OTN communication equipment performance prediction method based on deep neural integrated network

Abstract

The invention discloses an OTN communication equipment performance prediction method based on a deep neural integrated network. The method comprises the following steps: 1) collecting OTN communication equipment performance prediction data; 2) data standardization; 3) establishing a deep neural integrated network prediction model, and training the prediction model; establishing a deep neural integrated network prediction model, and training the prediction model by taking the three-dimensional vector group in the training set as input and the performance parameters as output; and 4) retaining the weight parameters obtained by training, and obtaining the trained deep neural integrated network prediction model to carry out OTN communication equipment performance prediction. According to the prediction method, the influence of the hardware equipment model on the actual performance of the equipment is considered, the relevance between the network topology structure and the working condition of the OTN equipment is also considered, and meanwhile, the equipment operation environment temperature and humidity are also brought into the decision-making basis category of the prediction model, so that the accuracy is relatively high.

Description

Technical Field

[0001] The present invention relates to communication device technologies, and in particular, to a method for predicting the performance of an OTN communication device based on a deep neural integration network. Background Art

[0002] As a cornerstone of the current social communication field, optical fiber communication technology undertakes data transmission tasks in fields such as the Internet, 5G communication, and enterprise management. Optical Transport Network (OTN) is a high-performance optical fiber communication network technology, especially suitable for long-distance and large-capacity data transmission. It plays a crucial role in modern communication networks and provides technical support for the informatization development of society. Its parameters mainly include transmission efficiency, signal quality, network reliability, etc. These parameters reflect the working efficiency, working quality, and supported service scope of the OTN network from different perspectives. Currently, there is still much room for optimization in OTN communication. On the one hand, engineers extend the network lifespan and increase transmission quality by optimizing the physical layer deployment of the OTN network and the optical signal transmission mode. In recent years, many scholars have also proposed the concept of Self-healing OTN, hoping to enable the OTN network to automatically switch the optical signal transmission path when a fault occurs through the introduction of relevant machine learning technologies, so that the network has a certain self-repair ability. On the other hand, people also start from the data structure and reduce the communication error rate by optimizing the frame structure of the transmitted data. All in all, under the current communication industry's pursuit of low latency and high throughput, the OTN communication system has great potential and room for optimization. The optimization process is inevitably inseparable from the prediction of performance parameters. Therefore, a low-cost and high-accuracy OTN performance parameter prediction method has become an urgent need in the current industry.

[0003] With the proposal of neural networks, the research in the field of artificial intelligence has advanced rapidly. In pursuit of the accuracy of prediction ability, deep neural networks (DNN), derived from classical multi-layer perceptrons, achieve the purpose of accurately predicting complex tasks by adding multi-hidden layer structures, non-linear activation functions, forward propagation, backward propagation, and optimization algorithms. Based on the above background, we propose a method for predicting the performance of an OTN communication device based on a deep neural integration network. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method for predicting the performance of an OTN communication device based on a deep neural integration network for the defects in the prior art.

[0005] The technical solution adopted by the present invention to solve its technical problems is as follows: An OTN communication device performance prediction method based on a deep neural ensemble network, comprising the following steps:

[0006] 1) Collect OTN communication device performance prediction data;

[0007] The OTN communication device performance prediction data includes: device model parameters, network topology structure parameters, environmental parameters, and performance parameters;

[0008] The device model parameters include interface model, fiber type, and special technology module model. Among them, the special technology module includes a coherent optical communication module and a dispersion compensation module;

[0009] The network topology structure parameters include network topology structure type and number of nodes;

[0010] The environmental parameters are the temperature and humidity of the OTN communication device working environment;

[0011] The performance parameters include data transmission rate, delay, and packet loss rate;

[0012] 2) Data standardization;

[0013] Perform encoding conversion on the device model parameters, network topology structure parameters, and environmental parameters to convert them into a 3D vector group;

[0014] Perform Z-score normalization on the performance parameters to convert them into continuous numerical values between 0 and 1;

[0015] Divide the converted data into a training set and a validation set;

[0016] 3) Establish a deep neural ensemble network prediction model and train the prediction model;

[0017] Establish a deep neural ensemble network prediction model, use the 3D vector group in the training set as the input, and the performance parameters as the output, and train the prediction model;

[0018] The input layer is designed as a feature neuron interface so that the input layer can receive the 3D vector group converted from the OTN device parameters. The input interface of the hidden layer corresponds to the output interface of the input layer, and the hidden layer is a three-layer hidden layer;

[0019] 4) Retain the obtained weight parameters after training, and obtain the trained deep neural ensemble network prediction model for OTN communication device performance prediction.

