Model training method and processing method of optical line terminal equipment

Abstract

The invention discloses a model training method and a processing method of optical line terminal equipment. The method comprises the following steps: acquiring historical operation data of optical line terminal equipment; performing data cleaning processing on the historical operation data, and performing sign extraction processing and labeling processing on the processed historical operation data to obtain a target data set; using the target data set to carry out iterative training on a plurality of classifiers, after training of each classifier is completed, adjusting the classifier participating in training and a weight coefficient of a training sample according to a classification error rate, and using the adjusted sample to train a next classifier; and combining the classifier generated by iterative training and the weight coefficient thereof into a target classifier, the target classifier being used for determining a cutover scheme corresponding to the optical line terminal device. The technical problem of resource waste caused by the fact that the operation energy consumption of the optical line terminal equipment cannot be efficiently reduced is solved.

Description

Technical Field

[0001] This application relates to the field of network technology and security. Specifically, it relates to a model training method and a processing method for an optical line terminal device. Background Art

[0002] With the continuous progress of technology and the rapid development of communication technology, optical line terminal (OLT) devices play a crucial role in communication networks. As a part of urban construction and an advocate for low-carbon development in the telecommunications industry.

[0003] However, with the rapid expansion of the OLT device deployment scale, the problem of network energy consumption has become increasingly prominent, bringing huge economic pressure to enterprises and operators. How to reduce the energy consumption of devices without affecting the user's Internet experience has always been a major problem in the industry.

[0004] In response to the above problems, no effective solution has been proposed yet. Summary of the Invention

[0005] This application provides a model training method and a processing method for an optical line terminal device to at least solve the technical problem of resource waste caused by the inability to efficiently reduce the operating energy consumption of the optical line terminal device.

[0006] According to one aspect of this application, a model training method is provided, including: obtaining historical operation data of an optical line terminal device, where the historical operation data includes: the number and location of wireless optical network interfaces of optical line terminal devices corresponding to different suppliers, device types, the number of retired port boards, the number of in-use ports, the number of idle ports, the number of in-use boards, the number of idle boards, the number of in-use frames, the number of idle frames, and the type of the computer room where the optical line terminal device is located; performing data cleaning processing on the historical operation data, and performing feature extraction processing and annotation processing on the processed historical operation data to obtain a target data set; using the target data set to iteratively train multiple classifiers. After completing the training of each classifier, according to the classification error rate, adjust the weights of the classifiers and training samples participating in the training, and use the adjusted samples to train the next classifier; combine the classifiers and their weight coefficients generated by the iterative training into a target classifier, where the target classifier is used to determine the cutover plan corresponding to the optical line terminal device.

[0007] Optionally, multiple classifiers are iteratively trained using a target dataset. After the training of each classifier is completed, according to the classification error rate, the weights of the classifiers and training samples participating in the training are adjusted, and the adjusted samples are used to train the next classifier, including: initializing the classifier weight coefficients, iteratively training the classifier using the target dataset until the preset number of classifiers is reached. Among them, the training process for each round is as follows: generate random numbers, where the number of random numbers is equal to the number of samples in the target dataset; sort the random numbers and determine the index sequence of the positions after sorting relative to the positions before sorting; reorder the target dataset according to the index sequence to obtain the first dataset, and in the first dataset, select a preset proportion of the training set and test set; initialize the weights of each training sample in the training set to obtain the target training set, where the weights of each training sample in the target training set are all equal and the sum of all weights is 1; train the classifier using the target training set, and when the preset stop condition is met, obtain the target classifier and save the target classifier to the classifier array; use the target classifier to process the test set to obtain the prediction result vector output by the target classifier; update the weight coefficients of the target classifier and the weight coefficients of the training samples according to the error rate between the prediction result vector and the true label.

[0008] Optionally, after combining the classifiers and their weight coefficients generated by the iterative training into the target classifier, the method further includes: when the classification error rate of the target classifier is greater than the preset threshold, compare the classification error rate of the target classifier with that of the classifiers generated by the iterative training in turn. If the classification error rate of the target classifier is greater than the classification error rate of the classifier being compared, adjust the model parameters of the target classifier according to the model parameters of the classifier being compared until the classification error rate of the target classifier is less than the preset threshold.

[0009] Optionally, adjusting the model parameters of the target classifier includes: finding the parameter combination that minimizes the classification error rate of the target classifier within a preset numerical range through cross-validation and grid search methods.

[0010] Optionally, performing physical sign extraction processing on the processed historical operation data includes: for the number of boards, extracting the number of different types of boards and the port density of the boards as features; for the type of the computer room where the optical line terminal device is located, extracting the scale information, power supply capacity information, and refrigeration capacity information of the computer room as features.

[0011] Optionally, performing data cleaning processing on the historical operation data includes: removing the data related to the Tianwang data and school data in the historical operation data.

[0012] According to another aspect of the present application, there is also provided a processing method for an optical line terminal device, including: obtaining the operation data of the optical line terminal device; inputting the operation data into a target classifier to obtain a cutover plan corresponding to the optical line terminal device output by the target classifier, where the target classifier is obtained by training through the above-mentioned model training method.

