Feedback optical input power prediction method based on neural network
By using a neural network-based feedback optical input power prediction method, the problem of inaccurate optical input power prediction in on-chip optical interconnects is solved, achieving the effects of saving power and adapting to dynamic traffic changes while meeting communication reliability requirements.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- XIDIAN UNIV
- Filing Date
- 2023-05-25
- Publication Date
- 2026-04-28
AI Technical Summary
In on-chip optical interconnects, existing technologies struggle to accurately predict optical input power, leading to power waste and communication unreliability. In particular, it is difficult to establish an accurate mapping relationship between optical signal-to-noise ratio and optical input power in high-order crosstalk noise and dynamic environments.
A neural network-based feedback optical input power prediction method is adopted. The neural network is trained to predict the optical input power based on the optical signal-to-noise ratio, and the model error and network environment error are corrected. The sample set is updated in real time to adapt to dynamic traffic changes.
It achieves power saving while meeting communication reliability requirements, adapts to dynamic traffic changes, and improves the accuracy of optical input power prediction and communication reliability.
Smart Images
Figure CN116707645B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of integrated circuit and optoelectronic integration technology, and specifically relates to a feedback optical input power prediction method based on neural networks. Background Technology
[0002] As the integration density of integrated circuits increases, the shortcomings of on-chip networks in terms of bandwidth, latency, reusability, and scalability are becoming increasingly apparent. Photonic Interconnects on Chips (PICs) are considered a promising interconnect method due to their advantages of high bandwidth, low latency, and low power consumption; among these, research on insertion loss, crosstalk, and power consumption is crucial for the performance evaluation of PICs.
[0003] In related technologies, there are two methods for adjusting the optical input power in on-chip optical interconnects (OSIs). The first method divides the optical input power into several levels based on its magnitude and adjusts these levels according to factors such as link utilization and bandwidth. The disadvantages of this method are that the boundaries for these levels are difficult to determine, and it may lead to additional power waste. The second method establishes a loss or crosstalk calculation model and calculates the optical input power based on the receiver sensitivity or the optical signal-to-noise ratio (OSNR) in the OSI. The disadvantage of this method is that it is difficult to fully consider all losses or crosstalk in the on-chip optical network, resulting in inaccurate calculated optical input power. In on-chip optical interconnects, signal loss includes various categories such as cross-link loss, propagation loss, bending loss, and loss through microring resonators (MR). Thermal loss is affected by on-chip temperature, and even for a specific transmission path, signal loss can change dynamically. Signal crosstalk includes not only crosstalk caused by other signals but also self-crosstalk noise (which is affected by the signal's own power). Furthermore, crosstalk itself can generate crosstalk noise again, becoming second-order crosstalk. In the complex internal structure of on-chip optical routers, higher-order crosstalk noise is difficult to estimate. Establishing a mapping relationship between optical signal-to-noise ratio and optical input power by calculating signal loss and crosstalk is usually inaccurate. Therefore, it is urgent to improve the above-mentioned deficiencies in the existing technology. Summary of the Invention
[0004] To address the aforementioned problems in the existing technology, this invention provides a feedback-based optical input power prediction method based on neural networks. The technical problem to be solved by this invention is achieved through the following technical solution:
[0005] In a first aspect, the present invention provides a feedback-based optical input power prediction method based on a neural network, comprising:
[0006] Based on the signal transmission quality requirements, determine the required optical signal-to-noise ratio at the receiving end of the destination node;
[0007] The optical signal-to-noise ratio is input into a trained neural network to obtain the predicted optical input power; wherein, the preset neural network is trained according to a sample set to obtain a trained neural network;
[0008] The predicted optical input power is corrected by model error correction and network environment error correction to obtain the corrected optical input power.
[0009] Based on the corrected optical input power, the optical input power of the signal transmitted at the source node is controlled so that the optical input power of the signal transmitted at the source node is equal to the corrected optical input power. The signal is then transmitted through the on-chip optical network and reaches the receiver at the destination node.
[0010] Determine whether the power of the signal received at the destination node is greater than the sensitivity of the receiver at the destination node. If so, obtain the corresponding optical signal-to-noise ratio based on the power of the signal received at the destination node, and combine the optical input power and optical signal-to-noise ratio of the signal transmitted at the source node to form a new sample. Update the samples in the sample set using a sliding window update method to train the preset neural network online.
