Link power control method and system of coherent optical wavelength division multiplexing transmission system

Abstract

The invention provides a link power control method and system for a coherent optical wavelength division multiplexing transmission system. The method comprises the following steps: establishing a coherent optical wavelength division multiplexing transmission simulation system; configuring the transmitting power of each channel in the wavelength division multiplexing system; taking the initial state of the transmitting power as the input of an execution neural network, obtaining adjustment values of transmitting power of different channels, and calculating the transmitting power of the channels through the initial state and the adjustment values of the transmitting power; configuring attenuation of each channel of a wavelength selection switch in the wavelength division multiplexing system; and taking the initial state of the wavelength selection switch as the input of the execution neural network, obtaining the attenuation adjustment value of the wavelength selection switch, and calculating the channel transmitting power through the initial state and the attenuation adjustment value. On the basis of a deep reinforcement learning algorithm, optimal configuration of power in a link is realized by adjusting emission work of different channels in a coherent optical wavelength division multiplexing system and attenuation of different channels in a wavelength selection switch, so that the communication quality of each channel of a receiving end after transmission is optimized.

Description

Technical Field

[0001] The present invention relates to the technical field of link power control, and in particular to a link power control method and system for a coherent optical wavelength division multiplexing transmission system. Background Art

[0002] Coherent optical wavelength division multiplexing system is the mainstream system of optical transmission. In the coherent optical wavelength division multiplexing system, the signals of each channel are affected by the nonlinearity of the optical fiber, and the transmission quality is inconsistent, which affects the overall transmission level of the system. Link power control is used to balance the communication quality of each channel. The current method of controlling link power based on deep learning relies on an accurate model of link transmission.

[0003] Patent application document CN116938383A discloses a nonlinear compensation method and system based on a module multiplexing system, including the following steps: selecting any one mode to be set as the second mode, and setting the other modes to the first mode; applying a frequency offset to the wavelength division multiplexed optical signal of the second mode by using a first optical frequency shifter; inputting the wavelength division multiplexed optical signal of the first mode and the wavelength division multiplexed optical signal of the second mode after the offset into the same mode division multiplexer, and outputting from the same mode division multiplexer after passing through the same multimode optical fiber; applying a reverse frequency offset to the wavelength division multiplexed optical signal of the second mode output from the mode division multiplexer by using a second optical frequency shifter; performing wavelength division multiplexing on the wavelength division multiplexed optical signal of the first mode output from the mode division multiplexer and the wavelength division multiplexed optical signal of the second mode to which the reverse frequency offset is applied, and completing nonlinear compensation. However, this patent cannot completely solve the existing technical problems, nor can it meet the requirements of the present invention. Summary of the invention

[0004] In view of the defects in the prior art, an object of the present invention is to provide a link power control method and system for a coherent optical wavelength division multiplexing transmission system.

[0005] The link power control method of the coherent optical wavelength division multiplexing transmission system provided by the present invention comprises:

[0006] Step 1: Build a coherent optical wavelength division multiplexing transmission simulation system, including setting the length of the optical fiber in the link, the parameters of the optical amplifier, and setting the overall transmission distance;

[0007] Step 2: Use the simulation environment to train the deep neural network and configure the transmit power of each channel in the wavelength division multiplexing system;

[0008] Step 3: The initial state of the transmit power is used as input into the trained execution neural network to obtain the adjustment value of the transmit power of different channels. The channel transmit power is calculated based on the initial state and the transmit power adjustment value.

[0009] Step 4: Use the simulation environment to train the deep neural network and configure the attenuation of each channel of the wavelength selective switch in the wavelength division multiplexing system;

[0010] Step 5: The initial state of the wavelength selective switch is used as input into the trained execution neural network to obtain the adjustment value of the wavelength selective switch attenuation, and the channel transmission power is calculated based on the initial state and the attenuation adjustment value.

