Intelligent optical fiber wiring collaborative scheduling method, device, equipment and medium

Abstract

The present application provides a method, device, equipment and medium for collaborative scheduling of intelligent fiber optic wiring, which belongs to the field of optical fiber communication technology. It trains a global LSTM model based on federated learning, obtains global model parameters and carries them into the local LSTM model of each intelligent fiber optic wiring robot to evaluate the health of each intelligent fiber optic wiring robot; abstracts the fiber cores of different intelligent fiber optic wiring robots into a logical resource pool and calculates their dynamic weights; determines the optimal resource range based on the dynamic weights, and determines the scheduling path that meets business requirements from the optimal resource range based on the search algorithm and reinforcement learning algorithm. Thus, it integrates federated learning, dynamic weight scheduling and reinforcement learning to build a three-layer collaborative architecture, which can automatically make resource allocation and path selection decisions according to different business needs, reduce manual intervention, reduce human errors, and realize intelligent operation and maintenance management.

Description

Technical Field

[0001] The present application relates to the field of optical fiber communication technology, and more specifically, to an intelligent optical fiber wiring collaborative scheduling method, device, equipment and medium. Background Art

[0002] With the development of technologies like 5G and the Industrial Internet, the number of fiber cores in computer rooms has skyrocketed (a single room can contain thousands of cores). Traditional manual O&M is inefficient, prone to operational errors, and the limited capacity of single-machine automation equipment makes it difficult to meet the demands of large-scale networking. Existing technologies urgently need to overcome bottlenecks such as rigid resource scheduling and insufficient multi-machine coordination to achieve intelligent, highly reliable, and dynamic O&M. Currently, fiber optic network O&M relies primarily on manual operations or single-machine automation equipment. Manual O&M requires technicians to plug and unplug patch cables on-site, which is time-consuming, prone to errors, and unsuitable for rapid service provisioning and restoration. While single-machine robots can operate autonomously, they are limited in capacity and lack multi-machine coordination. Summary of the Invention

[0003] In view of this, the purpose of this application is to provide an intelligent fiber optic distribution collaborative scheduling method, device, equipment and medium, which can realize dynamic scheduling of fiber optic resources within the cluster through a three-layer collaborative architecture of federated learning prediction-dynamic weight scheduling-reinforcement learning decision.

[0004] An embodiment of the present application provides a collaborative scheduling method for intelligent fiber optic wiring, which is applied to a cluster composed of multiple intelligent fiber optic wiring robots. The method includes the following steps:

[0005] Based on federated learning, a global LSTM model for predicting the health status of intelligent fiber optic wiring robots is trained to obtain global model parameters. These global model parameters are then loaded into the local LSTM model of each intelligent fiber optic wiring robot to evaluate the health of each intelligent fiber optic wiring robot.

[0006] The fiber cores of different intelligent fiber optic patching robots are abstracted into logical resource pools, and their dynamic weights are calculated based on the load of each logical resource pool and the health of the intelligent fiber optic patching robots contained therein;

[0007] An optimal resource range is determined based on the dynamic weight, and a scheduling path that meets business requirements is determined from the optimal resource range based on a search algorithm and a reinforcement learning algorithm.

[0008] In some embodiments, the method further comprises the following steps:

[0009] When changes in the network, equipment, or link status are detected, or when fluctuations in the volume of service requests exceed a set threshold, the dynamic weights are recalculated and the scheduling path is determined.

[0010] In some embodiments, the method of training a global LSTM model for predicting the health status of an intelligent fiber optic wiring robot based on federated learning to obtain global model parameters includes the following steps:

[0011] Initializing the global LSTM model and sending the initialized global model parameters to the local LSTM models of each intelligent fiber optic wiring robot;

[0012] Preprocessing the device data collected by the intelligent fiber optic wiring robot to construct time series data of different modalities, and training a local LSTM model based on the time series data;

[0013] The local training parameters obtained by each intelligent fiber optic wiring robot are used to perform parameter alignment compensation and parameter aggregation strategy on the global LSTM model to obtain the updated global model parameters.

[0014] In some embodiments, the device data includes optical power loss, the number of plugging and unplugging times of the optical fiber connector, the mileage of the lead screw, and environmental data. Preprocessing the device data collected by the intelligent optical fiber distribution robot to construct time series data of different modalities includes the following steps:

[0015] The number of plug-in / plug-out times within each window is counted based on a sliding window mechanism, and is divided into low-frequency plug-in / plug-out scenarios and high-frequency plug-in / plug-out scenarios according to a preset threshold value. In the low-frequency plug-in / plug-out scenario, the optical power loss is normalized using extreme value; in the high-frequency plug-in / plug-out scenario, the optical power loss is normalized using standard deviation.

