, and no quality problems.

[0081] The designed loss function is:

[0082] loss=-((1-ceil(y_true))*log(y0_pred)+y_true*log(y1_pred))

[0083] Among them, loss represents the loss function, ceil represents rounding up, y_true represents the quality difference label, and the quality difference label includes severe quality difference, moderate quality difference and no quality difference, y0_pred represents the probability that the ONU is predicted to be no quality difference, and y1_pred represents the probability that the ONU is predicted to be poor quality. In the actual experiment, the fault-free label y_true is 0, the severe quality difference label y_true is 1, and the moderate quality difference label y_true is 0.8.

[0084] Step 2: Training. The cross entropy loss is applied to the output results and labels of the deep learning model, and the stochastic gradient descent (SGD) method is used for back propagation to correct the deep model parameters. After multiple iterations, the required model is finally obtained.

[0085] The core invention of the dataset preparation in the training step is that the sample labels are not absolute 0 and 1, but have certain substitutability, for example, a combination of 0.3-0.7-0.9 can be selected, etc.; or a more fine-grained classification of quality difference can be performed, such as subdividing it into mild quality difference, moderate quality difference and severe quality difference and selecting corresponding sample labels.

[0086] Verification steps such as Figure 6 shown.

[0087] Step 1: Process the performance data and alarm data of the ONU.

[0088] Step 2: Load the deep learning model in the training phase.

[0089] Step 3: Input the data processed in step 1 into the deep learning model in step 2, and finally obtain the poor quality probability of each ONU.

[0090] Step 4: Set the judgment threshold. When the probability that the network classifies the ONU as poor quality is greater than the threshold, the ONU is judged as a poor quality ONU by the AI ​​model, and the poor quality ONU and its poor quality probability are output.

[0091] Embodiments of the present application The quality difference prediction method of ONU provided in the above embodiments of the present application is as follows:

[0092] 1) In terms of data processing, the performance data of the ONU in the medium and long term and the alarm data in the irregular corresponding time period. The ONU performance data is reported at the same time interval, while the type, quantity, and arrival time of the alarms involved in each ONU are irregular. The embodiment of the present application innovatively encodes the alarms and integrates them with the ONU performance data to adapt to the input of the deep learning model.

[0093] 2) In terms of deep learning models, the present invention innovatively designs deep learning models based on the characteristics of performance data and alarm data, classifies whether the ONU is of poor quality based on the input data, and predicts its failure probability. The one-dimensional causal dilation convolution is used as the basic unit, and the feature extraction network is designed in combination with the residual structure; the extracted features are used to calculate whether the ONU is of poor quality and the failure probability at the output layer.

[0094] 3) In terms of model training, the embodiment of the present application combines the results of on-site maintenance and innovatively divides the faults into severe quality problems and moderate quality problems according to the degree of the faults on site, assigns different labels to faults of different degrees, and designs a corresponding loss function to train the deep learning model.

[0095] The ONU quality difference prediction method provided in the above embodiments of the present application can at least achieve the following effects:

[0096] (1) High prediction accuracy: Practice has shown that it can accurately identify poor-quality ONUs when both the recall rate and accuracy are greater than 80%.

[0097] (2) The method is highly adaptable: Through in-depth analysis of performance data and alarm data over a period of time, a deep learning model for fault analysis is constructed using an AI algorithm. The model automatically learns the relationship between ONU performance, alarms, and faults in the medium and long term, reducing the trial and error of manual feature extraction.

[0098] (3) Fast calculation: In this embodiment, the AI ​​algorithm of the deep learning model does not need to be trained separately for each object. The trained deep learning model will be adapted to all objects. Therefore, for the client, it only needs to wait for one inference time. The algorithm proposed in this embodiment requires fewer layers, so the calculation is faster.

[0099] Figure 7 A schematic diagram of the structure of an ONU quality difference prediction device provided in an embodiment of the present application is shown, and the device 700 includes the following modules.

[0100] The collection module 702 is used to collect characteristic data, where the characteristic data includes performance data and alarm data of the ONU.

[0101] The processing module 704 is used to pre-process the characteristic data to form a characteristic matrix; wherein the vector length of the performance data in the characteristic matrix is ​​the same as the vector length of the alarm data.

[0102] The prediction module 706 is used to input the feature matrix into a deep learning model to obtain the poor quality probability of the ONU; wherein the deep learning model is used to predict the poor quality probability of the ONU.

[0103] In the embodiment of the present application, the collected ONU performance data and alarm data are pre-processed to form a feature matrix, and then the poor quality probability of the ONU is predicted through a deep learning model, so that the poor quality ONU can be identified with a high accuracy, so that the rectification is completed through active operation and maintenance, and the Internet experience of users in the whole network is improved. The maintenance cost of the ONU is low and the efficiency is high; at the same time, the vector length of the performance data in the feature matrix is ​​the same as the vector length of the alarm data, so that the feature matrix is ​​more in line with the input of the deep learning model, which is conducive to improving the accuracy of the poor quality probability of the ONU predicted by the deep learning model.

