CN113840190A - Optical path transmission quality prediction method based on ANN loss function optimization - Google Patents

Abstract

The invention discloses an optical path transmission quality prediction method based on ANN loss function optimization, wherein a regularization term is introduced based on traditional MSE and MAE loss functions, and the high estimation value and proportion of a model are reduced by giving a larger penalty to the high estimation value, so that the aim of reducing the maximum positive deviation of the model is fulfilled. Compared with the traditional QoT estimation model based on MSE and MAE, the method provided by the invention can greatly reduce the maximum positive deviation value of the model while paying attention to the accuracy of the model, thereby achieving the purposes of improving the distribution efficiency of the optical path modulation format and improving the capacity of the optical network.

Description

Optical path transmission quality prediction method based on ANN loss function optimization

Technical Field

The invention relates to the technical field of optical communication, in particular to an optical path transmission quality prediction method based on ANN loss function optimization.

Background

The emergence of new technologies such as cloud computing, edge computing, internet of things, virtual reality, artificial intelligence and 5G leads to explosive growth of network data. As one of the most important infrastructures for network data transmission, the optical transport network has a rapidly increasing capacity demand, and further development of a high-capacity transmission system is required to improve the network utilization rate.

At present, in a conventional optical transmission network, in order to satisfy smooth deployment service of an optical network in a life cycle of the optical network, a sufficient margin needs to be reserved for the optical network at a deployment stage of the optical network, where the margin mainly includes a system margin, an unallocated margin, and a design margin. The system margin is reserved margin due to device aging, network load change, polarization effect and the like; unallocated margin is a margin reserved due to a transmission capacity of a device and a deviation of a transmission distance from an actual demand; the design margin is a margin reserved due to inaccuracy of estimation of signal transmission quality.

Therefore, an accurate optical path transmission quality prediction method is one of key technologies for realizing a low-margin optical network, and predicting the transmission quality (QoT) of an optical path before deployment is a crucial step for the optimal design of the optical network. On one hand, whether the optical path can be connected or not can be judged, on the other hand, on the premise of ensuring the reliability of the optical path, the available highest-order modulation format is distributed to the optical path according to the predicted value of the optical path, so that the actual available capacity of the network is improved, and low-margin transmission is realized.

The traditional signal quality estimation models are divided into two types, one type is an accurate analysis model, an accurate QoT value can be obtained by a method for solving an optical transmission equation, but the calculation time is long, and the method is not suitable for being applied to large networks and dynamic networks; the other is an approximate estimation model, the calculation speed of the model is high, but the design margin is introduced due to the uncertainty of the model, and the spectrum efficiency is reduced. In view of the successful application of machine learning in both the network layer and the physical layer, in order to balance the computational complexity and accuracy of QoT prediction well, the physical layer behavior can be implicitly captured by the change of data based on the data collected from the optical network by using a machine learning method, so as to predict the QoT value of the link where no connection is established according to the link information of the established connection.

The two main solutions at present are based on QoT estimation using classification and regression models in machine learning. The main idea of the classification model is to judge whether the link request can be established, and the main idea of the regression model is to directly predict the estimation value of the output QoT. Machine learning classification models generally require more balanced data to train, and data imbalance is a typical existing problem in classification models, which generally means that the number of samples in different classes is greatly different. Regression models are less dependent on data than classification models and can provide more effective information, including estimates of signal quality and margins, etc.

In recent years, Artificial Neural Networks (ANN) have proven to be a promising QoT estimation technique, with some work achieving good performance. The results of the transmission quality estimation are generally classified into two categories, i.e., a predicted value greater than an actual value (hereinafter referred to as "overestimation") and a predicted value less than an actual value (hereinafter referred to as "underestimation"). Overestimating means that the predicted value is larger than the actual value, which may result in improper setting of the optical path and unreliable optical network. To solve this problem, a margin is usually introduced in the network to guarantee a certain reliability, which can be set according to the maximum high evaluation value of the model. On the other hand, underestimation means that the predicted value is smaller than the actual value, and the optical path also selects a lower modulation format, resulting in lower network capacity. In order to ensure the maximum reliability and capacity of the optical network, when deploying the optical path, a margin is generally considered, the margin is subtracted from a predicted value of the optical path, and a highest-order modulation format is selected. The optical path deployment scenario can therefore be expressed as:

OSNR_p-margin≥FEC_limit

wherein OSNR_pRepresenting the predicted value of the optical path, margin is the design margin introduced due to model estimation inaccuracy, and this value is typically set according to the maximum positive deviation of the model. FEC (Forward error correction)_limitRefers to a threshold required by the modulation format used, above which signals are considered recoverable and "error-free".