[0011] The beneficial effects of this invention are:
[0012] This invention provides a feedback-based optical input power prediction method based on a neural network. According to a given optical signal-to-noise ratio (OSN), the method uses a trained neural network to predict the optical input power feedback-wise. It then corrects for model errors and network environment errors in the predicted optical input power. Based on the corrected optical input power, the method controls the optical input power of the signal transmitted at the source node to be equal to the corrected optical input power. This signal is then transmitted through the on-chip optical network to the receiver at the destination node, where it is converted into an OSN. The optical input power and OSN of the signal transmitted at the source node are combined to form a new sample. A sliding window update method is used to update the sample set of the pre-set neural network, enabling online training of the pre-set neural network. On the one hand, this method corrects for model errors and network environment errors in the optical input power, saving power while meeting communication reliability requirements. On the other hand, real-time online training of the pre-set neural network allows it to adapt to dynamic traffic changes within the on-chip optical network.
[0013] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description
[0014] Figure 1 This is a flowchart of a feedback optical input power prediction method based on neural networks provided in an embodiment of the present invention;
[0015] Figure 2 This is a flowchart of the process of training a preset neural network provided in an embodiment of the present invention;
[0016] Figure 3 This is a schematic diagram of an updated sample set provided in an embodiment of the present invention. Detailed Implementation
[0017] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.
[0018] In addressing the problems in existing technologies, such as the difficulty in estimating high-order crosstalk noise within the complex internal structure of on-chip optical routers, and the inaccuracy in establishing the mapping relationship between optical signal-to-noise ratio (SNR) and optical input power by calculating signal loss and crosstalk, this invention provides a neural network-based feedback optical input power prediction method. This method predicts optical input power based on a given required SNR and corrects the prediction result based on the error, thereby saving power while meeting communication reliability requirements.
[0019] The main idea of this invention is to save optical input power while meeting the requirements of communication reliability, namely, the minimum optical signal-to-noise ratio (OSN). By determining the minimum OSN, and by obtaining the mapping relationship between OSN and optical input power, the optical input power can be predicted based on a given OSN. Neural networks possess excellent self-learning and adaptive capabilities, and can be used to realize the nonlinear mapping relationship between OSN and optical input power. After training the neural network, it can predict the required optical input power based on a given OSN. Furthermore, to address the issue that the predicted optical input power may not be accurate, potentially leading to communication unreliability, error correction is applied to the predicted optical input power to improve communication reliability.
[0020] Please see Figure 1 As shown, Figure 1 This is a flowchart of a neural network-based feedback optical input power prediction method provided in an embodiment of the present invention. The neural network-based feedback optical input power prediction method provided by the present invention includes:
[0021] S101. Determine the required optical signal-to-noise ratio (OSNR) at the destination node receiver based on the signal transmission quality requirements.
[0022] Specifically, the Optical Network-on-Chip (ONoC) provided in this embodiment transmits an optical signal at the source node. The optical signal is guided into an optical fiber and coupled into a waveguide. The resonant wavelength of a microring resonator (MR) is modulated with the wavelength of the optical signal. The modulated optical signal is then transmitted along the waveguide to the receiver at the destination node. The optical signal is then filtered, converted by a photodetector, amplified by a transimpedance amplifier, and the modulated data at the wavelength of the optical signal is extracted to obtain the optical signal-to-noise ratio (OSNR). During this process, some losses occur, such as signal loss L (in dB). Furthermore, crosstalk also occurs, such as signal crosstalk noise power P. crosstalk (Unit: mW); These losses and crosstalk noise power affect the optical input power of the optical signal, further affecting the optical signal-to-noise ratio (OSNR) at the receiver; while the OSNR of the signal in the on-chip optical interconnect represents the signal transmission quality. In practice, it is desirable to achieve a lower optical input power P input Achieving a high optical signal-to-noise ratio (OSNR) ensures signal transmission quality while saving power.
[0023] Therefore, this embodiment studies the relationship between optical signal-to-noise ratio (OSNR) and optical input power, and its expression is:
[0024]
[0025] According to the above formula, there is a single mapping relationship between the optical signal-to-noise ratio and the optical input power. By adjusting the optical input power P... input This allows for achieving an optical signal-to-noise ratio that meets transmission reliability requirements with lower optical input power.
[0026] Meeting transmission reliability requires a minimum optical signal-to-noise ratio (OSNR), meaning the required OSNR must be greater than the minimum OSNR. threshold OSNR here threshold To meet the optical signal-to-noise ratio threshold for reliable transmission.