[0011] Preferably, the step 2 comprises:

[0012] Step 2.1: Take the initial value of the transmission power as the starting state, take the starting state as the input of the execution neural network, and obtain the output as the adjustment value of the transmission power. Calculate the channel transmission power through the starting state and the transmission power adjustment value, and then transmit it to obtain the result at the receiving end; obtain the signal-to-noise ratio of each channel at the receiving end, calculate the corresponding reward through the signal-to-noise ratio obtained at the receiving end and the target signal-to-noise ratio, and update the starting state to the next state; collect the starting state, the transmission power adjustment value, the next state and the reward into a data pair, store them in the data pool, and generate the data in the data pool by continuously interacting with the execution neural network and the simulation transmission system;

[0013] Step 2.2: Train the critical neural network with the data in the data pool, use the current state and execution action in the data pair as the input of the critical neural network, and obtain its output; subtract the output from the reward function and take the absolute value as the loss function of the critical neural network, and update the parameters of the critical neural network by the gradient descent method;

[0014] Step 2.3: Train the execution neural network through the data in the data pool and the critical neural network, use the current state in the data pair as the input of the execution neural network to obtain the output of the execution neural network, use the output and the current state as the input of the critical neural network to obtain the output of the critical neural network; use the opposite number of the output of the critical neural network as the loss function, and update the parameters of the execution neural network through the gradient descent method.

[0015] Preferably, the channel transmit power is calculated by the initial state and the transmit power adjustment value, and the expression is:

[0016] P ch =P 0 +P Δ

[0017] Among them, P ch is the channel transmission power, P 0 is the initial state, P Δ is the power adjustment value;

[0018] The corresponding reward is calculated by the signal-to-noise ratio obtained at the receiving end and the target signal-to-noise ratio. The expression is:

[0019] R i =||SNR i -SNR target || 2

[0020] Among them, R i For reward, SNR i is the signal-to-noise ratio at the receiving end, SNR target is the target signal-to-noise ratio.

[0021] Preferably, step 4 comprises:

[0022] Step 4.1: Take the initial attenuation value as the starting state, take the starting state as the input of the execution neural network, and obtain the output as the attenuation adjustment value. Calculate the attenuation of each channel through the starting state and the attenuation adjustment value, then transmit and obtain the result at the receiving end; finally, obtain the signal-to-noise ratio of each channel at the receiving end, calculate the corresponding reward through the signal-to-noise ratio obtained at the receiving end and the target signal-to-noise ratio, and update the starting state to the next state; collect the starting state, attenuation adjustment value, next state and reward into a data pair, store them in the data pool, and generate data in the data pool through the continuous interaction between the execution neural network and the simulation transmission system;

[0023] Step 4.2: Use the data in the data pool to train the critical neural network;

[0024] Step 4.3: Use the data in the data pool and the critical neural network to train the executive neural network.

[0025] Preferably, the attenuation of each channel is calculated through the initial state and the attenuation adjustment value, and the expression is:

[0026] A ch =A 0 +A Δ

[0027] Among them, A ch is the channel attenuation, A 0 is the initial state, A Δ is the attenuation adjustment value.

[0028] The link power control system of the coherent optical wavelength division multiplexing transmission system provided by the present invention comprises:

[0029] Module M1: Build a coherent optical wavelength division multiplexing transmission simulation system, including setting the length of the optical fiber in the link, the parameters of the optical amplifier, and setting the overall transmission distance;

[0030] Module M2: Use the simulation environment to train the deep neural network and configure the transmit power of each channel in the wavelength division multiplexing system;

[0031] Module M3: takes the initial state of the transmit power as input into the trained execution neural network, obtains the adjustment value of the transmit power of different channels, and calculates the channel transmit power based on the initial state and the transmit power adjustment value;

[0032] Module M4: Use the simulation environment to train deep neural networks and configure the attenuation of each channel of the wavelength selective switch in the wavelength division multiplexing system;

[0033] Module M5: The initial state of the wavelength selective switch is used as input into the trained execution neural network to obtain the adjustment value of the wavelength selective switch attenuation, and the channel transmission power is calculated based on the initial state and the attenuation adjustment value.