The existing literature trains a neural network by using Mean Square Error (MSE) and absolute error (MAE) as loss functions, both of which are symmetric loss functions, giving the same penalty for both overestimation and underestimation, mainly focusing on the average accuracy of the model, i.e. the average error of all samples, and not additionally focusing on the maximum positive deviation of the model, i.e. the design margin of the network. In order to ensure the reliability of the optical path, an operator and an equipment provider may allocate a modulation format for the optical path deployment by using a difference between a predicted value of the signal quality of the optical path and a design margin, and the design margin is too high, which may cause a modulation format to be selected too low during the optical path deployment, thereby reducing the network capacity.

Therefore, in order to solve the problem, it is necessary to optimize the loss function of the ANN, so as to reduce the maximum positive deviation of the model while paying attention to the accuracy of the model, thereby increasing the network capacity to the maximum extent.

Disclosure of Invention

Aiming at the problems, the invention provides an optical path transmission quality prediction method based on ANN loss function optimization by introducing regularization terms based on the existing MSE and MAE loss functions, and the method greatly reduces the maximum positive deviation value of a model while paying attention to the accuracy of the model so as to improve the distribution efficiency of an optical path modulation format and further improve the capacity of an optical network.

In order to achieve the above purpose, the invention provides the following technical scheme:

an optical path transmission quality prediction method based on ANN loss function optimization comprises the following steps:

the first step is as follows: initializing a neural network model structure, setting learning rate and iteration times, and initializing a weight for each layer randomly;

the second step is that: inputting training data into a neural network, and obtaining a corresponding output result according to a forward propagation algorithm;

the third step: calculating the deviation between the model output and a given reference output according to the loss function α _ MSE or α _ MSE;

wherein y represents a true value, y ^ represents a predicted value, alpha is a penalty coefficient, I (x) is a regularization term, when the predicted value is greater than the true value, namely when the predicted value is overestimated, extra penalty is given by adjusting the value of alpha, and when the predicted value is less than the true value, namely when the predicted value is underestimated, the regularization term has a value of 0;

the fourth step: the weight of the neuron is adjusted by adopting a random gradient descent method, so that the prediction error of the neural network model is minimum;

the fifth step: repeating the forward transmission and backward propagation of the training data set among the neural networks, and updating the weight of the neuron to minimize the global loss of each learning iteration.

Compared with the prior art, the invention has the beneficial effects that:

the invention provides an ANN loss function optimization-based optical path transmission quality prediction method, wherein a loss function is mainly based on the existing symmetric loss function MAE and MSE, a regularization item is introduced, when a predicted value is larger than a real value, extra punishment is given to a model, and the purpose of reducing the maximum positive deviation of the model is realized, so that the distribution efficiency of an optical path modulation format is improved, and the capacity of an optical network is further improved.

Drawings

In order to more clearly illustrate the embodiments of the present application or technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments described in the present invention, and other drawings can be obtained by those skilled in the art according to the drawings.

Fig. 1 is a schematic flow chart of an optical path transmission quality prediction method based on ANN loss function optimization according to the present invention.

Detailed Description

For a better understanding of the present solution, the method of the present invention is described in detail below with reference to the accompanying drawings.

1. Loss function design

The loss functions in most research literature at present are symmetric loss functions, such as: the MSE mean square error function, MAE, both penalties for overestimation and underestimation are the same.