[0013] According to another aspect of the present application, there is also provided a model training device, including: an obtaining module, configured to obtain the historical operation data of the optical line terminal device, where the historical operation data includes: the number and location of the wireless optical network interfaces of the optical line terminal devices corresponding to different suppliers, device types, the number of port boards taken out of service, the number of in-use ports, the number of idle ports, the number of in-use boards, the number of idle boards, the number of in-use frames, the number of idle frames, and the type of the computer room where the optical line terminal device is located; a processing module, configured to perform data cleaning processing on the historical operation data, and perform feature extraction processing and annotation processing on the processed historical operation data to obtain a target data set; a training module, configured to perform iterative training on multiple classifiers using the target data set, and after completing the training of each classifier, adjust the weights of the classifiers and training samples participating in the training according to the classification error rate, and use the adjusted samples to train the next classifier; a combining module, configured to combine the classifiers generated by iterative training and their weight coefficients into a target classifier, where the target classifier is used to determine the cutover plan corresponding to the optical line terminal device.

[0014] According to another aspect of the present application, there is also provided a non-volatile storage medium, where the storage medium includes a stored program, and when the program runs, it controls the device where the storage medium is located to execute the above-mentioned model training method.

[0015] According to another aspect of the present application, there is also provided an electronic device, including: a memory and a processor, where the processor is configured to run the program stored in the memory, and when the program runs, it executes the above-mentioned model training method.

[0016] According to another aspect of the present application, there is also provided a computer program, where when the computer program is executed by a processor, it implements the above-mentioned model training method.

[0017] According to another aspect of the present application, there is also provided a computer program product, where the computer program product includes a non-volatile computer-readable storage medium, and the non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the above-mentioned model training method.

[0018] In this application, historical operation data of an optical line terminal device is obtained. The historical operation data includes: the number and location of wireless optical network interfaces of optical line terminal devices corresponding to different suppliers, device types, the number of retired port boards, the number of in-use ports, the number of idle ports, the number of in-use boards, the number of idle boards, the number of in-use frames, the number of idle frames, and the type of the computer room where the optical line terminal device is located. The historical operation data is subjected to data cleaning processing, and the processed historical operation data is subjected to feature extraction processing and annotation processing to obtain a target data set. The target data set is used to iteratively train multiple classifiers. After the training of each classifier is completed, according to the classification error rate, the weights of the classifiers and training samples participating in the training are adjusted, and the adjusted samples are used to train the next classifier. The classifiers generated by iterative training and their weight coefficients are combined into a target classifier. The target classifier is used to determine the cutover scheme corresponding to the optical line terminal device, achieving the purpose of efficiently reducing the operation energy consumption of the optical line terminal device, thereby realizing the technical effect of avoiding resource waste, and further solving the technical problem of resource waste caused by the inability to efficiently reduce the operation energy consumption of the optical line terminal device. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings described herein are used to provide a further understanding of the present application and form a part of the present application. The illustrative embodiments and descriptions thereof of the present application are used to explain the present application and do not constitute an improper limitation to the present application. In the drawings:

[0020] Figure 1 is a flowchart of a model training method according to an embodiment of the present application;

[0021] Figure 2 is a flowchart of an adaptive support vector machine algorithm according to an embodiment of the present application;

[0022] Figure 3 is a structural diagram of a model training device according to an embodiment of the present application;

[0023] Figure 4 is a hardware structure block diagram of a computer terminal of a model training method according to an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0024] In order to enable those skilled in the art to better understand the solution of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0025] It should be noted that the terms "first", "second", etc. in the description, claims and the above-mentioned drawings of the present application are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present application described here can be implemented in an order other than those illustrated or described here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0026] According to an embodiment of the present application, a method embodiment of a model training method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. And although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0027] Figure 1 is a flowchart of a model training method according to an embodiment of the present application, as Figure 1 shown, the method includes the following steps:

[0028] Step S102, obtaining historical operation data of an optical line terminal device, where the historical operation data includes: the number and location of wireless optical network interfaces of optical line terminal devices corresponding to different suppliers, device types, the number of port boards taken out of service, the number of in-use ports, the number of idle ports, the number of in-use boards, the number of idle boards, the number of in-use frames, the number of idle frames, and the type of the computer room where the optical line terminal device is located.

[0029] In step S102, it is first necessary to ensure that the detailed historical operation data of the optical line terminal equipment can be accessed and collected. The following data can be obtained in the network operation and maintenance department: the number and location of the wireless optical network PON interfaces of the OLT equipment corresponding to different suppliers: record the total number of PON ports on each supplier's equipment and the specific board location. Equipment type: record the specific model of the OLT equipment to facilitate subsequent equipment classification and performance analysis. The number of port boards withdrawn from the network: count the number of port boards that are no longer used in the past period of time. The number of ports in use: count the number of PON ports currently in use, which is used to characterize the active users and business load of the equipment. The number of idle ports: count the number of unused PON ports, which is used to evaluate the potential capacity and optimization possibilities of the equipment. The number of boards in use: record the number of boards currently in use, which is used to characterize the overall load and resource allocation of the equipment. The number of idle boards: count the number of unused boards, which is used to identify resource redundancy. The number of frames in use: record the number of frames currently in use, which is used to evaluate the physical space usage of the equipment. The number of idle frames: count the number of unused frames, which is used to plan the physical layout and heat dissipation requirements of the equipment. Type of room where the optical line terminal equipment is located: Collect information about the room where the equipment is located, including its size, location, and environmental conditions, to assess the potential impact on equipment operation.

[0030] Step S104, performing data cleaning processing on the historical operation data, and performing vital sign extraction processing and labeling processing on the processed historical operation data to obtain a target data set.