[0020] According to the above solution, in step 2), the encoding of the device model parameters, network topology structure parameters, and environmental parameters is as follows:

[0021] The device model parameter encoding is to perform conventional digital encoding on the interface model, fiber type, and special technology module model;

[0022] The network topology structure parameter encoding is that the first digit is the structure type flag bit, and the subsequent digits represent the number of nodes; for example, "star structure - 12 nodes" is encoded as 012;

[0023] The environmental parameter encoding includes a three - digit temperature encoding and a three - digit humidity encoding.

[0024] According to the above - mentioned scheme, in step 3), the loss function used for model training is:

[0025]

[0026] where y i is the actual value of the i - th sample in the training set, is the predicted value made by the model based on the features of the i - th sample, and n is the data length.

[0027] According to the above - mentioned scheme, in step 3), the activation function of the model uses the sigmoid function to introduce a non - linear factor, which is expressed as follows:

[0028]

[0029] w,b = minLoss DNN (w,b)

[0030] where w is the weight parameter, b is the bias, x is the input vector, and y is the output vector.

[0031] The beneficial effects produced by the present invention are:

[0032] 1. The prediction scheme conceived by the present invention not only takes into account the influence of the hardware device model on the actual performance of the device, but also considers the correlation between the network topology structure and the working conditions of the OTN device. At the same time, the temperature and humidity of the device operating environment are also incorporated into the decision - making basis category of the prediction model. Therefore, this prediction method has high accuracy.

[0033] 2. The present invention uses a deep neural ensemble network as the training model framework. Compared with traditional machine learning, it can give full play to the characteristic parameters of the sample data set, making the obtained model weight parameters perform better in solving actual prediction problems. Description of the Drawings

[0034] The present invention will be further described below in conjunction with the drawings and embodiments. In the drawings:

[0035] Figure 1 is the flowchart of the method of the embodiment of the present invention;

[0036] Figure 2 It is a schematic diagram of the training process of the deep neural integration network model according to an embodiment of the present invention. Detailed implementation manners

[0037] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0038] As Figure 1 shown, a method for predicting the performance of an OTN communication device based on a deep neural integration network includes the following steps:

[0039] 1) Collect performance prediction data of the OTN communication device;

[0040] The performance prediction data of the OTN communication device includes: device model parameters, network topology structure parameters, environmental parameters, and performance parameters;

[0041] The device model parameters include interface model, fiber type, and special technology module model. Among them, the special technology module includes a coherent optical communication module and a dispersion compensation module;

[0042] The network topology structure parameters include network topology structure type and number of nodes;

[0043] The environmental parameters are the temperature and humidity of the working environment of the OTN communication device;

[0044] The performance parameters include data transmission rate, delay, and packet loss rate;

[0045] 2) Data standardization;

[0046] Encode the device model parameters, network topology structure parameters, and environmental parameters into a 3D vector group;

[0047] Encode the device model parameters, network topology structure parameters, and environmental parameters as follows:

[0048] Encode the device model parameters as conventional digital encoding of the interface model, fiber type, and special technology module model;

[0049] Encode the network topology structure parameters as the first digit being the structure type flag bit, and the subsequent digits representing the number of nodes; for example, "star structure - 12 nodes" is encoded as 012; for an OTN communication system with an automatic healing function or other dynamic topology structures, it can also be marked and distinguished to fully consider its impact.

[0050] Encode the environmental parameters including three-digit encoding of temperature and three-digit encoding of humidity.

[0051] Perform Z-score normalization on the performance parameters to convert them into continuous numerical values between 0 and 1;

[0052]

[0053] where X norm is the normalization result, μ is the mean of the collected data, σ is the standard deviation of the collected data, and X is the original numerical value.

[0054] Finally, use the python format conversion tool to convert the data file format into a 3×N dataset format;

[0055] Make the converted data into an available dataset and divide it according to 3:1 as the training set and validation set of the deep neural ensemble network;

[0056] 3) Establish a deep neural ensemble network prediction model and train the prediction model;

[0057] Establish a deep neural ensemble network prediction model, use the 3D vector group in the training set as the input, and the performance parameter as the output, and train the prediction model;

[0058] The input layer is designed as a feature neuron interface so that the input layer can receive the 3D vector group converted from the OTN device parameters. The input interface of the hidden layer corresponds to the output interface of the input layer, and the hidden layer is a three-layer hidden layer;

[0059] The loss function used for model training is:

[0060]

[0061] where y i is the actual value of the i-th sample in the training set, is the prediction value made by the model from the features of the i-th sample, and n is the data length.