[0016] The cumulative fatigue is calculated based on the number of plug-in and pull-out times, the screw mileage and the time decay effect, which is used to measure the cumulative stress and loss that the equipment is subjected to during use;

[0017] Obtaining the timestamp of each plugging and unplugging operation, and converting it into a first time code and a second time code according to the periodicity of plugging and unplugging of the intelligent optical fiber distribution robot;

[0018] The normalized optical power loss, cumulative fatigue, number of plug-in and unplugging times, first time code, second time code, and environmental data are aligned by time step to form a two-dimensional input tensor as time series data of different modalities.

[0019] In some embodiments, determining the optimal resource range based on the dynamic weight includes the following steps:

[0020] Determining the priority of the logical resource pool according to the size of the dynamic weight, and determining the priority of each intelligent optical fiber distribution robot in the logical resource pool according to the health of the intelligent optical fiber distribution robot and the number of idle fiber cores;

[0021] A set number of logical resource pools and the intelligent fiber optic wiring robots contained therein are selected in order of priority as the optimal resource range.

[0022] In some embodiments, determining a scheduling path that meets business requirements from the optimal resource range based on a search algorithm and a reinforcement learning algorithm includes the following steps:

[0023] A search algorithm is used based on network topology information to calculate a set of feasible paths from a source node to a destination node within the optimal resource range. If no path that meets the service requirements exists, the selected logical resource pool and the number of intelligent fiber optic wiring robots contained therein are expanded in order of priority, and the optimal resource range is re-determined until a path that meets the service requirements is found.

[0024] Obtain the current state of the optical fiber network and use a greedy strategy to select a path from the set of feasible paths as an action to be executed; wherein, if the generated random number is less than the set exploration rate, a path is randomly selected from the set of feasible paths as an action to be executed; if the generated random number is greater than the set exploration rate, the path with the largest expected cumulative reward is selected from the set of feasible paths as an action to be executed;

[0025] Calculate the immediate reward after the action is executed based on the set reward strategy, and calculate the expected cumulative reward after the action is executed based on the immediate reward, the set learning rate and discount factor, the expected cumulative reward when the action is not executed, and the maximum expected cumulative reward among all feasible paths;

[0026] Iteratively optimize the path based on the calculated expected value of cumulative rewards to determine the scheduling path that meets business requirements.

[0027] In some embodiments, the search algorithm adopts the A-star algorithm, and the status of the optical fiber network includes network topology information, dynamic weights of the logical resource pool, and the health of the intelligent optical fiber distribution robot; the network topology information includes the connection relationship between nodes, link length, and optical power attenuation.

[0028] In some embodiments, an intelligent fiber optic wiring collaborative scheduling device is further provided, which is applied to a cluster composed of multiple intelligent fiber optic wiring robots, and the device includes:

[0029] A federated learning health prediction module is used to train a global LSTM model for predicting the health status of intelligent fiber optic wiring robots based on federated learning, obtain global model parameters, and implement the global model parameters into the local LSTM model of each intelligent fiber optic wiring robot to evaluate the health of each intelligent fiber optic wiring robot;

[0030] A dynamic weight partitioning module is used to abstract the fiber cores of different intelligent fiber optic patching robots into logical resource pools, and calculate the dynamic weight of each logical resource pool according to the load of the logical resource pool and the health of the intelligent fiber optic patching robots contained therein;

[0031] A reinforcement learning path matching module is used to determine an optimal resource range based on the dynamic weight, and to determine a scheduling path that meets business requirements from the optimal resource range based on a search algorithm and a reinforcement learning algorithm.

[0032] In some embodiments, an electronic device is also provided, comprising: a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor and the memory communicate through the bus, and when the machine-readable instructions are executed by the processor, the steps of any one of the above-mentioned intelligent fiber optic distribution collaborative scheduling methods are performed.

[0033] In some embodiments, a computer-readable storage medium is further provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of any one of the above-mentioned intelligent fiber optic distribution collaborative scheduling methods are executed.