[0031] The collected historical operation data is cleaned to remove incomplete, erroneous or duplicate data to ensure the accuracy and reliability of subsequent analysis. After data cleaning, the processed data is subjected to feature extraction and annotation, including: 1. Labeling: Add a label to each piece of data to identify the type of computer room it belongs to, the status of the equipment (such as whether it is off the network), etc., to facilitate the identification and understanding of the meaning of the data during model training. 2. Feature processing: Standardize the digital features in the data and encode the classification features to ensure that all features can be input into the machine learning model in a consistent manner.

[0032] Step S106, iteratively train multiple classifiers using the target data set. After completing the training of each classifier, adjust the weight coefficients of the classifiers and training samples involved in the training according to the classification error rate, and use the adjusted samples to train the next classifier.

[0033] In step S106, using the cleaned and labeled target dataset, multiple SVM classifiers are iteratively trained through the Adaboost algorithm. In each iteration, according to the classification error rate of the previous round of classifiers, the weights of the samples in the dataset are adjusted so that the misclassified samples receive more attention in the next round of training. After completing the training of each classifier, the weight coefficients of the participating classifiers and their training samples are adjusted by calculating the classification error rate to ensure that the model can gradually improve the classification accuracy of the data.

[0034] Step S108, combine the classifiers generated by iterative training and their weight coefficients into a target classifier, where the target classifier is used to determine the cutover plan corresponding to the optical line terminal device.

[0035] After completing the iterative training of all classifiers, these classifiers and their corresponding weight coefficients are combined to generate the final target classifier. This target classifier can predict and evaluate the optimal cutover plan of the device based on the input OLT device operation data.

[0036] Among them, the cutover plan corresponding to the optical line terminal device aims to optimize network resources, improve device efficiency, reduce energy consumption, and adapt to network changes. These plans can be very specific. The following uses several examples to illustrate the possible contents of common OLT device cutover plans:

[0037] 1. Background: There are multiple boards on a certain OLT device, but not all PON ports are fully utilized. Some ports are idle, while others are close to full load. Cutover plan: Use an intelligent analysis model to predict the traffic demand and user activities of each PON port, and migrate the users on the ports with lower traffic to the idle ports or ports with lower load to achieve reallocation of resources and reduce device energy consumption.

[0038] 2. Background: Multiple OLT devices are deployed in a certain area, but with the slowdown of user growth, the resource utilization rate of some devices is low. Cutover plan: Identify and shut down the OLT devices with low resource utilization rate, migrate the affected users to other devices, and perform frame combination on the chassis at the same time.

[0039] 3. Background: During the business peak period, the PON ports of some OLT devices are overloaded, while there is resource waste during the low peak period. Cutover plan: According to historical data and business forecasts, dynamically adjust the operating status of the device. For example, during the low peak period, migrate some users to other OLT devices to achieve dynamic balance of resources.

[0040] 4. Background: Operators are facing the dual pressures of energy conservation and emission reduction and reducing operation costs. Cutover solution: Considering equipment utilization rate, energy consumption, and maintenance costs comprehensively, formulate energy conservation and emission reduction solutions, such as migrating users to equipment with higher energy efficiency or reducing equipment energy consumption through software optimization.

[0041] According to the above steps, obtain the historical operation data of the optical line terminal equipment, where the historical operation data includes: the number and location of the wireless optical network interfaces of the optical line terminal equipment corresponding to different suppliers, equipment type, the number of retired port boards, the number of in-use ports, the number of idle ports, the number of in-use boards, the number of idle boards, the number of in-use frames, the number of idle frames, and the type of the computer room where the optical line terminal equipment is located; perform data cleaning processing on the historical operation data, and perform feature extraction processing and annotation processing on the processed historical operation data to obtain a target data set; use the target data set to iteratively train multiple classifiers. After completing the training of each classifier, according to the classification error rate, adjust the weights of the classifiers and training samples participating in the training, and use the adjusted samples to train the next classifier; combine the classifiers generated by iterative training and their weight coefficients into a target classifier, where the target classifier is used to determine the cutover solution corresponding to the optical line terminal equipment, achieving the purpose of efficiently reducing the operation energy consumption of the optical line terminal equipment, thereby realizing the technical effect of avoiding resource waste.

[0042] The following makes an exemplary description and explanation of Figure 1 the steps shown.

[0043] According to some alternative embodiments of the present application, multiple classifiers are iteratively trained using a target data set. After the training of each classifier is completed, according to the classification error rate, the weights of the classifiers and the training samples participating in the training are adjusted, and the adjusted samples are used to train the next classifier. This can be achieved through the following method: Initialize the classifier weight coefficients, and iteratively train the classifier using the target data set until the preset number of classifiers is reached. Among them, the training process for each round is as follows: Generate random numbers, where the number of random numbers is equal to the number of samples in the target data set; Sort the random numbers and determine the index sequence of the positions after sorting relative to the positions before sorting; According to the index sequence, reorder the target data set to obtain the first data set, and in the first data set, select a preset proportion of the training set and the test set; Initialize the weights of each training sample in the training set to obtain the target training set, where the weights of each training sample in the target training set are all equal, and the sum of all weights is 1; Use the target training set to train the classifier. When the preset stop condition is met, obtain the target classifier and save the target classifier to the classifier array; Use the target classifier to process the test set to obtain the prediction result vector output by the target classifier; According to the error rate between the prediction result vector and the true label, update the weight coefficient of the target classifier and update the weight coefficient of the training sample.

[0044] Specifically, determine the number of classifiers trained using the Adaboost algorithm, for example, set it to 10 classifiers. Assign initial weight coefficients to each classifier. Initially, the weights of all classifiers can be set to be equal. For example, if the number of classifiers is 10, the initial weight of each classifier is 1 / 10.