[0034] Preferably, the module M2 comprises:

[0035] Module M2.1: Take the initial value of the transmission power as the starting state, take the starting state as the input of the execution neural network, and obtain the output as the adjustment value of the transmission power. Calculate the channel transmission power through the starting state and the transmission power adjustment value, and then transmit it to obtain the result at the receiving end; obtain the signal-to-noise ratio of each channel at the receiving end, calculate the corresponding reward through the signal-to-noise ratio obtained at the receiving end and the target signal-to-noise ratio, and update the starting state to the next state; collect the starting state, the transmission power adjustment value, the next state and the reward into a data pair, store them in the data pool, and generate the data in the data pool through the continuous interaction between the execution neural network and the simulation transmission system;

[0036] Module M2.2: Train the critical neural network with the data in the data pool, use the current state and execution action in the data pair as the input of the critical neural network, and obtain its output; subtract the output from the reward function and take the absolute value as the loss function of the critical neural network, and update the parameters of the critical neural network through the gradient descent method;

[0037] Module M2.3: Train the execution neural network through the data in the data pool and the critical neural network, use the current state in the data pair as the input of the execution neural network to obtain the output of the execution neural network, use the output and the current state as the input of the critical neural network to obtain the output of the critical neural network; use the opposite number of the output of the critical neural network as the loss function, and update the parameters of the execution neural network through the gradient descent method.

[0038] Preferably, the channel transmit power is calculated by the initial state and the transmit power adjustment value, and the expression is:

[0039] P ch =P 0 +P Δ

[0040] Among them, P ch is the channel transmission power, P 0 is the initial state, P Δ is the power adjustment value;

[0041] The corresponding reward is calculated by the signal-to-noise ratio obtained at the receiving end and the target signal-to-noise ratio. The expression is:

[0042] R i =||SNR i -SNR target || 2

[0043] Among them, R i For reward, SNR i is the signal-to-noise ratio at the receiving end, SNR target is the target signal-to-noise ratio.

[0044] Preferably, the module M4 comprises:

[0045] Module M4.1: Take the initial attenuation value as the starting state, take the starting state as the input of the execution neural network, and obtain the output as the attenuation adjustment value. Calculate the attenuation of each channel through the starting state and the attenuation adjustment value, then transmit and obtain the result at the receiving end; finally, obtain the signal-to-noise ratio of each channel at the receiving end, calculate the corresponding reward through the signal-to-noise ratio obtained at the receiving end and the target signal-to-noise ratio, and update the starting state to the next state; collect the starting state, attenuation adjustment value, next state and reward into data pairs, store them in the data pool, and generate data in the data pool through the continuous interaction between the execution neural network and the simulation transmission system;

[0046] Module M4.2: Use data from the data pool to train a critical neural network;

[0047] Module M4.3: Use data from the data pool and critique neural networks to train and execute neural networks.

[0048] Preferably, the attenuation of each channel is calculated through the initial state and the attenuation adjustment value, and the expression is:

[0049] A ch =A 0 +A Δ

[0050] Among them, A ch is the channel attenuation, A 0 is the initial state, A Δ is the attenuation adjustment value.

[0051] Compared with the prior art, the present invention has the following beneficial effects:

[0052] (1) Based on a deep reinforcement learning algorithm, the present invention achieves optimal configuration of power in the link by adjusting the transmission power of different channels in the coherent optical wavelength division multiplexing system and the attenuation of different channels in the wavelength selective switch, thereby optimizing the communication quality of each channel at the receiving end after transmission;

[0053] (2) The present invention uses deep reinforcement learning to enable the neural network to be trained without relying on an accurate differentiable model, thereby solving the problem of different communication qualities of different channels in a coherent optical communication system and enabling different channels to obtain the same signal-to-noise ratio. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0055] Figure 1 For the generation of training data and the construction of data pool;

[0056] Figure 2 Flowchart for evaluating network training;

[0057] Figure 3 To perform network training flow chart;

[0058] Figure 4 Flow chart for setting the transmitting power and wavelength selection switch;

[0059] Figure 5 To set the transmit power optimization result. DETAILED DESCRIPTION

[0060] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can also be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0061] Example 1

[0062] The present invention provides a link power control method for a coherent optical wavelength division multiplexing transmission system based on deep reinforcement learning, comprising:

[0063] Step 1: Build a coherent optical wavelength division multiplexing transmission simulation system, including setting the length of the optical fiber in the link, the parameters of the optical amplifier, and setting the overall transmission distance;

[0064] Step 2: Use the simulation environment to train the deep neural network so that the execution neural network can configure the transmission power of each channel in the wavelength division multiplexing system;

[0065] Step 2.1: Select a reasonable initial value of the transmit power as the starting state, use the starting state as the input of the neural network, and obtain the output as the adjusted value of the transmit power. Then, calculate the channel transmit power according to formula (1) based on the starting state and the adjusted value of the transmit power, and then transmit it to obtain the result at the receiving end.