[0034] The intelligent fiber optic wiring collaborative scheduling method, device, equipment and medium described in this application are applied to a cluster composed of multiple intelligent fiber optic wiring robots. Based on federated learning, a global LSTM model for predicting the health status of the intelligent fiber optic wiring robot is trained to obtain global model parameters, and the global model parameters are loaded into the local LSTM model of each intelligent fiber optic wiring robot to evaluate the health of each intelligent fiber optic wiring robot; the fiber cores of different intelligent fiber optic wiring robots are abstracted into logical resource pools, and their dynamic weights are calculated based on the load of each logical resource pool and the health of the intelligent fiber optic wiring robots contained therein; the optimal resource range is determined based on the dynamic weights, and a scheduling path that meets business requirements is determined from the optimal resource range based on a search algorithm and a reinforcement learning algorithm. Thus, federated learning, dynamic weight scheduling and reinforcement learning are integrated to construct a three-layer collaborative architecture that can automatically make resource allocation and path selection decisions according to different business needs, reduce manual intervention, reduce human errors, and realize intelligent operation and maintenance management. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0036] Figure 1 A flowchart of the intelligent optical fiber distribution collaborative scheduling method according to an embodiment of the present application is shown;

[0037] Figure 2 A flowchart of obtaining global model parameters by training a global LSTM model for predicting the health status of an intelligent fiber optic wiring robot based on federated learning according to an embodiment of the present application is shown;

[0038] Figure 3 A flowchart of determining a scheduling path that meets business requirements from the optimal resource range based on a search algorithm and a reinforcement learning algorithm according to an embodiment of the present application is shown;

[0039] Figure 4 The figure shows a schematic diagram of the structure of the intelligent optical fiber distribution collaborative scheduling device according to an embodiment of the present application;

[0040] Figure 5 A schematic structural diagram of an electronic device according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It should be understood that the drawings in the present application only serve the purpose of illustration and description and are not used to limit the scope of protection of the present application. In addition, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate the operations implemented according to some embodiments of the present application. It should be understood that the operations of the flowcharts can be implemented out of sequence, and steps without logical context can be reversed or implemented simultaneously. In addition, those skilled in the art, under the guidance of the contents of this application, can add one or more other operations to the flowchart, or remove one or more operations from the flowchart.

[0042] In addition, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present application.

[0043] It should be noted that the term "comprising" will be used in the embodiments of the present application to indicate the existence of the features declared thereafter, but does not exclude the addition of other features.

[0044] In view of the technical problems raised by the background technology, the present application provides an intelligent fiber optic distribution collaborative scheduling method, device, equipment and medium, which can realize dynamic scheduling of fiber optic resources within the cluster through a three-layer collaborative architecture of federated learning prediction-dynamic weight scheduling-reinforcement learning decision.

[0045] See the instructions attached Figure 1 The present application provides an intelligent fiber optic wiring collaborative scheduling method, which is applied to a cluster composed of multiple intelligent fiber optic wiring robots, including the following steps:

[0046] S1. Based on federated learning, a global LSTM model for predicting the health status of the intelligent fiber optic wiring robot is trained to obtain global model parameters. The global model parameters are then loaded into the local LSTM model of each intelligent fiber optic wiring robot to evaluate the health of each intelligent fiber optic wiring robot.

[0047] S2. Abstract the fiber cores of different intelligent fiber optic patching robots into logical resource pools, and calculate the dynamic weight of each logical resource pool based on the load of the logical resource pool and the health of the intelligent fiber optic patching robots contained therein;

[0048] S3. Determine an optimal resource range based on the dynamic weight, and determine a scheduling path that meets business requirements from the optimal resource range based on a search algorithm and a reinforcement learning algorithm.

[0049] In step S1, the distributed federated learning framework is mainly used to realize the distributed training of local data and global model aggregation of the intelligent optical fiber wiring robot, so as to improve the adaptability and accuracy of the health monitoring of the intelligent optical fiber wiring robot.

[0050] See the instructions attached Figure 2 The global LSTM model for predicting the health status of the intelligent optical fiber distribution robot based on federated learning training is used to obtain global model parameters, including the following steps:

[0051] S101, initializing the global LSTM model, and sending the initialized global model parameters to the local LSTM model of each intelligent optical fiber distribution robot;

[0052] S102: Preprocessing the device data collected by the intelligent optical fiber distribution robot to construct time series data of different modalities, and training a local LSTM model based on the time series data;

[0053] S103: Utilize the local training parameters obtained by each intelligent fiber optic wiring robot to perform parameter alignment compensation and parameter aggregation strategy on the global LSTM model to obtain updated global model parameters.

[0054] Specifically, in step S101, the global LSTM model constructed for predicting the health status of the intelligent fiber optic wiring robot can effectively process time series data and capture the changing patterns of connector plugging and unplugging frequency, cumulative fatigue, etc. over time. After completing the model initialization, global model parameters such as LSTM layer weights, fully connected layer weights, LSTM layer biases and fully connected layer biases are sent to each intelligent fiber optic wiring robot.