[0045] For each round of training (assuming a total of 10 rounds of training, with each round corresponding to the generation of a classifier), perform the following steps: Generate a sequence of random numbers equal in quantity to the number of samples in the target dataset. For example, if the target dataset has 100 samples, then generate 100 random numbers. Use Python's sorting method to sort the generated random numbers, obtaining the sorted numbers and their indices in the original sequence. This index sequence will be used for subsequent reordering of the dataset. Reorder the target dataset according to the sorted index sequence to obtain a new first dataset. Randomly select a portion from the first dataset as the training set (e.g., at a ratio of 1:5, select 20% of the data as the training set), and the rest as the test set. This ensures that different data subsets are used for each training, enhancing the generalization ability of the model. Set the initial weights for each sample in the training set. For example, if the training set has 20 samples, then the initial weight for each sample is 1 / 20, ensuring that the sum of the weights of all samples is 1. Use the initialized training set to train the classifier. After training is complete, use the test set to test the classifier, obtaining the predicted result vector output by the classifier and comparing it with the true labels of the test set to calculate the classification error rate. Adjust the weight coefficient of the current classifier according to the classification error rate. If the error rate of the classifier is low, its weight will increase; if the error rate is high, the weight will decrease. This step ensures that classifiers that perform better in subsequent iterations will have greater influence. Similarly, update the weights of the training set samples according to the difference between the predicted result vector of the classifier and the true labels. Samples that are misclassified will be given higher weights, causing subsequent training to focus more on these difficult samples and improving the overall performance of the model.

[0046] After each round of training, save the obtained target classifier into the classifier array until all classifiers are trained. Through multiple rounds of training, the weight coefficients of the classifiers in the classifier array will be continuously optimized to reflect their importance in the overall model.

[0047] Finally, an ensemble classifier model composed of multiple classifiers is obtained, where the weight coefficient of each classifier reflects its performance and importance in the model. This model can be used to more accurately predict and process the cutover plan of optical line terminal devices, improve the utilization efficiency of network resources, and reduce device energy consumption at the same time.

[0048] Preferably, after combining the classifier generated by iterative training and its weight coefficients into a target classifier, the following steps may also be performed: When the classification error rate of the target classifier is greater than a preset threshold, the classifier generated by iterative training and the target classifier are successively compared based on the classification error rate. If the classification error rate of the target classifier is greater than that of the classifier being compared, the model parameters of the target classifier are adjusted according to the model parameters of the classifier being compared until the classification error rate of the target classifier is less than the preset threshold.

[0049] Specifically, determine a preset classification error rate threshold (e.g., 5%). Set the initial model parameters of the target classifier, which may include the kernel function type of SVM, regularization parameter C, gamma parameter, etc.

[0050] Use the target classifier to predict a preset test data set, and calculate its classification error rate. This step identifies the number of misclassified samples by comparing the prediction results with the true labels of the test set and calculates the error rate. Check whether the classification error rate of the target classifier is greater than the preset threshold. If it is greater, proceed to the next step; otherwise, the classifier parameters meet the requirements, and the process can be stopped and the current target classifier can be used.

[0051] If the error rate of the target classifier is greater than the preset threshold, select a classifier from the array of classifiers generated during iterative training for comparison. This classifier should have shown better performance in previous training rounds, and its classification error rate is lower than that of the current target classifier. Analyze the model parameters of the comparison classifier to determine which parameters contribute to its better classification performance. It could be the choice of kernel function, the magnitude of the regularization parameter, gamma value, etc.

[0052] According to these parameters of the comparison classifier, correspondingly adjust the model parameters of the target classifier. After adjusting the parameters, use the target classifier to predict the test data set again and calculate the new classification error rate. If the new error rate is still greater than the preset threshold, continue the loop, select the next comparison classifier and adjust the parameters; if the error rate is less than the preset threshold, stop the loop, and the parameter settings of the current target classifier are valid.

[0053] Once the classification error rate of the target classifier is lower than the preset threshold, the model parameter adjustment process terminates. At this time, the target classifier will be used as the final model for predicting and optimizing the cutover plan of the optical line terminal device.

[0054] In the above process, the key point is to use the classifier with better performance during iterative training as a guide to adjust the parameters of the target classifier in order to obtain a more optimal model. The parameter adjustment can be performed manually or optimized through automated algorithms (such as grid search, random search, etc.) to find the best parameter combination.

[0055] Preferably, adjusting the model parameters of the target classifier can be achieved by the following method: searching for a parameter combination that minimizes the classification error rate of the target classifier within a preset value range through cross-validation and grid search methods.

[0056] Specifically, define the range of adjustable parameters in the target classifier. For example, for a support vector machine, the parameters may include the regularization parameter C, the kernel function parameter gamma, etc. Determine the candidate value range for each parameter. For example, the candidate value of C may be [0.1, 1, 10, 100], and the candidate value of gamma may be [0.001, 0.01, 0.1, 1]. Then, create a parameter grid, which will contain a list of all parameter combinations.

[0057] Divide the target dataset into K equal-sized subsets, which is the basis of cross-validation. For example, if K=5, the dataset will be divided into 5 subsets.

[0058] For each parameter combination in the parameter grid, perform the following steps: Select one subset as the validation set and the remaining subsets as the training set. Use the current parameter combination to train the target classifier on the training set. Apply the trained classifier to the validation set and calculate the classification error rate on the validation set. Record the average classification error rate of each parameter combination on all K validation sets.