[0027] It should be noted that in this embodiment, a feedback approach is adopted, and the optical input power is predicted using the prediction method provided by this invention, given the required optical signal-to-noise ratio (OSNR) at the receiving end.
[0028] S102. Input the given required optical signal-to-noise ratio (OSNR) into the trained neural network to obtain the predicted optical input power; wherein, the preset neural network is trained according to the sample set to obtain the trained neural network.
[0029] Specifically, please see Figure 2 As shown, Figure 2 This is a flowchart illustrating the process of training a preset neural network according to an embodiment of the present invention. In this embodiment, the trained neural network is first obtained, and the specific process is as follows:
[0030] S1021. Determine the adjustment range P of the optical input power. min ~P max Among them, P min P represents the minimum optical input power. max This represents the maximum value of the optical input power.
[0031] Specifically, among multiple communication pairs, based on the worst-case optical signal-to-noise ratio (OSNR)... wc and the receiver sensitivity P at the destination node sensitivity Determine the adjustment range P of the optical input power. min ~P max Among them, the worst-case optical signal-to-noise ratio (OSNR) wc The minimum optical signal-to-noise ratio generated by all communication tasks during transmission is given by L. wc (Unit: dB, and L) wc >0), the corresponding signal crosstalk is P. crosstalk,wc (Unit: mW) The receiver sensitivity at the destination node is determined by the performance of the optical device itself, and a preset optical signal-to-noise ratio (OSNR) threshold is set to ensure transmission reliability. threshold (Unit: dB), which is the signal-to-noise ratio threshold that satisfies the maximum bit error rate.
[0032] Among them, the maximum value of the optical input power P max The expression is:
[0033]
[0034] Minimum optical input power P min The expression is:
[0035] P min =P sensitivity .
[0036] S1022. Based on the adjustment range of the optical input power, obtain the optical input power P of the transmitted signal at the source node in the t-th time slot. input (t), whose expression is:
[0037]
[0038] Where m is the total number of time slots.
[0039] In this embodiment, the signal transmitted at the source node has an optical input power P.input (t) During transmission, signal loss and crosstalk occur, resulting in low signal power received at the target node. In extreme cases, the signal may be undetectable by the photodetector. In this case, the minimum optical input power P needs to be updated. min Specifically:
[0040] Over K consecutive time slots, the optical input power P of the signal transmitted at the source node is... input (t) is the same, and its expression is:
[0041] P input (t-K+1)=P input (t-K+2)=…=P input (t);
[0042] The signal power received at the destination node is P. signal (t), when P signal (t) is greater than P sensitivity When the signal is detected, the optical signal-to-noise ratio osnr(t) is obtained, and when P signal (t) is less than P sensitivity At this point, the signal is not detected, meaning the optical signal-to-noise ratio osnr(t) is missing, and the minimum value P of the updated optical input power is used. min , making P min =P input (t).
[0043] It should be noted that the minimum value of the optical input power is continuously updated during the optical input power prediction process.
[0044] S1023. Obtain the optical input power of m time slots and the optical signal-to-noise ratio corresponding to the optical input power, and construct a sample set, the expression of which is:
[0045]
[0046] The sample set contains m samples.
[0047] Specifically, the optical input power P at the source node is continuously adjusted. input (t), obtain the corresponding optical signal-to-noise ratio osnr(t) at the destination node, and construct the sample set {osnr,P input The larger the number of samples in the sample set, the more samples are selected, and the more accurate the trained neural network model will be. However, the latency and power costs caused by collecting the sample set are also higher.
[0048] S1023, Based on the sample set {osnr,P} input The system trains a pre-defined neural network, taking the optical signal-to-noise ratio as input and outputting the trained and predicted optical input power.
[0049] Also includes:
[0050] Obtain the test sample set;
[0051] The optical signal-to-noise ratio in the test sample set is input into the trained neural network to obtain the predicted optical input power.
[0052] The mean square error between the predicted optical input power and the actual optical input power in the test sample set is used as the loss function, and its expression is:
[0053]
[0054] Where MAE is the loss function, P predict (t) represents the predicted optical input power, P input (t) represents the actual optical input power in the test sample set, and n represents the number of samples in the test sample set;
[0055] By calculating the loss function, the weights and biases in the preset neural network are updated, and the trained neural network is also updated.