[0055] In step S102, after receiving the initialized global model parameters, each intelligent fiber optic patching robot trains its local LSTM model based on locally collected and preprocessed time series data. The device data collected by the intelligent fiber optic patching robot includes optical power loss of the fiber channel, the number of fiber connector plugging and unplugging cycles, temperature, humidity, vibration intensity (time series data), screw mileage, and other optical, mechanical, and environmental data. Data preprocessing is performed as follows:

[0056] (1) The number of plug-in and unplug-out times in each window is counted based on a sliding window mechanism, and is divided into a low-frequency plug-in and unplug-out scenario and a high-frequency plug-in and unplug-out scenario according to a preset threshold value; wherein, in the low-frequency plug-in and unplug-out scenario, the optical power loss is normalized by extreme value; in the high-frequency plug-in and unplug-out scenario, the optical power loss is normalized by standard deviation.

[0057] Specifically, optical power loss is dynamically normalized by plugging and unplugging frequency segmentation. This is because, in low-frequency plugging and unplugging scenarios, there are fewer plugging and unplugging operations, resulting in relatively stable optical power loss data and a high noise content. In high-frequency plugging and unplugging scenarios, plugging and unplugging are frequent, leading to dramatic fluctuations in optical power loss (e.g., a sudden surge in loss upon plugging and unplugging). The optical power loss data is then divided into multiple subsets based on plugging and unplugging frequency ranges (e.g., low and high). Normalization parameters (mean, standard deviation, or extreme value) are independently calculated for each subset, adapting to differences in data distribution at different frequencies. By segmenting the plugging and unplugging frequency, "high-frequency amplification and low-frequency smoothing" is achieved, directly linking the fluctuation characteristics of optical power loss to the plugging and unplugging frequency, improving the model's responsiveness to extreme scenarios.

[0058] In specific implementations, a sliding window mechanism can be used to define short-term time windows Tshort (in days) and count the number of plug-ins F within each window. Because both low-frequency and high-frequency plug-ins exist, statistics on the number of plug-ins within a long-term time window Tlong are added in addition to the plug-in count within the time window. Tlong is composed of multiple consecutive time windows Tshort. The average plug-in frequency within the Tlong time window is evaluated using an exponentially weighted moving average (EWMA).

[0059]

[0060] in, For the forgetting factor, The larger the value, the smaller the impact of historical data. Ft represents the number of plug-ins and unplugs in the time window t; In one embodiment, the average plug-in and plug-out frequency in the Tlong time window is divided into low frequency and high frequency by a preset threshold value.

[0061] In a low-frequency plug-in / plug-out scenario, extreme value normalization is performed on the optical power loss.

[0062] Take the maximum and minimum values ​​within the Tlong time window and calculate the normalized optical power loss data:

[0063]

[0064] Among them, P is the optical power loss data collected by a certain intelligent optical fiber wiring robot. is the calculated normalized data, is the maximum value of P in the Tlong time window, is the minimum value of P in the Tlong time window, S is the scaling factor (range: 0.7~0.9, default: 0.8), and O is the offset (default 0.1). The purpose of setting S and O is to compress the normalized range to the middle interval and ensure that the minimum value is not zero, ensuring the physical meaning of the value.

[0065] In a high-frequency plug-in / plug-out scenario, the optical power loss is normalized using standard deviation.

[0066]

[0067] in, is the average optical power loss in the window, is the standard deviation of optical power loss within the window.

[0068] (2) The accumulated fatigue factor AEF is calculated based on the number of plugging and unplugging times, the mileage of the screw and the time decay effect to measure the accumulated stress and loss that the equipment is subjected to during use.

[0069]

[0070] in, For plug-in dynamic attenuation, The basic damage of a single plug and unplug is usually 0.15; Overload force multiple, usually 1.2, effective when Fk>0.8 Fmax; is the stress relaxation coefficient, which is usually 0.02 / hour for optical fiber connectors, depending on the material;

[0071] The dynamic wear of the screw mileage, As the basic wear rate, stainless steel guide rails are usually 0.003; is the nonlinear wear acceleration index, usually 2.1; The speed threshold refers to the point in time during the operation of an intelligent fiber optic patching robot where the lead screw speed reaches a certain value, which significantly impacts the wear of the device. This value is determined based on the device's mechanical properties and design specifications.

[0072] (3) Convert the timestamp of each plug-in and plug-out operation into a time code.

[0073] Among them, the timestamp of the plugging and unplugging operation records the specific time when each plugging and unplugging operation occurs. It can provide information about the time regularity and periodicity of the plugging and unplugging operations, and convert the timestamp into a numerical method suitable for model processing as the collection input data set. Convert the plugging and unplugging timestamp into a time interval, that is, the time difference (hours) between the current time and the most recent plugging and unplugging. According to the periodic characteristics of the intelligent fiber optic wiring robot plugging and unplugging, monthly and daily periodicity are selected to collect the periodic information of the plugging and unplugging operations. Among them, time code 1 represents periodicity within 30 days: convert the timestamp into hours within 30 days (0-720) and embed it through sine / cosine coding: ;

[0074] Time code 2 indicates intraday periodicity: Convert the timestamp to the hour of the day (0-23) and similarly perform sine / cosine encoding: .