[0066] P ch =P 0 +P Δ (1)

[0067] Among them, P ch is the channel transmission power, P 0 is the initial state, P Δ is the power adjustment value.

[0068] Finally, the signal-to-noise ratio of each channel is obtained at the receiving end. The corresponding reward is calculated according to formula (2) based on the signal-to-noise ratio obtained at the receiving end and the target signal-to-noise ratio, and the starting state is updated to the next state.

[0069] R i =||SNR i -SNR target || 2 (2)

[0070] Among them, R i For reward, SNR i is the signal-to-noise ratio at the receiving end, SNR target is the target signal-to-noise ratio.

[0071] The starting state, transmission power adjustment value, next state and reward are collected as data pairs and stored in the data pool. The data in the data pool is generated by continuously interacting with the execution neural network and the simulation transmission system, such as Figure 1 .

[0072] Step 2.2: Train the critical neural network with the data in the data pool, such as Figure 2 The current state and execution action in the data pair are used as the input of the critical neural network to obtain its output; the output is subtracted from the reward function and the absolute value is taken as the loss function of the critical neural network, and the parameters of the critical neural network are updated by the gradient descent method.

[0073] Step 2.3: Train the execution neural network using the data in the data pool and the critical neural network, such as Figure 3 The current state in the data pair is used as the input of the execution neural network to obtain its output; the output and the current state are used as the input of the critical neural network to obtain the output of the critical neural network; the opposite of the output is used as the loss function, and then the parameters of the execution neural network are updated by the gradient descent method.

[0074] Step 3: Take the initial state of the transmit power as input and input it into the trained execution neural network to obtain the adjustment value of the transmit power of different channels. Then, the channel transmit power is calculated according to formula (1) based on the initial state and the transmit power adjustment value.

[0075] Step 4: Use the simulation environment to train the deep neural network so that the executing neural network can configure the attenuation of each channel of the wavelength selective switch in the wavelength division multiplexing system, such as Figure 4 .

[0076] Step 4.1: Select a reasonable initial attenuation value as the starting state, use the starting state as the input of the neural network, and obtain the output as the attenuation adjustment value. Then, the attenuation of each channel is calculated according to formula (3) based on the starting state and the attenuation adjustment value, and then the transmission is performed to obtain the result at the receiving end, such as Figure 5 .

[0077] A ch =A 0 +A Δ (3)

[0078] Among them, A ch is the channel attenuation, A 0 is the initial state, A Δ is the attenuation adjustment value.

[0079] Finally, the signal-to-noise ratio of each channel is obtained at the receiving end. According to the signal-to-noise ratio obtained at the receiving end and the target signal-to-noise ratio, the corresponding reward is calculated according to formula (2), and the starting state is updated to the next state.

[0080] The starting state, decay adjustment value, next state and reward are collected as data pairs and stored in the data pool. The data in the data pool is generated by the continuous interaction between the execution neural network and the simulation transmission system.

[0081] Step 4.2: Use the same process as step 2.2 to train the critical neural network using the data in the data pool.

[0082] Step 4.3: Use the same process as step 2.2 to train the executive neural network using the data in the data pool and the critical neural network.

[0083] Step 5: Take the initial state of the wavelength selective switch as input and input it into the trained execution neural network to obtain the attenuation adjustment value of the wavelength selective switch. Then, the channel transmission power is calculated according to formula (3) based on the initial state and attenuation adjustment value.

[0084] Example 2

[0085] The present invention also provides a link power control system of a coherent optical wavelength division multiplexing transmission system. The link power control system of the coherent optical wavelength division multiplexing transmission system can be realized by executing the process steps of the link power control method of the coherent optical wavelength division multiplexing transmission system, that is, those skilled in the art can understand the link power control method of the coherent optical wavelength division multiplexing transmission system as a preferred implementation of the link power control system of the coherent optical wavelength division multiplexing transmission system.