[0056] S103. Correct the predicted optical input power by model error correction and network environment error correction, and obtain the corrected optical input power.
[0057] Specifically, in this embodiment, the predicted optical input power is corrected using model error parameters. The process of obtaining the model error parameters includes:
[0058] Obtain the test sample set;
[0059] The optical signal-to-noise ratio in the test sample set is input into the trained neural network to obtain the predicted optical input power.
[0060] Based on the actual optical input power of the transmitted signal at the source node corresponding to the optical signal-to-noise ratio and the predicted optical input power, the model error parameter error1 is obtained, and its expression is:
[0061]
[0062] Where n is the number of samples in the test sample set, P input (t) represents the actual optical input power of the transmitted signal at the source node, P predict (t) represents the predicted optical input power during testing.
[0063] It should be noted that in this embodiment, the number of samples in the test sample set is n = 0.1 × m, and the samples in the test sample set are processed to remove incomplete values.
[0064] The predicted optical input power is corrected using network environment error parameters. The process of obtaining network environment error parameters includes:
[0065] In K consecutive time slots, the source node emits the same optical input power P. input (t), whose expression is:
[0066] P input (t-K+1)=P input (t-K+2)=…=P input (t);
[0067] In the i-th consecutive time slot, obtain the maximum optical signal-to-noise ratio (PSNR) maxosnr corresponding to the transmitted signal at the source node. K_i and minimum optical signal-to-noise ratio minosnr K_i Obtain the difference Δosnr between the maximum and minimum optical signal-to-noise ratios. K_i and obtain their mean.
[0068] Based on the difference Δosnr K_i and mean The network environment error parameter error2 is obtained, and its expression is:
[0069]
[0070] Among them, 0 <i≤m / K。
[0071] It should be noted that the model error is caused by the preset neural network, while the network environment error refers to the error caused by the on-chip optical network environment.
[0072] S104. Based on the corrected optical input power, control the optical input power of the signal transmitted at the source node to make the optical input power of the signal transmitted at the source node equal to the corrected optical input power. The signal is then transmitted through the on-chip optical network to the receiver at the destination node.
[0073] Specifically, in this embodiment, at a certain time slot t, ignoring modeling errors and on-chip network environment errors, a given required signal-to-noise ratio (OSNR) is input into a trained neural network, and the predicted optical input power is output. Based on the trained neural network, without considering changes in the on-chip network environment, the relationship between the given required OSNR and the optical input power is modeled, i.e., f(·). Assuming the influence of the on-chip network environment, including flow load and wavelength allocation, is g(·), then the optical input power P is obtained from the given required optical OSNR. input The mapping expression is:
[0074] P input= g(f(OSNR),t);
[0075] In the above formula, t represents the time slot. This formula indicates the change from the required optical signal-to-noise ratio (OSNR) to the optical input power P. input The relationship between them is influenced by both the trained neural network model and the dynamic on-chip optical network environment.
[0076] Based on the above reasoning, it can be concluded that even with the same optical signal-to-noise ratio (OSNR), different optical input powers (P) may be generated in different time slots. input Even in harsh environments of on-chip optical networks, the optical input power predicted based on the minimum optical signal-to-noise ratio may not meet the minimum optical signal-to-noise ratio requirement of the target node receiver. Therefore, the error between the predicted optical input power and the actual optical input power is divided into two parts. One part is caused by the modeling accuracy of the predictive controller, called the model error, denoted as error1. The other part is caused by the dynamic network environment, called the network environment error, denoted as error2.
[0077] Therefore, error correction includes model error correction and network environment error correction.
[0078] First, model error correction is performed, and the optical input power after model error correction is P′. input (t), because to meet the requirements of communication reliability, the optical input power P′ after model error correction is (t), input (t) should be slightly greater than the predicted optical input power P. predict (t); Optical input power P′ after model error correction input The expression for (t) is:
[0079]
[0080] Secondly, after model error correction, network environment error correction is performed, and the optical input power P for network environment error correction is... input The expression for (t) is:
[0081]
[0082] The optical input power P obtained after model error correction and network environment error correction is... input (t), the source node can control the input power of the optical signal, which is used to transmit on-chip optical signals.