[0104] Optionally, as an embodiment, the collection module 702 is used to collect the performance data of x cycles, wherein the performance data is collected t times in each cycle; collect the alarm data of x cycles, x, t are positive integers; the processing module 704 is used to process the collected performance data into a feature matrix; encode each collected alarm data into a vector, and based on the vector length of the performance data in the feature matrix of the performance data, copy the vector of the alarm data so that the vector length of the performance data is the same as the vector length of the alarm data.

[0105] Optionally, as an embodiment, each vector of the alarm data includes at least one of the following: the number of occurrences of the alarm data; and the alarm duration.

[0106] Optionally, as an embodiment, the prediction module 706 is used to input the feature matrix into the input layer of the deep learning model, and the input layer is composed of a residual structure composed of a one-dimensional causal dilation convolution; the feature matrix after processing by the input layer passes through an intermediate layer, and the intermediate layer is used to encode, integrate and adjust the shape of the feature matrix; the feature matrix after processing by the intermediate layer passes through an output layer, and the output layer is used to output the quality difference probability of the ONU.

[0107] Optionally, as an embodiment, a model training module is further included, which is used to train the deep learning model based on the following loss function:

[0108] loss=-((1-ceil(y_true))*log(y0_pred)+y_true*log(y1_pred))

[0109] Among them, loss represents the loss function, ceil represents rounding up, y_true represents the quality difference label, and the quality difference label includes severe quality difference, moderate quality difference and no quality difference, y0_pred represents the probability that the ONU is predicted to have no quality difference, and y1_pred represents the probability that the ONU is predicted to have poor quality.

[0110] Optionally, as an embodiment, the model training module is used to calculate the cross entropy loss on the label of the ONU and the output result of the deep learning model in training based on the loss function; use the stochastic gradient descent method to back-propagate the cross entropy loss to correct the parameters of the deep learning model in training, and obtain the trained deep learning model after multiple iterations.

[0111] Optionally, as an embodiment, the processing module 704 is further used to output prompt information when the poor quality probability of the ONU is greater than a preset determination threshold, wherein the prompt information is used to indicate the poor quality of the ONU and / or the poor quality probability of the ONU.

[0112] Optionally, as an embodiment, the performance data is used to characterize at least one of the following of the ONU: flow rate, bandwidth utilization, transmit and receive optical power, attenuation, temperature, voltage, current, and positioning frame loss; the alarm data is used to characterize at least one of the following of the ONU: signal loss, signal weakening, many optical path bit errors, power off, power failure, and disconnection.

[0113] The device 700 provided in the embodiment of the present application can execute the various methods described in the foregoing method embodiments, and realize the functions and beneficial effects of the various methods described in the foregoing method embodiments, which will not be repeated here.

[0114] Figure 8 A schematic diagram of the hardware structure of an electronic device that implements the embodiment of the present application is shown. Referring to the figure, at the hardware level, the electronic device includes a processor, and optionally, an internal bus, a network interface, and a memory. Among them, the memory may include a memory, such as a high-speed random access memory (Random-Access Memory, RAM), and may also include a non-volatile memory (non-volatile memory), such as at least one disk storage, etc. Of course, the electronic device may also include hardware required for other services.

[0115] The processor, the network interface and the memory may be interconnected through an internal bus, which may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one bidirectional arrow is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0116] The memory is used to store the program. Specifically, the program may include a program code, and the program code includes a computer operation instruction. The memory may include a memory and a non-volatile memory, and provides instructions and data to the processor.

[0117] The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it, forming a device for locating the target user at the logical level. The processor executes the program stored in the memory and is specifically used to execute: Figure 1-6 The method disclosed in the illustrated embodiment realizes the functions and beneficial effects of each method described in the foregoing method embodiments, which will not be described in detail here.

[0118] The above application Figure 1-6 The method disclosed in the illustrated embodiment can be applied to a processor or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in the processor or an instruction in the form of software. The above processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiments of the present application can be directly embodied as a hardware decoding processor for execution, or a combination of hardware and software modules in the decoding processor for execution. The software module can be located in a storage medium mature in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0119] The electronic device can also execute the methods described in the foregoing method embodiments, and realize the functions and beneficial effects of the methods described in the foregoing method embodiments, which will not be repeated here.