[0086] The link power control system of the coherent optical wavelength division multiplexing transmission system provided by the present invention includes: module M1: building a coherent optical wavelength division multiplexing transmission simulation system, including setting the length of the optical fiber in the link, the parameters of the optical amplifier, and setting the overall transmission distance; module M2: using the simulation environment to train the deep neural network, and configuring the transmission power of each channel in the wavelength division multiplexing system; module M3: using the starting state of the transmission power as input, inputting it into the trained execution neural network, obtaining the adjustment value of the transmission power of different channels, and calculating the channel transmission power through the starting state and the transmission power adjustment value; module M4: using the simulation environment to train the deep neural network, and configuring the attenuation of each channel of the wavelength selection switch in the wavelength division multiplexing system; module M5: using the starting state of the wavelength selection switch as input, inputting it into the trained execution neural network, obtaining the adjustment value of the attenuation of the wavelength selection switch, and calculating the channel transmission power through the starting state and the attenuation adjustment value.

[0087] The module M2 includes: module M2.1: taking the initial value of the transmission power as the starting state, taking the starting state as the input of the execution neural network, and obtaining the output as the adjustment value of the transmission power, calculating the channel transmission power through the starting state and the transmission power adjustment value, and then transmitting, and obtaining the result of the receiving end; obtaining the signal-to-noise ratio of each channel at the receiving end, calculating the corresponding reward through the signal-to-noise ratio obtained at the receiving end and the target signal-to-noise ratio, and updating the starting state to the next state; combining the starting state, the transmission power adjustment value, the next state and the reward into a data pair, storing them in a data pool, and generating data in the data pool by continuously interacting with the execution neural network and the simulation transmission system; module M2.2: through the data The critical neural network is trained with data from the data pool, and the current state and execution action in the data pair are used as the input of the critical neural network to obtain its output; the output is subtracted from the reward function and the absolute value is taken as the loss function of the critical neural network, and the parameters of the critical neural network are updated by the gradient descent method; Module M2.3: The execution neural network is trained with data from the data pool and the critical neural network, and the current state in the data pair is used as the input of the execution neural network to obtain the output of the execution neural network, and the output and the current state are used as the input of the critical neural network to obtain the output of the critical neural network; the opposite of the output of the critical neural network is used as the loss function, and the parameters of the execution neural network are updated by the gradient descent method.

[0088] The channel transmit power is calculated by the initial state and the transmit power adjustment value. The expression is:

[0089] P ch =P 0 +P Δ

[0090] Among them, P ch is the channel transmission power, P 0 is the initial state, P Δ is the power adjustment value;

[0091] The corresponding reward is calculated by the signal-to-noise ratio obtained at the receiving end and the target signal-to-noise ratio. The expression is:

[0092] R i =||SNR i -SNR target || 2

[0093] Among them, R i For reward, SNR i is the signal-to-noise ratio at the receiving end, SNR target is the target signal-to-noise ratio.

[0094] The module M4 includes: module M4.1: taking the initial attenuation value as the starting state, taking the starting state as the input of the execution neural network, and obtaining the output as the attenuation adjustment value, calculating the attenuation of each channel through the starting state and the attenuation adjustment value, and then transmitting and obtaining the result at the receiving end; finally obtaining the signal-to-noise ratio of each channel at the receiving end, calculating the corresponding reward through the signal-to-noise ratio obtained at the receiving end and the target signal-to-noise ratio, and updating the starting state to the next state; combining the starting state, attenuation adjustment value, next state and reward into a data pair, storing them in a data pool, and generating data in the data pool through continuous interaction between the execution neural network and the simulation transmission system; module M4.2: using the data in the data pool to train the critical neural network; module M4.3: using the data in the data pool and the critical neural network to train the execution neural network.

[0095] The attenuation of each channel is calculated through the initial state and attenuation adjustment value. The expression is:

[0096] A ch =A 0 +A Δ

[0097] Among them, A ch is the channel attenuation, A 0 is the initial state, A Δ is the attenuation adjustment value.

[0098] Those skilled in the art know that, in addition to implementing the system, device and its various modules provided by the present invention in a purely computer-readable program code, it is entirely possible to implement the same program in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers and embedded microcontrollers by logically programming the method steps. Therefore, the system, device and its various modules provided by the present invention can be considered as a hardware component, and the modules included therein for implementing various programs can also be considered as structures within the hardware component; the modules for implementing various functions can also be considered as both software programs for implementing the method and structures within the hardware component.