[0059] Among all parameter combinations, the one with the lowest average classification error rate is selected as the optimal solution. This parameter combination will be used to train the final model to obtain the best classification performance.

[0060] The optimal parameter combination found in cross-validation is used to train the entire target dataset to generate the final target classifier. This classifier should have a lower classification error rate on unseen data and improve the generalization ability of the model.

[0061] Through the above steps, combined with the robustness evaluation of cross-validation and the parameter optimization of grid search, it can be ensured that the target classifier finds the optimal parameter combination within the given parameter range.

[0062] Preferably, the processed historical operation data is subjected to vital sign extraction processing, including: for the number of boards, extracting the number of different types of boards and the port density of the boards as features; for the type of computer room where the optical line terminal equipment is located, extracting the scale information, power supply capacity information and cooling capacity information of the computer room as features.

[0063] As some optional embodiments of the present application, data cleaning processing is performed on the historical operation data, including: eliminating data related to Skynet data and school data in the historical operation data.

[0064] Another model training method is also provided in an embodiment of this application, which specifically includes the following steps:

[0065] Step S1, data collection: First, the operation data of the OLT device needs to be collected. This data includes the type of the computer room where it is located, the device type, the number of boards, the number of ports (including: the number of retired ports on the port board, the number of in-use ports, the number of idle ports, etc.), the number of boards (including the number of in-use boards, the number of idle boards, etc.), and the number of frames (including the number of in-use frames, the number of idle frames, etc.).

[0066] Step S101, collect the device type of the OLT device, and the devices can be classified according to different manufacturers, such as Huawei, ZTE, etc. Put the collected device types into the database to obtain the OLT device data set.

[0067] Step S102, for the OLT devices of different manufacturers, collect the type of the computer room where they are located, the number of boards, the number of retired ports on the port board, the number of in-use ports, the number of idle ports, the number of in-use boards, the number of idle boards, the number of in-use frames, and the number of idle frames respectively, and put them into the database.

[0068] Step S2, data preprocessing: Preprocess the collected data, including data cleaning, data conversion, data standardization, etc., so that the data meets the input requirements of the deep learning model.

[0069] Step S201, due to special reasons, first clean out the Tianwang data and school data, and then convert the remaining data. Label the computer room information, device information, port information, etc. extracted in step S1. The processed data is used as the sample set for feature extraction.

[0070] Step S202, establish the OLT device quantity set P = {P a1 , P a2 , …, P st , …, P nN}, where P st is the t-th board of the s-th manufacturer, n is the total number of manufacturers, and N is the total number of OLT device boards.

[0071] Step S203, establish the OLT device frame quantity set Q = {Q 1 , Q 2 , …, Q t , …, Q n}, where Q t represents the number of frames corresponding to t.

[0072] Step S204, establish the OLT device board quantity set T = {T a1 , T a2 , …, T st , …, T nN}, where T st represents the number of PON ports of the P-th st board card.

[0073] Step S205: Correlate OLT devices from different manufacturers, computer room types, the number of board cards, the number of retired port board cards, the number of in-use ports, the number of idle ports, the number of in-use board cards, the number of idle board cards, the number of in-use chassis, and the number of idle chassis to obtain a unified data set. Remove null values and duplicate values to obtain a preprocessed data set D = {(x(1), y(1)), …, (x(|D|), y(|D|))}.

[0074] Step S31: Model construction and training. Figure 2 is a flowchart of an adaptive support vector machine algorithm according to an embodiment of the present application. As Figure 2 shown, the algorithm specifically includes the following steps.

[0075] Step S301: According to the data set D obtained in step S2, during each round of iteration of this data set, the training set randomly selected and generated is T, where the k-th round is T k , then the training set T k can be expressed as:

[0076] T k =(x 1 , y 1 ), (x 2 , y 2 ), …, (x n , y n ), x i ∈X, y i ∈Y

[0077] where X and Y are the value ranges of the data attribute x and the data class label y respectively.

[0078] Step S302: Set the number of SVM classifiers to be trained by the Adaboost boosting method to 10.

[0079] Step S303: Generate a set of random numbers. Since random extraction is to be performed, it is required that the number of this set of random numbers is the same as the number of samples in the data set. Then use the generated random numbers as parameters and sort them using the built-in sort method in python. Specifically, first sort the original data and return the index sequence of the sorted numbers corresponding to the numbers before sorting. Then, substitute this index sequence into the data set to obtain a shuffled data set. Next, select a part from this shuffled data set as the final training set for training the SVM classifier in this loop; the other part is used as the final test set.

[0080] Step S304: Initialize the weights of each sample in the final training set so that these weights are all equal and their sum is equal to 1.

[0081] Step S305: Determine whether the current number of loops has exceeded the number of SVM classifiers to be trained. If not, jump to Step S306; if so, jump to Step S313.

[0082] Step S306: Randomly divide the final test set into an AdaBoost training set and an AdaBoost test set in a ratio of 1:5. Then, use the current AdaBoost training set to train an SVM classifier. Finally, save the trained SVM classifier into the array of trained SVM classifiers for subsequent use.

[0083] Step S307: Use the SVM classifier trained in Step S306 to test the AdaBoost test set and obtain the result vector. Match the vector to be classified with the class label vector of the AdaBoost test set to obtain the prediction result of this classification. If a sample is correctly classified, the corresponding element in the test result vector is 0; if a sample is misclassified, the corresponding element in the test result vector is 1. In the above way, a test result comparison vector composed of 0s and 1s can be obtained, so as to know which samples are correctly classified and which samples are misclassified.