[0120] Of course, in addition to software implementation methods, the electronic device of the present application does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc. That is to say, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0121] The embodiment of the present application also provides a computer-readable storage medium, wherein the computer-readable medium stores one or more programs, and when the one or more programs are executed by an electronic device including multiple application programs, the electronic device executes Figure 1-6 The method disclosed in the illustrated embodiment realizes the functions and beneficial effects of each method described in the foregoing method embodiments, which will not be described in detail here.

[0122] The computer-readable storage medium includes a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0123] Furthermore, an embodiment of the present application also provides a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, and when the program instructions are executed by a computer, the following process is implemented: Figure 1-6 The method disclosed in the illustrated embodiment realizes the functions and beneficial effects of each method described in the foregoing method embodiments, which will not be described in detail here.

[0124] In short, the above description is only a preferred embodiment of the present application and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0125] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0126] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0127] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0128] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

Claims

1. A method for predicting quality difference of an optical network unit (ONU), comprising: Collecting characteristic data, wherein the characteristic data includes performance data and alarm data of the ONU; Preprocessing the characteristic data to form a characteristic matrix; wherein the vector length of the performance data in the characteristic matrix is ​​the same as the vector length of the alarm data; The feature matrix is ​​input into a deep learning model to obtain the poor quality probability of the ONU; wherein the deep learning model is used to predict the poor quality probability of the ONU.

2. The method according to claim 1, wherein: The collecting of characteristic data includes: collecting the performance data of x cycles, wherein the performance data is collected t times in each cycle; collecting the alarm data of x cycles, where x and t are positive integers; The preprocessing of the characteristic data includes: processing the collected performance data into a characteristic matrix; encoding each collected alarm data into a vector, and based on the vector length of the performance data in the characteristic matrix of the performance data, copying the vector of the alarm data so that the vector length of the performance data is the same as the vector length of the alarm data.

3. The method according to claim 2, wherein: Each vector of the alarm data includes at least one of the following: the number of occurrences of the alarm data; and the alarm duration.

4. The method according to any one of claims 1 to 3, wherein: The step of inputting the feature matrix into a deep learning model to obtain the poor quality probability of the ONU includes: Inputting the feature matrix into an input layer of the deep learning model, wherein the input layer is composed of a residual structure consisting of a one-dimensional causal dilated convolution; The feature matrix processed by the input layer passes through an intermediate layer, and the intermediate layer is used to perform encoding integration and shape adjustment on the feature matrix; The feature matrix processed by the intermediate layer passes through the output layer, and the output layer is used to output the quality difference probability of the ONU.

5. The method according to claim 1, wherein: Before collecting the feature data, the method further includes: obtaining the deep learning model through training based on the following loss function: loss=-((1-ceil(y_true))*log(y0_pred)+y_true*log(y1_pred)) Among them, loss represents the loss function, ceil represents rounding up, y_true represents the quality difference label, and the quality difference label includes severe quality difference, moderate quality difference and no quality difference, y0_pred represents the probability that the ONU is predicted to have no quality difference, and y1_pred represents the probability that the ONU is predicted to have poor quality.

6. The method according to claim 5, wherein: The deep learning model obtained by training based on the loss function includes: Based on the loss function, the cross entropy loss is calculated for the label of the ONU and the output result of the deep learning model in training; the cross entropy loss is back-propagated using the stochastic gradient descent method to correct the parameters of the deep learning model in training, and the trained deep learning model is obtained after multiple iterations.

7. The method according to claim 1, wherein: After inputting the feature matrix into a deep learning model to obtain the poor quality probability of the ONU, the method further includes: When the probability of poor quality of the ONU is greater than a preset determination threshold, a prompt message is output, where the prompt message is used to indicate the poor quality of the ONU and / or the probability of poor quality of the ONU.

8. The method according to claim 1, wherein: The performance data is used to characterize at least one of the following of the ONU: flow rate, bandwidth utilization, transmit and receive optical power, attenuation, temperature, voltage, current, and positioning frame loss; The alarm data is used to characterize at least one of the following of the ONU: signal loss, signal weakening, many optical path bit errors, power off, power failure, and line loss.

9. An ONU quality difference prediction device, comprising: A collection module, used for collecting characteristic data, wherein the characteristic data includes performance data and alarm data of the ONU; A processing module, used for preprocessing the characteristic data to form a characteristic matrix; wherein the vector length of the performance data in the characteristic matrix is ​​the same as the vector length of the alarm data; A prediction module is used to input the feature matrix into a deep learning model to obtain the poor quality probability of the ONU; wherein the deep learning model is used to predict the poor quality probability of the ONU.

10. An electronic device comprising: processor; as well as A memory arranged to store computer executable instructions which, when executed, cause the processor to perform the steps of the method of any one of claims 1 to 8.

11. A computer-readable medium storing one or more programs, which, when executed by an electronic device including a plurality of application programs, causes the electronic device to execute the steps of the method according to any one of claims 1 to 8.

Images