[0099] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A link power control method for a coherent optical wavelength division multiplexing transmission system, characterized in that: include: Step 1: Build a coherent optical wavelength division multiplexing transmission simulation system, including setting the length of the optical fiber in the link, the parameters of the optical amplifier, and setting the overall transmission distance; Step 2: Use the simulation environment to train the deep neural network and configure the transmit power of each channel in the wavelength division multiplexing system; Step 3: The initial state of the transmit power is used as input into the trained execution neural network to obtain the adjustment value of the transmit power of different channels. The channel transmit power is calculated based on the initial state and the transmit power adjustment value. Step 4: Use the simulation environment to train the deep neural network and configure the attenuation of each channel of the wavelength selective switch in the wavelength division multiplexing system; Step 5: The initial state of the wavelength selective switch is used as input into the trained execution neural network to obtain the adjustment value of the wavelength selective switch attenuation, and the channel transmission power is calculated based on the initial state and the attenuation adjustment value.

2. The link power control method of the coherent optical wavelength division multiplexing transmission system according to claim 1, characterized in that: The step 2 comprises: Step 2.1: Take the initial value of the transmission power as the starting state, take the starting state as the input of the execution neural network, and obtain the output as the adjustment value of the transmission power. Calculate the channel transmission power through the starting state and the transmission power adjustment value, and then transmit it to obtain the result at the receiving end; obtain the signal-to-noise ratio of each channel at the receiving end, calculate the corresponding reward through the signal-to-noise ratio obtained at the receiving end and the target signal-to-noise ratio, and update the starting state to the next state; collect the starting state, the transmission power adjustment value, the next state and the reward into a data pair, store them in the data pool, and generate the data in the data pool by continuously interacting with the execution neural network and the simulation transmission system; Step 2.2: Train the critical neural network with the data in the data pool, use the current state and execution action in the data pair as the input of the critical neural network, and obtain its output; subtract the output from the reward function and take the absolute value as the loss function of the critical neural network, and update the parameters of the critical neural network by the gradient descent method; Step 2.3: Train the execution neural network through the data in the data pool and the critical neural network, use the current state in the data pair as the input of the execution neural network to obtain the output of the execution neural network, use the output and the current state as the input of the critical neural network to obtain the output of the critical neural network; use the opposite number of the output of the critical neural network as the loss function, and update the parameters of the execution neural network through the gradient descent method.

3. The link power control method of the coherent optical wavelength division multiplexing transmission system according to claim 2, characterized in that: The channel transmit power is calculated by the initial state and the transmit power adjustment value. The expression is: P ch =P0+P Δ Among them, P ch is the channel transmission power, P0 is the initial state, P Δ is the power adjustment value; The corresponding reward is calculated by the signal-to-noise ratio obtained at the receiving end and the target signal-to-noise ratio. The expression is: R i =||SNR i -SNR target || 2 Among them, R i For reward, SNR i is the signal-to-noise ratio at the receiving end, SNR target is the target signal-to-noise ratio.

4. The link power control method of the coherent optical wavelength division multiplexing transmission system according to claim 2, characterized in that: The step 4 comprises: Step 4.1: Take the initial attenuation value as the starting state, take the starting state as the input of the execution neural network, and obtain the output as the attenuation adjustment value. Calculate the attenuation of each channel through the starting state and the attenuation adjustment value, then transmit and obtain the result at the receiving end; finally, obtain the signal-to-noise ratio of each channel at the receiving end, calculate the corresponding reward through the signal-to-noise ratio obtained at the receiving end and the target signal-to-noise ratio, and update the starting state to the next state; collect the starting state, attenuation adjustment value, next state and reward into a data pair, store them in the data pool, and generate data in the data pool through the continuous interaction between the execution neural network and the simulation transmission system; Step 4.2: Use the data in the data pool to train the critical neural network; Step 4.3: Use the data in the data pool and the critical neural network to train the executive neural network.

5. The link power control method of the coherent optical wavelength division multiplexing transmission system according to claim 4, characterized in that: The attenuation of each channel is calculated through the initial state and attenuation adjustment value. The expression is: TO ch =A0+A Δ Among them, A ch is the channel attenuation, A0 is the initial state, A Δ is the attenuation adjustment value.