[0084] Step S308: Evaluate the performance of the classifier by calculating the training error of this classification. Then, calculate the weight of the classifier for this classification prediction according to the training error and the prediction result. Finally, save the calculated weight into the array of classifier weights for subsequent use.

[0085] Step S309: Since the weight of the classifier has been obtained, calculate the new weights of the correctly classified and misclassified samples according to the formula.

[0086] Step S310: Update the calculated new weights into the sample weight array.

[0087] Step S311: Standardize the sample weight array so that the sum of all the weights in it is equal to 1.

[0088] Step S312: Jump to Step S305.

[0089] Step S313: Training is completed, and an array of classifiers containing several SVM classifiers is obtained.

[0090] Step S4, model prediction and planning optimization. After the model is trained, the model can be used to obtain the optimal cutover plan for the OLT device, and the prediction result can be used for the planning of the PON port location.

[0091] Step S401, use the first one of a series of SVM classifiers trained in the first stage as a single SVM classifier to compare with the integrated classifier.

[0092] Step S402, repeatedly execute Step S401 until the overall model error reaches within the set range.

[0093] Step S403, finally obtain the optimization and energy-saving evaluation model of the access network device, and obtain the optimal cutover plan for the OLT device.

[0094] Through the above steps, using the improved deep learning algorithm to optimize the PON port location, computer room type, and number of boards of the device, optimize resource allocation, and dynamically adjust the device status, the energy consumption of the OLT device can be effectively reduced. In addition, by combining various data features of the access network device, such as the type of computer room, device type, number of boards, ports, boards, and frames, and through the deep learning algorithm, the PON port location is reasonably planned, the device status is better identified, and the device utilization rate is improved.

[0095] Figure 3 It is a structural diagram of a model training device according to an embodiment of the present application, as Figure 3 shown. The device includes:

[0096] An acquisition module 31, configured to acquire historical operation data of an optical line terminal device, where the historical operation data includes: the number and location of wireless optical network interfaces of optical line terminal devices corresponding to different suppliers, device type, number of port boards taken out of service, number of in-use ports, number of idle ports, number of in-use boards, number of idle boards, number of in-use frames, number of idle frames, and type of computer room where the optical line terminal device is located.

[0097] A processing module 32, configured to perform data cleaning processing on the historical operation data, and perform feature extraction processing and annotation processing on the processed historical operation data to obtain a target data set.

[0098] A training module 33, configured to perform iterative training on multiple classifiers using the target data set. After the training of each classifier is completed, according to the classification error rate, adjust the weights of the classifiers and training samples participating in the training, and use the adjusted samples to train the next classifier.

[0099] The combined module 34 is used to combine the classifiers and their weight coefficients generated by iterative training into a target classifier, where the target classifier is used to determine the cutover plan corresponding to the optical line terminal device.

[0100] Optionally, multiple classifiers are iteratively trained using the target dataset. After the training of each classifier is completed, according to the classification error rate, the weight coefficients of the classifiers and training samples participating in the training are adjusted, and the adjusted samples are used to train the next classifier, including: initializing the classifier weight coefficients, iteratively training the classifier using the target dataset until the preset number of classifiers is reached, where the training process for each round is as follows: generating random numbers, where the number of random numbers is equal to the number of samples in the target dataset; sorting the random numbers and determining the index sequence of the positions after sorting relative to the positions before sorting; reordering the target dataset according to the index sequence to obtain the first dataset, and selecting a preset proportion of the training set and test set from the first dataset; initializing the weights of each training sample in the training set to obtain the target training set, where the weights of each training sample in the target training set are all equal and the sum of all weights is 1; training the classifier using the target training set, and obtaining the target classifier when the preset stop condition is met, and saving the target classifier to the classifier array; processing the test set using the target classifier to obtain the predicted result vector output by the target classifier; updating the weight coefficients of the target classifier and the weight coefficients of the training samples according to the error rate between the predicted result vector and the true label.

[0101] Optionally, after combining the classifiers and their weight coefficients generated by iterative training into a target classifier, the following steps can also be performed: when the classification error rate of the target classifier is greater than the preset threshold, compare the target classifier with the classifiers generated by iterative training in turn based on the classification error rate. If the classification error rate of the target classifier is greater than the classification error rate of the classifier being compared, adjust the model parameters of the target classifier according to the model parameters of the classifier being compared until the classification error rate of the target classifier is less than the preset threshold.

[0102] Optionally, adjusting the model parameters of the target classifier includes: finding the parameter combination that minimizes the classification error rate of the target classifier within the preset numerical range through cross-validation and grid search methods.

[0103] Optionally, performing feature extraction processing on the processed historical operation data includes: for the number of boards, extracting the number of different types of boards and the port density of the boards as features; for the type of the computer room where the optical line terminal device is located, extracting the scale information, power supply capacity information, and refrigeration capacity information of the computer room as features.

[0104] Optionally, perform data cleaning on the historical operation data, including: removing the data related to the sky network data and school data from the historical operation data.

[0105] It should be noted that the above Figure 3 Each of the modules can be a program module (for example, a set of program instructions for implementing a specific function), or a hardware module. For the latter, it can be presented in the following forms, but not limited to: the manifestation form of each of the above modules is a processor, or the functions of each of the above modules are implemented by a processor.

[0106] It should be noted that Figure 3 For the preferred implementation manners of the illustrated embodiments, reference can be made to Figure 1 the relevant descriptions of the illustrated embodiments, which will not be elaborated herein.