[0075] Then, the time series data of different modes are aligned by time step to form a two-dimensional input tensor , as time series data of different modes. Considering that the plugging and unplugging of optical fiber connectors has a great impact on the performance and health of intelligent optical fiber wiring robots, the number of plugging and unplugging of optical fiber connectors at each time step is added to Xt as a new feature. The Min-Max normalization method can be used to convert the plugging and unplugging times n into .

[0076] Among them, when the local LSTM model of the intelligent fiber optic wiring robot is trained based on the locally collected and preprocessed time series data, the mean square error (MSE) loss function is used to calculate the mean square error between the predicted value and the true value to evaluate the accuracy of the model prediction.

[0077]

[0078] Where L represents the loss function value, which is used to measure the difference between the model's prediction and the true value. The smaller the loss function value, the closer the model's prediction is to the true value, and the better the model performance. N represents the total number of time series data samples used in model training, including information such as optical power loss, accumulated fatigue, and plug-in / plug-out frequency. The larger the number of samples, the more comprehensive the model's learning features and the stronger the generalization ability of the model training results. represents the true health of the intelligent fiber optic wiring robot labeled for the i-th sample; It represents the health prediction value of the intelligent fiber optic wiring robot obtained by analyzing the i-th sample, that is, the model analyzes the input sample data according to the currently set parameters. The output result obtained after forward propagation calculation. During the local LSTM model training process, the model parameters are continuously adjusted to make As close as possible , thereby reducing the loss function L value.

[0079] Furthermore, during the training of the intelligent fiber optic patching robot health status prediction model, considering the imbalance of data volume among different intelligent fiber optic patching robots and the fact that the processing efficiency of intelligent fiber optic patching robots is limited by computing power, the momentum stochastic gradient descent algorithm (Momentum SGD) was used to adjust the model parameters to obtain the smallest possible L value. The parameter adjustment process is as follows:

[0080]

[0081] in, Represents the parameter value of the t-th iteration model, that is, the latest parameter value after this update. Representative The model parameter value of the iteration, that is, the parameter value at the end of the previous iteration. represents the momentum at the tth iteration, which represents the cumulative effect of the gradient update direction; Calculated using the following formula:

[0082]

[0083] Among them, t represents the current iteration step, which indicates the current number of iterations. The model will be trained for multiple iterations, and the model parameters will be updated in each iteration. is the momentum coefficient, which ranges from [0,1] and is usually 0.9. Determines the degree of influence of previous momentum on current momentum. The larger μ is, the greater the influence of past gradient information on the current update, and the more the model can maintain the previous direction of movement during the update process; the smaller μ is, the greater the influence of the current gradient is. It is The momentum at the iteration represents the momentum value calculated at the previous iteration. The learning rate controls the step size of each parameter update. If the learning rate is too large, the model may oscillate around the optimal solution or even diverge. If the learning rate is too small, the model converges very slowly. Represents the loss function L with respect to the parameter θ in the The gradient at the iteration. The gradient represents the rate of change of the loss function at the current parameter value. Its direction points to the direction in which the loss function increases fastest, and the negative gradient direction is the direction in which the loss function decreases fastest.

[0084] In step S103, after the intelligent fiber optic distribution robot completes the local LSTM model parameter optimization, it feeds back the updated LSTM layer weights, fully connected layer weights, LSTM layer biases, and fully connected layer bias parameters to the global LSTM model, performs parameter alignment compensation and parameter aggregation strategies, and obtains updated global model parameters.

[0085] Specifically, due to the significant differences in the optical power normalization parameters between low-frequency plugging and high-frequency plugging intelligent fiber optic patching robots, feature scaling compensation is performed using the local standard deviation and global standard deviation of accumulated cycle statistics before federation aggregation:

[0086]

[0087] in, The original model parameters of the i-th intelligent fiber optic wiring robot obtained after local LSTM model training are trained based on the local data of device i and reflect the operating characteristics of device i itself; is the standard deviation of the local optical power loss of the i-th intelligent fiber optic wiring robot within the set accumulation period, reflecting the degree of dispersion of the local optical power loss data of device i. It is compared with the global standard deviation to determine the degree of difference between the data characteristics of device i and the global data characteristics; It is the standard deviation of the current optical power loss data of all intelligent fiber optic wiring robots. It represents the degree of global data dispersion and is used to unify the characteristic scales of different intelligent fiber optic wiring robots. By calculating the standard deviation of the optical power loss data of all devices, a global standard can be obtained to measure the fluctuation of data from different devices and avoid the influence of low-frequency plugging and unplugging of intelligent fiber optic wiring robots on the model. After feature scaling compensation, the local model parameters of the i-th device are obtained. This parameter not only takes into account the local optical power loss data characteristics of device i, but also adjusts the parameters by comparing them with the global standard deviation to make them more consistent with the requirements of the global model parameters.