[0083] S105. Determine whether the power of the received signal at the destination node is greater than the sensitivity of the receiver at the destination node. If so, obtain the corresponding optical signal-to-noise ratio based on the power of the received signal at the destination node, and form a new sample by combining the optical input power and optical signal-to-noise ratio of the transmitted signal at the source node. Update the samples in the sample set using a sliding window update method to train the preset neural network online.
[0084] Specifically, in this embodiment, considering the dynamic changes in on-chip bandwidth, in order to adapt the preset neural network model to the dynamic changes in on-chip bandwidth, the sample set is updated and the preset neural network is trained online based on the training of the preset neural network.
[0085] Please see Figure 3 As shown, Figure 3 This is a schematic diagram of an updated sample set provided in an embodiment of the present invention. In time slot t, it is determined whether the power of the signal received at the destination node is greater than the sensitivity of the receiver at the destination node. If so, the optical signal-to-noise ratio (SNR) corresponding to time slot t is obtained based on the power of the signal received at the destination node. A new sample is then formed by combining the optical input power of the signal transmitted at the source node and the SNR. The sample set is updated using a sliding window update method. Figure 3 As shown, the sample set stores m samples from time slot tm to time slot t-1. When the sample (osnr(t), P) of time slot t is obtained... input After (t)), the new sample (osnr(t),P input (t)) is put into the sample set, and the sample (osnr(tm),P) of the time slot tm is added. input (tm)) Remove the sample set. In the preset online training process of the neural network, osnr(t) is the actual optical signal-to-noise ratio obtained during the transmission process. In step S104, when the trained neural network obtains the predicted optical input power, the input of the trained model is the given optical signal-to-noise ratio OSNR, which theoretically guarantees that the optical signal-to-noise ratio threshold value that meets the reliability requirements is met.
[0086] In summary, the present invention provides a feedback-based optical input power prediction method based on a neural network. According to a given optical signal-to-noise ratio (OSN), the method predicts the optical input power feedback-based through a trained neural network. It then corrects for model errors and network environment errors in the predicted optical input power. Based on the corrected optical input power, the method controls the optical input rate of the transmitted signal at the source node to be equal to the corrected optical input power. This signal is then transmitted through the on-chip optical network to the receiver at the destination node, where it is converted into an OSN. The optical input power and OSN of the transmitted signal at the source node are combined to form a new sample. A sliding window update method is used to update the sample set of the pre-set neural network, enabling online training of the pre-set neural network. On the one hand, the correction of model errors and network environment errors in the optical input power saves power while meeting communication reliability requirements. On the other hand, real-time online training of the pre-set neural network allows it to adapt to dynamic traffic changes within the on-chip optical network.
[0087] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations are intended to cover non-exclusive inclusion, such that an article or device comprising a list of elements includes not only those elements but also other elements not expressly listed. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the article or device comprising said element. Terms such as "connected" or "linked" are not limited to physical or mechanical connections but can include electrical connections, whether direct or indirect. The orientations or positional relationships indicated by terms such as "upper," "lower," "left," and "right" are based on the orientations or positional relationships shown in the accompanying drawings and are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as limiting the invention.
[0088] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features or characteristics described may be combined in any suitable manner in one or more embodiments or examples. In addition, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.
[0089] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.
Claims
1. A feedback-based optical input power prediction method based on neural networks, characterized in that, include: Based on the signal transmission quality requirements, determine the required optical signal-to-noise ratio at the receiving end of the destination node; The optical signal-to-noise ratio is input into a trained neural network to obtain the predicted optical input power; wherein, a preset neural network is trained based on a sample set to obtain a trained neural network. The predicted optical input power is corrected for model error and network environment error to obtain the corrected optical input power. Based on the corrected optical input power, the optical input power of the signal transmitted at the source node is controlled so that the optical input power of the signal transmitted at the source node is equal to the corrected optical input power. The signal is then transmitted through the on-chip optical network to the receiver at the destination node. Determine whether the power of the signal received at the destination node is greater than the sensitivity of the receiver at the destination node. If so, obtain the corresponding optical signal-to-noise ratio based on the power of the signal received at the destination node, and combine the optical input power of the signal transmitted at the source node and the optical signal-to-noise ratio to form a new sample. Update the samples in the sample set using a sliding window update method to train the preset neural network online.