[0062] The activation function of the model uses the sigmiod function to introduce a non-linear factor, which is expressed as follows:

[0063]

[0064] w,b=minLoss DNN (w,b)

[0065] where w is the weight parameter, b is the bias, x is the input vector, and y is the output vector.

[0066] First, use the sigmoid function to introduce a non-linear factor and complete the numerical range migration at the same time, then use the MAE function to calculate the model loss, and the weight parameter exists as part of the denominator exponential factor in the sigmoid function.

[0067] As Figure 2 shown in the training process of the deep neural integration network model used in the present invention, after the OTN device information data is input into the neural network as the training set, classification and non-linearization operations are first completed in the input layer and propagated backward to the subsequent layers. The middle layer is responsible for calculating the loss function and continuously optimizing the weight parameters using the gradient descent algorithm. At the same time, the data will be propagated forward in the middle layer, causing the weight parameters to be iterated repeatedly, and the prediction accuracy of the model will increase. In the validation session of the training, the output layer predicts the results and validates them. The validation process itself will also be used as learning materials for the neural network to improve its latter function. After the training is completed, the training parameters can be optimized according to the confusion matrix of the model. Appropriate learning rates and batch data volumes will improve the training effect.

[0068] 4) Retain the obtained weight parameters after training, and obtain the trained deep neural integration network prediction model for OTN communication device performance prediction.

[0069] After obtaining the input parameters, the model will quickly and accurately generate corresponding performance prediction results, covering a series of performance indicators such as bandwidth utilization, transmission delay, and data loss rate. These results provide key decision-making support for network administrators, including network optimization, device maintenance, and fault diagnosis. Through real-time monitoring and prediction, potential problems can be detected in a timely manner, improving network performance and stability.

[0070] It should be understood that for those of ordinary skill in the art, improvements or transformations can be made according to the above description, and all such improvements and transformations should fall within the protection scope of the appended claims of the present invention.

Claims

1. A method for predicting OTN communication equipment performance based on a deep neural integration network, characterized in that: The following steps are involved: 1) Collect OTN communication equipment performance prediction data; The OTN communication equipment performance prediction data includes: equipment model parameters, network topology parameters, environmental parameters and performance parameters; The equipment model parameters include interface model, optical fiber type and special technology module model, wherein the special technology module includes a coherent optical communication module and a dispersion compensation module; The network topology parameters include the network topology type and the number of nodes; The environmental parameters are the temperature and humidity of the working environment of the OTN communication equipment; The performance parameters include data transmission rate, delay and packet loss rate; 2) Data standardization; Encode the equipment model parameters, network topology parameters and environment parameters and convert them into a 3D vector group; Normalize the performance parameters into continuous values ​​between 0 and 1; Divide the transformed data into training set and validation set; 3) Establish a deep neural ensemble network prediction model and train the prediction model; Establish a deep neural ensemble network prediction model, use the 3D vector group in the training set as input and the performance parameters as output to train the prediction model; The input layer is designed as a characteristic neuron interface so that the input layer receives a 3D vector group converted from the OTN device parameters. The hidden layer input interface corresponds to the input layer output interface, and the hidden layer has three hidden layers. 4) The weight parameters obtained through training are retained to obtain the trained deep neural integration network prediction model for OTN communication equipment performance prediction.

2. The method for predicting OTN communication equipment performance based on deep neural integration network according to claim 1 is characterized in that: In step 2), the device model parameters, network topology parameters and environment parameters are encoded as follows: The equipment model parameter coding is the conventional digital coding of the interface model, optical fiber type and special technology module model; The network topology parameter is encoded as follows: the first bit is the structure type flag, and the subsequent numbers represent the number of nodes; The environmental parameter coding includes three-digit coding of temperature and three-digit coding of humidity.

3. The OTN communication equipment performance prediction method based on deep neural integration network according to claim 1 is characterized in that: In step 3), the loss function used in model training is: Among them, y i is the actual value of the i-th sample in the training set, is the predicted value made by the model based on the i-th sample feature, and n is the data length.

4. The method for predicting OTN communication equipment performance based on deep neural integration network according to claim 3 is characterized in that: In step 3), the activation function of the model uses the sigmiod function to introduce a nonlinear factor, which is expressed as follows: w,b=minLoss DNN (w,b) Among them, w is the weight parameter, b is the bias, x is the input vector, and y is the output vector.

5. An electronic device, characterized in that: include: one or more processors; as well as a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors are caused to execute the method according to any one of claims 1 to 4.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 4 is implemented.

Images