[0088] Plugging and unplugging frequency reflects differences in device usage intensity. Different intelligent fiber optic patching robots have different plugging and unplugging frequencies, which impacts optical power loss. At low plugging and unplugging frequencies, optical power loss data is relatively stable but contains a high noise component. At high plugging and unplugging frequencies, optical power loss surges and fluctuates dramatically at the moment of plugging and unplugging. Considering the weighting effects of low- and high-frequency plugging and unplugging devices during parameter aggregation enables more accurate learning of the relationship between optical power loss and plugging and unplugging operations, improving the model's responsiveness to device health in extreme plugging and unplugging scenarios and ensuring that model output more closely reflects the actual device operation.

[0089] Due to the continuous wear and aging of optical and mechanical components of intelligent fiber-optic patching robots during long-term operation, cumulative fatigue measures the cumulative stress and wear experienced by the device, combining factors such as the number of plugging and unplugging cycles and the mileage of the leadscrew, along with the time decay effect. Introducing cumulative fatigue allows the model to comprehensively consider the health risks accumulated over the device's long-term operation, rather than focusing solely on current plugging and unplugging operations. This model supplements information from the dimension of long-term device aging, complementing short-term performance indicators such as optical power loss.

[0090]

[0091] in, Represents the global model parameters updated after parameter aggregation, by updating the local model parameters Perform weighted summation. is the number of local optical power loss data samples of the i-th intelligent fiber optic wiring robot; The sum of the local optical power loss data samples of all intelligent fiber optic wiring robots; is the historical average plugging and unplugging frequency (times / hour) of the i-th intelligent fiber optic wiring robot; The global average plug-in / plug-out frequency measures the plug-in / plug-out frequency level of the entire device. is the accumulated fatigue of the i-th intelligent optical fiber distribution robot; Indicates the average cumulative fatigue of all intelligent fiber optic wiring robots, measuring the fatigue level of the overall equipment; is the regulating factor, Increasing will increase the weight of the high-frequency plugging and unplugging intelligent fiber optic wiring robot. Usually 0.5; The activity coefficient is determined based on the number of plug-ins and unplugs in the past 30 days. (like =1+number of plugging and unplugging times in the last 30 days), thereby ensuring that intelligent fiber optic wiring robots with different plugging and unplugging frequencies contribute more reasonably to the calculation results of the global model parameters.

[0092] By taking into account the data volume, plug-in and unplugging frequency, and activity differences of different devices in a weighted manner, the parameters of intelligent fiber optic wiring robots with more plug-in and unplugging times have greater weight when aggregating the global model. At the same time, the characteristics of intelligent fiber optic wiring robots with low plug-in and unplugging frequencies are not ignored, making the global model more reasonable in integrating the information of each local model, and improving the adaptability and prediction accuracy of the health status of different devices.

[0093] After the global LSTM model completes dual weight aggregation and parameter alignment, the global model parameters will be updated This data is distributed to each intelligent fiber-optic patching robot. Each intelligent fiber-optic patching robot collects fiber-related data for the current moment and a historical period (e.g., the past 24 hours), including optical power loss, plug-in / plug-out times, temperature, humidity, and so on. This data is organized into time series data in the same format as used during training, and an input tensor X is constructed. This prepared input tensor X is then fed into the trained local LSTM model, which then outputs a predicted health value for the intelligent fiber-optic patching robot.