[0107] Figure 4 The hardware structure block diagram of a computer terminal for implementing the model training method is shown. As Figure 4 shown, the computer terminal 40 may include one or more (shown as 402a, 402b,..., 402n in the figure) processors 402 (the processor 402 may include, but is not limited to, a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 404 for storing data, and a transmission module 406 for communication functions. In addition, it may further include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which can be included as one of the ports of the BUS bus), a network interface, a power supply, and / or a camera. Those of ordinary skill in the art can understand that Figure 4 the structure shown is only schematic and does not limit the structure of the above electronic device. For example, the computer terminal 40 may further include more or fewer components than those Figure 4 shown, or have a different configuration from that Figure 4 shown.

[0108] It should be noted that the above one or more processors 402 and / or other data processing circuits can generally be referred to as "data processing circuits" herein. The data processing circuit can be embodied in whole or in part as software, hardware, firmware, or any arbitrary combination thereof. In addition, the data processing circuit can be a single independent processing module, or be incorporated in whole or in part into any one of the other elements in the computer terminal 40. As involved in the embodiments of the present application, the data processing circuit is a kind of processor control (such as the selection of a variable resistor terminal path connected to an interface).

[0109] The memory 404 can be used to store software programs and modules of application software, such as the program instruction / data storage device corresponding to the model training method in the embodiments of the present application. The processor 402 executes various functional applications and data processing by running the software programs and modules stored in the memory 404, that is, implements the above-mentioned model training method. The memory 404 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memories. In some instances, the memory 404 may further include a memory remotely disposed relative to the processor 402, and these remote memories can be connected to the computer terminal 40 through a network. Examples of the above network include but are not limited to the Internet, intranet, local area network, mobile communication network, and combinations thereof.

[0110] The transmission module 406 is used to receive or send data via a network. Specific examples of the above network may include a wireless network provided by a communication provider of the computer terminal 40. In one instance, the transmission module 406 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one instance, the transmission module 406 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0111] The display can be, for example, a touch-screen liquid crystal display (LCD), and the liquid crystal display enables a user to interact with the user interface of the computer terminal 40.

[0112] It should be noted here that in some alternative embodiments, the above Figure 4 shown computer terminal may include hardware elements (including circuits), software elements (including computer code stored on a computer-readable medium), or a combination of both hardware elements and software elements. It should be pointed out that Figure 4 is only an example of a specific specific instance, and is intended to show the types of components that may exist in the above computer terminal.

[0113] It should be noted that Figure 4 the shown computer terminal is used to execute Figure 1 the shown model training method, so the relevant explanations in the execution method of the above commands are also applicable to this electronic device, which will not be elaborated here.

[0114] The embodiments of the present application also provide a non-volatile storage medium. The non-volatile storage medium includes a stored program, wherein when the program runs, it controls the device where the storage medium is located to execute the above model training method.

[0115] A program for a non-volatile storage medium to perform the following functions: obtaining historical operation data of an optical line terminal device, where the historical operation data includes: the number and location of wireless optical network interfaces of optical line terminal devices corresponding to different vendors, device types, the number of port cards taken out of service, the number of in-use ports, the number of idle ports, the number of in-use boards, the number of idle boards, the number of in-use frames, the number of idle frames, and the type of the computer room where the optical line terminal device is located; performing data cleaning processing on the historical operation data, and performing feature extraction processing and annotation processing on the processed historical operation data to obtain a target data set; using the target data set to perform iterative training on multiple classifiers, and after completing the training of each classifier, adjusting the weights of the classifiers and training samples participating in the training according to the classification error rate, and using the adjusted samples to train the next classifier; combining the classifiers generated by the iterative training and their weight coefficients into a target classifier, where the target classifier is used to determine the cutover plan corresponding to the optical line terminal device.

[0116] An embodiment of the present application also provides an electronic device, including: a memory and a processor, where the processor is used to run a program stored in the memory, and when the program runs, it executes the above model training method.

[0117] The processor is used to run a program that performs the following functions: obtaining historical operation data of an optical line terminal device, where the historical operation data includes: the number and location of wireless optical network interfaces of optical line terminal devices corresponding to different vendors, device types, the number of port cards taken out of service, the number of in-use ports, the number of idle ports, the number of in-use boards, the number of idle boards, the number of in-use frames, the number of idle frames, and the type of the computer room where the optical line terminal device is located; performing data cleaning processing on the historical operation data, and performing feature extraction processing and annotation processing on the processed historical operation data to obtain a target data set; using the target data set to perform iterative training on multiple classifiers, and after completing the training of each classifier, adjusting the weights of the classifiers and training samples participating in the training according to the classification error rate, and using the adjusted samples to train the next classifier; combining the classifiers generated by the iterative training and their weight coefficients into a target classifier, where the target classifier is used to determine the cutover plan corresponding to the optical line terminal device.

[0118] The serial numbers of the above embodiments of the present application are only for description and do not represent the advantages and disadvantages of the embodiments.

[0119] In the above embodiments of the present application, the descriptions of the various embodiments have their own focuses. For parts not detailed in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0120] In the above embodiments of the present application, the collected information is information and data authorized by the user or fully authorized by all parties. Moreover, for the processing of relevant data such as collection, storage, use, processing, transmission, provision, disclosure, and application, all comply with relevant laws, regulations, and standards, necessary protection measures are taken, it does not violate public order and good customs, and corresponding operation entrances are provided for users to choose to authorize or reject.