6. A link power control system for a coherent optical wavelength division multiplexing transmission system, characterized in that: include: Module M1: Build a coherent optical wavelength division multiplexing transmission simulation system, including setting the length of the optical fiber in the link, the parameters of the optical amplifier, and setting the overall transmission distance; Module M2: Use the simulation environment to train the deep neural network and configure the transmit power of each channel in the wavelength division multiplexing system; Module M3: takes the initial state of the transmit power as input into the trained execution neural network, obtains the adjustment value of the transmit power of different channels, and calculates the channel transmit power based on the initial state and the transmit power adjustment value; Module M4: Use the simulation environment to train the deep neural network and configure the attenuation of each channel of the wavelength selective switch in the wavelength division multiplexing system; Module M5: The initial state of the wavelength selective switch is used as input into the trained execution neural network to obtain the adjustment value of the wavelength selective switch attenuation, and the channel transmission power is calculated based on the initial state and the attenuation adjustment value.

7. The link power control system of the coherent optical wavelength division multiplexing transmission system according to claim 6, characterized in that: The module M2 comprises: Module M2.1: Take the initial value of the transmission power as the starting state, take the starting state as the input of the execution neural network, and obtain the output as the adjustment value of the transmission power. Calculate the channel transmission power through the starting state and the transmission power adjustment value, and then transmit it to obtain the result at the receiving end; obtain the signal-to-noise ratio of each channel at the receiving end, calculate the corresponding reward through the signal-to-noise ratio obtained at the receiving end and the target signal-to-noise ratio, and update the starting state to the next state; collect the starting state, the transmission power adjustment value, the next state and the reward into a data pair, store them in the data pool, and generate the data in the data pool through the continuous interaction between the execution neural network and the simulation transmission system; Module M2.2: Train the critical neural network with the data in the data pool, use the current state and execution action in the data pair as the input of the critical neural network, and obtain its output; subtract the output from the reward function and take the absolute value as the loss function of the critical neural network, and update the parameters of the critical neural network through the gradient descent method; Module M2.3: Train the execution neural network through the data in the data pool and the critical neural network, use the current state in the data pair as the input of the execution neural network to obtain the output of the execution neural network, use the output and the current state as the input of the critical neural network to obtain the output of the critical neural network; use the opposite number of the output of the critical neural network as the loss function, and update the parameters of the execution neural network through the gradient descent method.

8. The link power control system of the coherent optical wavelength division multiplexing transmission system according to claim 7, characterized in that: The channel transmit power is calculated by the initial state and the transmit power adjustment value. The expression is: P ch =P0+P Δ Among them, P ch is the channel transmission power, P0 is the initial state, P Δ is the power adjustment value; The corresponding reward is calculated by the signal-to-noise ratio obtained at the receiving end and the target signal-to-noise ratio. The expression is: R i =||SNR i -SNR target || 2 Among them, R i For reward, SNR i is the signal-to-noise ratio at the receiving end, SNR target is the target signal-to-noise ratio.

9. The link power control system of the coherent optical wavelength division multiplexing transmission system according to claim 7, characterized in that: The module M4 comprises: Module M4.1: Take the initial attenuation value as the starting state, take the starting state as the input of the execution neural network, and obtain the output as the attenuation adjustment value. Calculate the attenuation of each channel through the starting state and the attenuation adjustment value, then transmit and obtain the result at the receiving end; finally, obtain the signal-to-noise ratio of each channel at the receiving end, calculate the corresponding reward through the signal-to-noise ratio obtained at the receiving end and the target signal-to-noise ratio, and update the starting state to the next state; collect the starting state, attenuation adjustment value, next state and reward into data pairs, store them in the data pool, and generate data in the data pool through the continuous interaction between the execution neural network and the simulation transmission system; Module M4.2: Use data from the data pool to train a critical neural network; Module M4.3: Use data from the data pool and critique neural networks to train and execute neural networks.

10. The link power control system of the coherent optical wavelength division multiplexing transmission system according to claim 9, characterized in that: The attenuation of each channel is calculated through the initial state and attenuation adjustment value. The expression is: TO ch =A0+A Δ Among them, A ch is the channel attenuation, A0 is the initial state, A Δ is the attenuation adjustment value.

Images