[0094] Based on a federated learning framework, this approach addresses the difficulty in collaboratively optimizing different intelligent fiber optic patching robot models and in fully leveraging overall data features. Through distributed modeling and global model aggregation, training results from individual intelligent fiber optic patching robots can be shared, improving model performance and adaptability. This enables the model to more comprehensively capture the operating patterns of each intelligent fiber optic patching robot in the network and accurately assess the health of the equipment. This helps meet the O&M requirements of different intelligent fiber optic patching robots in large-scale fiber optic networks and improve overall O&M efficiency. Furthermore, plugging and unplugging event markers and accumulated fatigue counters are embedded in the LSTM model. This approach addresses the difficulty in capturing the combined effects of instantaneous operation and long-term aging of intelligent fiber optic patching robots. It overcomes the limitations of traditional models that can only process single-type or static data, enabling dynamic and comprehensive modeling of device status. This ensures accurate assessment of device health, enabling timely identification of potential equipment issues, ensuring stable operation of intelligent fiber optic patching robots, and enhancing fiber optic network reliability. Furthermore, parameter aggregation considers both plugging and unplugging frequency and accumulated fatigue for dual weighting, while implementing an event alignment compensation mechanism. This solves the generalization problem of heterogeneous data from multiple devices, improves the limitation of traditional federated aggregation strategies that are difficult to handle data differences between devices, ensures that the global model reasonably integrates the local model information of each intelligent fiber optic wiring robot, enhances the adaptability and prediction accuracy of the health status of different devices, and enables the model to be better applied to actual complex intelligent fiber optic wiring robot scenarios.

[0095] In step S2, the main task is to construct a logical resource pool. Specifically, the optical fiber cores connected to different intelligent fiber optic patching robots are abstracted into logical resource pools. A unique logical identifier is assigned to the range of fiber cores managed by each intelligent fiber optic patching robot. For example, logical resource pool 1 includes two intelligent fiber optic patching robots A and B. Intelligent fiber optic patching robot A manages fiber cores 1-288; intelligent fiber optic patching robot B manages fiber cores 289-480. The resource pool logical identifiers corresponding to intelligent fiber optic patching robots A and B are Pool_1; logical resource pool 2 includes one intelligent fiber optic patching robot C. Intelligent fiber optic patching robot C manages fiber cores 481-768. The corresponding resource pool logical identifier is Pool_2. The logical identifiers of the fiber cores managed by all intelligent fiber optic patching robots are then integrated to form a global logical resource pool.

[0096] When a business request requires the selection of two intelligent fiber optic wiring robots to connect to different fiber cores, the dynamic weight is calculated based on the real-time load of each logical resource pool and the average health of the fiber core. , to balance the load of the intelligent fiber optic wiring robot and prioritize the assignment of tasks to the fiber core area managed by the intelligent fiber optic wiring robot with high health.

[0097]

[0098] in, is the dynamic weight of the i-th logical resource pool. The higher the weight, the more priority the logical resource pool will be selected when allocating tasks. 、 is the weight coefficient used to adjust the relative importance of load conditions and average core health, ; is the number of currently idle fiber cores in the i-th logical resource pool, reflecting the load of the logical resource pool. The more idle fiber cores, the lower the load. is the total number of cores in the i-th logical resource pool; The average fiber core health of the i-th logical resource pool is obtained by averaging the health scores of all fiber cores in the logical resource pool. It reflects the health status of the fiber cores. The higher the score, the better the fiber core health.

[0099]

[0100] in, The number of intelligent fiber optic wiring robots in the logical resource pool; Represents the health score of the i-th intelligent fiber optic wiring robot with a value range of 0-1. is the predicted health value of the i-th intelligent fiber optic wiring robot, expressed in percentage; The number of idle fiber cores connected to the i-th intelligent fiber optic patching robot. This parameter reflects the remaining resources of the intelligent fiber optic patching robot that can be used to assign tasks. The more idle fiber cores there are, the more likely this intelligent fiber optic patching robot will be considered for the next task assignment.

[0101] In step S3, when determining the optimal resource range, the priority of the logical resource pool is first determined according to the size of the dynamic weight, and the priority of each intelligent fiber optic distribution robot in the logical resource pool is determined according to the health of the intelligent fiber optic distribution robot and the number of idle fiber cores; then, a set number of logical resource pools and the intelligent fiber optic distribution robots contained therein are selected in order of priority as the optimal resource range, so that reinforcement learning performs fiber core path selection based on the logical resource pool, rather than in a large and disordered set of physical fiber cores, thereby greatly reducing the search space.

[0102] See the appendix of the manual. Figure 3 The method of determining a scheduling path that meets business requirements from the optimal resource range based on a search algorithm and a reinforcement learning algorithm includes the following steps:

[0103] S301. Calculate a set of feasible paths from a source node to a destination node within the optimal resource range using a search algorithm based on network topology information; if no path that meets the service requirements exists, expand the selected logical resource pool and the number of intelligent fiber optic distribution robots contained therein in order of priority, and redefine the optimal resource range until a path that meets the service requirements is found;

[0104] S302. Obtain the current state of the optical fiber network and select a path from the set of feasible paths using a greedy strategy as an action to be executed; wherein, if the generated random number is less than a set exploration rate, a path is randomly selected from the set of feasible paths as an action to be executed; if the generated random number is greater than the set exploration rate, a path with the largest expected cumulative reward is selected from the set of feasible paths as an action to be executed;

[0105] S303, calculating an immediate reward after the action is executed based on the set reward strategy, and calculating an expected cumulative reward after the action is executed based on the immediate reward, the set learning rate and discount factor, the expected cumulative reward when the action is not executed, and the maximum expected cumulative reward among all feasible paths;

[0106] S304: Perform iterative optimization of the path based on the calculated expected value of the cumulative reward to determine a scheduling path that meets the business requirements.