[0121] In several embodiments provided by the present application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the described device embodiments are merely illustrative. For example, the division of the units can be a logical function division. In actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling, direct coupling, or communication connection to each other can be through some interfaces. The indirect coupling or communication connection of units or modules can be in an electrical or other form.

[0122] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0123] In addition, in each embodiment of the present application, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0124] If the above-mentioned integrated units are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the relevant technology, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present application. And the aforementioned storage medium includes: USB flash drives, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), mobile hard disks, magnetic disks, or optical discs and other various media that can store program codes.

[0125] The above are only the preferred embodiments of the present application. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present application, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present application.

Claims

1. A model training method, characterized in that: include: Obtaining historical operation data of the optical line terminal equipment, wherein the historical operation data includes: the number and location of the wireless optical network interfaces of the optical line terminal equipment corresponding to different suppliers, the type of equipment, the number of port boards withdrawn from the network, the number of ports in use, the number of idle ports, the number of boards in use, the number of idle boards, the number of frames in use, the number of idle frames, and the type of the computer room where the optical line terminal equipment is located; Performing data cleaning processing on the historical operation data, and performing vital sign extraction processing and labeling processing on the processed historical operation data to obtain a target data set; Iteratively train multiple classifiers using the target data set, and after completing the training of each classifier, adjust the weight coefficients of the classifiers involved in the training and the training samples according to the classification error rate, and use the adjusted samples to train the next classifier; The classifiers generated by iterative training and their weight coefficients are combined into a target classifier, wherein the target classifier is used to determine a cutover solution corresponding to the optical line terminal device.

2. The method according to claim 1, characterized in that Iteratively training multiple classifiers using the target data set, adjusting the weight coefficients of the classifiers involved in the training and the training samples according to the classification error rate after completing the training of each classifier, and using the adjusted samples to train the next classifier, including: Initialize the classifier weight coefficient, and iteratively train the classifier using the target data set until the preset number of classifiers is reached, wherein each round of training process is as follows: Generate random numbers, wherein the number of the random numbers is equal to the number of samples in the target data set; Sorting the random numbers and determining an index sequence of positions after sorting relative to positions before sorting; Reordering the target data set according to the index sequence to obtain a first data set, and selecting a training set and a test set of preset proportions from the first data set; Initializing the weight of each training sample in the training set to obtain a target training set, wherein the weight of each training sample in the target training set is all equal and the sum of all weights is 1; Using the target training set to train a classifier, obtaining a target classifier when a preset stop condition is met, and saving the target classifier to a classifier array; Processing the test set using the target classifier to obtain a prediction result vector output by the target classifier; According to the error rate between the prediction result vector and the true label, the weight coefficient of the target classifier is updated, and the weight coefficient of the training sample is updated.

3. The method according to claim 1, characterized in that After combining the classifiers generated by iterative training and their weight coefficients into a target classifier, the method further includes: When the classification error rate of the target classifier is greater than a preset threshold, the classifiers generated by iterative training are used in turn to perform comparisons with the target classifier based on the classification error rate. If the classification error rate of the target classifier is greater than the classification error rate of the classifier compared with it, the model parameters of the target classifier are adjusted according to the model parameters of the classifier compared with it until the classification error rate of the target classifier is less than the preset threshold.

4. The method according to claim 3, characterized in that The model parameters of the target classifier are adjusted, including: searching for a parameter combination that minimizes the classification error rate of the target classifier within a preset value range through cross-validation and grid search methods.

5. The method according to claim 1, characterized in that Perform vital sign extraction on the processed historical operation data, including: For the number of boards, extracting the number of different types of boards and the port density of the boards as features; For the type of the machine room where the optical line terminal equipment is located, the scale information, power supply capacity information and cooling capacity information of the machine room are extracted as features.

6. The method according to claim 1, characterized in that The historical operation data is cleaned, including: removing data related to Skynet data and school data from the historical operation data.

7. A processing method for an optical line terminal device, characterized in that: include: Obtaining the operation data of optical line terminal equipment; The operation data is input into a target classifier to obtain a cutover plan corresponding to the optical line terminal device output by the target classifier, wherein the target classifier is obtained by training using the model training method described in any one of claims 1 to 6.

8. A model training device, characterized in that: include: An acquisition module is used to acquire historical operation data of an optical line terminal device, wherein the historical operation data includes: the number and location of wireless optical network interfaces of the optical line terminal device corresponding to different suppliers, the type of equipment, the number of port boards withdrawn from the network, the number of ports in use, the number of idle ports, the number of boards in use, the number of idle boards, the number of frames in use, the number of idle frames, and the type of the computer room where the optical line terminal device is located; A processing module, used for performing data cleaning processing on the historical operation data, and performing vital sign extraction processing and labeling processing on the processed historical operation data to obtain a target data set; A training module, used to iteratively train multiple classifiers using the target data set, and after completing the training of each classifier, adjust the weight coefficients of the classifiers and training samples involved in the training according to the classification error rate, and use the adjusted samples to train the next classifier; The combination module is used to combine the classifiers and weight coefficients generated by iterative training into a target classifier, wherein the target classifier is used to determine the cutover solution corresponding to the optical line terminal device.

9. A non-volatile storage medium, characterized in that: The non-volatile storage medium includes a stored program, wherein when the program is running, the device where the non-volatile storage medium is located is controlled to execute the model training method described in any one of claims 1 to 6.

10. An electronic device, characterized in that: include: A memory and a processor, wherein the processor is used to run a program stored in the memory, wherein the program executes the model training method described in any one of claims 1 to 6 when running.

11. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the model training method described in any one of claims 1 to 6 is implemented.

Images