2. The neural network-based feedback optical input power prediction method according to claim 1, characterized in that, The step of training a preset neural network based on a sample set to obtain a trained neural network includes: Determine the adjustment range of optical input power. ;in, This represents the minimum optical input power. This represents the maximum value of the optical input power. Based on the adjustment range of the optical input power, obtain the first... The optical input power of the transmitter at the source node of each time slot Its expression is: ； in, The total number of time slots; Get The optical input power of each time slot is calculated, and the optical signal-to-noise ratio corresponding to the optical input power is obtained to construct a sample set, the expression of which is: ； The samples in the sample set include indivual; The optical signal-to-noise ratio in the sample set is input into the preset neural network, the preset neural network is trained, and the trained and predicted optical input power is output.
3. The feedback-based optical input power prediction method based on neural networks according to claim 2, characterized in that, The maximum value of the optical input power The expression is: ； The minimum value of the optical input power The expression is: ； in, To meet the optical signal-to-noise ratio threshold for transmission reliability, Minimum optical signal-to-noise ratio Corresponding signal crosstalk, Minimum optical signal-to-noise ratio The corresponding signal loss, This represents the receiver sensitivity at the target node.
4. The neural network-based feedback optical input power prediction method according to claim 3, characterized in that, Also includes: Update the minimum value of the optical input power. ; In continuous K In each time slot, the optical input power of the signal transmitted at the source node is... The same, its expression is: ； The power of the signal received at the destination node is ,when Greater than When the signal is detected, the optical signal-to-noise ratio is obtained. Less than At that time, the signal was not detected, making Update the minimum value of the optical input power. .
5. The neural network-based feedback optical input power prediction method according to claim 2, characterized in that, Also includes: Obtain the test sample set; The optical signal-to-noise ratio in the test sample set is input into the trained neural network to obtain the test-predicted optical input power. The mean square error between the predicted optical input power and the actual optical input power in the test sample set is used as the loss function, and its expression is: ； in, For loss function, To test the predicted optical input power, To test the actual optical input power in the sample set, The number of samples in the test sample set; By calculating the loss function, the weights and biases in the preset neural network are updated, and the trained neural network is also updated.
6. The neural network-based feedback optical input power prediction method according to claim 1, characterized in that, The predicted optical input power is corrected using model error parameters. The process of obtaining the model error parameters includes: Obtain the test sample set; The optical signal-to-noise ratio in the test sample set is input into the trained neural network to obtain the test-predicted optical input power. Based on the actual optical input power of the transmitted signal at the source node corresponding to the optical signal-to-noise ratio and the predicted optical input power, the model error parameters are obtained. Its expression is: ； in, The number of samples in the test sample set, This represents the actual optical input power of the signal transmitted at the source node. To test the predicted optical input power, This is the index of the time slot.
7. The neural network-based feedback optical input power prediction method according to claim 1, characterized in that, The predicted optical input power is corrected using network environment error parameters, and the process of obtaining the network environment error parameters includes: In continuous In each time slot, the optical input power of the signal transmitted at the source node is... The same, its expression is: ； In the In a continuous time slot, Obtain the maximum optical signal-to-noise ratio corresponding to the transmitted signal at the source node. and minimum optical signal-to-noise ratio Obtain the difference between the maximum and minimum optical signal-to-noise ratio. and obtain their mean. ; Based on the difference and mean Obtain the network environment error parameters Its expression is: ； in, , For the index of the time slot, This represents the total number of time slots.
8. The feedback-based optical input power prediction method based on neural networks according to claim 1, characterized in that, Light input power corrected for model errors The expression is: ； in, These are the model error parameters. For the predicted optical input power, This is the index of the time slot.
9. The feedback-based optical input power prediction method based on neural networks according to claim 1, characterized in that, Optical input power corrected for network environment errors The expression is: ； in, For network environment error parameters, The optical input power after model error correction. This is the index of the time slot.
10. The neural network-based feedback optical input power prediction method according to claim 1, characterized in that, The optical input power of the signal transmitted at the source node and the optical signal-to-noise ratio constitute a new sample. The samples in the sample set are updated using a sliding window update method, including: exist In the time slot, determine whether the power of the signal received at the destination node is greater than the sensitivity of the receiver at the destination node. If so, obtain the signal based on the power of the signal received at the destination node. The optical signal-to-noise ratio corresponding to the time slot is calculated, and the optical input power of the transmitted signal at the source node and the optical signal-to-noise ratio are combined to form a new sample. Then, the sample is set together. The sample corresponding to the time slot is removed, and the new sample is moved into the sample set as... The sample set of time slots, This represents the total number of time slots.