In order to reduce the maximum positive bias of the model, a regularization term i (x) is introduced based on MAE and MSE, and is expressed as formula (3):

wherein, y represents the true value,

and expressing a predicted value, wherein alpha is a penalty coefficient to achieve the aim of reducing the maximum overestimation, when the predicted value is larger than the true value, namely when the overestimation is carried out, an additional penalty is given by adjusting the value of alpha, and when the predicted value is smaller than the true value, namely when the underestimation is carried out, the value of the regularization term is 0.

We introduce this regularization term based on MSE and MAE, proposing two new loss functions, α _ MSE and α _ MAE, expressed as:

2. optical path transmission quality deviation estimation method based on artificial neural network

Artificial neural networks are suitable for modeling systems with complex and nonlinear relationships between inputs and outputs. Generally, the structure of an artificial neural network model consists of three types of layers: an input layer, a hidden layer, and an output layer. The input layer is composed of many neurons corresponding to different input data characteristics, for example: information such as channel transmitting power, channel opening state, optical path length, optical path span number and the like; the hidden layers are connected with the input layer, all the features of the input layer are combined together, and then the hidden layers are connected to the output layer to form a neural network, wherein the output layer is generally an OSNR value corresponding to an optical path. An activation function is applied to each neuron from the input layer to the output layer. What we need to do is to set the hyper-parameters of the model, such as the number of layers, the number of neurons in each layer, the activation function and the learning rate, and the neural network can adjust the weight of neurons in each model layer through the training process of the neural network, automatically fit the data distribution and obtain the result closest to the expected output value.

The method for predicting the optical path transmission quality based on ANN loss function optimization applies the loss function to the QoT estimation process, and the whole steps of the scheme are shown in figure 1.

The first step is as follows: initializing a neural network model structure, setting hyper-parameters such as learning rate and iteration times, and randomly initializing weights for each layer.

The second step is that: and inputting training data into the neural network, and obtaining a corresponding output result according to a forward propagation algorithm.

The third step: the deviation between the model output and a given reference output is calculated from the loss function α _ MSE or α _ MSE.

The fourth step: and (3) adjusting the weight of the neuron by adopting a random gradient descent method to minimize the prediction error of the neural network model.

The fifth step: repeating the forward transmission and backward propagation of the training data set among the neural networks, and updating the weight of the neuron to minimize the global loss of each learning iteration.

After a certain learning process iteration, the complete artificial neural network model is applied to a new input data set to predict the target value. The main content algorithm is shown in table 1.

According to the optical path transmission quality prediction method based on ANN loss function optimization, regularization terms are introduced based on traditional MSE and MAE loss functions, high estimation value and proportion of a model are reduced in a mode of giving higher punishment to the high estimation value, and therefore the purpose of reducing the maximum positive deviation of the model is achieved. Compared with the traditional QoT estimation model based on MSE and MAE, the method provided by the invention can greatly reduce the maximum positive deviation value of the model while paying attention to the accuracy of the model, thereby achieving the purposes of improving the distribution efficiency of the optical path modulation format and improving the capacity of the optical network.

The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: it is to be understood that modifications may be made to the technical solutions described in the foregoing embodiments, or equivalents may be substituted for some of the technical features thereof, but such modifications or substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims (1)

1. An optical path transmission quality prediction method based on ANN loss function optimization is characterized by comprising the following steps:

the first step is as follows: initializing a neural network model structure, setting learning rate and iteration times, and initializing a weight for each layer randomly;

the second step is that: inputting training data into a neural network, and obtaining a corresponding output result according to a forward propagation algorithm;

the third step: calculating the deviation between the model output and a given reference output according to the loss function α _ MSE or α _ MSE;

wherein y represents a true value, y ^ represents a predicted value, alpha is a penalty coefficient, I (x) is a regularization term, when the predicted value is greater than the true value, namely when the predicted value is overestimated, extra penalty is given by adjusting the value of alpha, and when the predicted value is less than the true value, namely when the predicted value is underestimated, the regularization term has a value of 0;

the fourth step: the weight of the neuron is adjusted by adopting a random gradient descent method, so that the prediction error of the neural network model is minimum;

the fifth step: repeating the forward transmission and backward propagation of the training data set among the neural networks, and updating the weight of the neuron to minimize the global loss of each learning iteration.

Images