[0107] In step S301, a search algorithm is used to screen a set of feasible paths that meet service requirements within the optimal resource range determined based on the priority. To determine the path search range, several logical resource pools with the highest weights are first selected to ensure that the source and destination nodes belong to these logical resource pools. Then, for each selected logical resource pool, the intelligent fiber optic patching robot participating in the path search is determined based on its internal intelligent fiber optic patching robot priority queue. The set of fiber cores managed by the selected intelligent fiber optic patching robot serves as the initial range for the path search. Thus, based on the logical resource pools and the priorities of the intelligent fiber optic patching robots, a relatively small set of logical resource pools, intelligent fiber optic patching robots, and fiber cores is selected from the vast set of physical fiber cores, ensuring fiber core quality and balanced service core load. This provides a reasonable starting space for subsequent path search. Based on network topology information, a search algorithm is then used to calculate a set of feasible paths from the source node to the destination node from this set of fiber cores, which serves as the action space for reinforcement learning.

[0108] It should be noted that due to the selection of a smaller logical resource pool, intelligent fiber optic patching robot, and fiber core set, there is a possibility that a feasible path cannot be found from this fiber core set. If this occurs, the logical resource pools are expanded downwards in descending order according to the logical resource pool priority queue. The optimal resource range is then re-determined based on the intelligent fiber optic patching robot queue priority until a feasible path is found or all logical resource pools are traversed.

[0109] In one embodiment, the A-star algorithm is used to calculate the set of optional paths. :When a service request is received and the AZ starting and ending nodes are determined, the possible fiber core paths from the starting point to the end point are calculated based on the network topology information through the A-star algorithm. The A-star algorithm uses a heuristic search based on the breadth-first search algorithm to estimate the cost of node n to the target node. When selecting the next node to be expanded at each step, the A-star algorithm will give priority to The node with the smallest value.

[0110]

[0111] in, is the actual cost from the starting node to node n, represented by the normalized sum of the physical lengths of all optical fiber lines from the starting node to node n; is the estimated cost from node n to the target node, represented by the normalized value of the shortest number of hops from node n to the target node; is the health loss cost of the intelligent fiber optic wiring robot corresponding to node n ( is the normalized score of the robot's health, ranging from 0 to 1); are weights, and the sum of the three is 1. In this way, when calculating the link, not only the impact of optical cable length on optical signal attenuation is taken into account, but also the connection relationship of subsequent nodes and the health of the currently selected intelligent fiber optic wiring robot are considered, minimizing the impact of the intelligent fiber optic wiring robot's jumper connection on optical power loss.

[0112] In steps S302-S304, for each path in the set of feasible paths, its value is evaluated based on the current Q value, that is, learning is done by maintaining a Q value table to record the long-term cumulative reward expectation of performing a certain action in a certain state. In the fiber optic wiring scenario, each row of the Q value table corresponds to the real-time state s at a certain time t, and each column corresponds to a possible action (i.e., the fiber core connection path, calculated by the A-star algorithm). The evaluation value after the action is executed is , and its update formula is:

[0113]

[0114] in, In state The fiber core connection path action is not executed The Q value stored at To perform core connection path action The Q value after the current state The expected long-term cumulative reward of choosing this path; The learning rate is between 0 and 1. It controls the degree to which new information is learned each time the Q value is updated. The closer it is to 1, the greater the impact of the newly obtained reward information on the update of the Q value; The closer it is to 0, the more the Q value update depends on previous experience; To perform an action The immediate reward obtained after the action is used to measure the direct effect of the action; Discount factor, ranging from [0, 1]. It is used to measure the importance of future rewards. The closer it is to 1, the more importance is placed on future rewards, and the longer-term impact of current actions on subsequent states and rewards is considered. The closer it is to 0, the more it focuses on immediate rewards; Indicates execution of an action The new state to which the action is transferred reflects the change in state caused by the execution of the action; In the new state All feasible fiber core connection paths The maximum Q value in the set represents the maximum expected long-term cumulative reward from choosing the optimal path in the new state. Initially, the Q value for each path in the set of feasible paths is typically initialized to a default value, such as 0. This is because sufficient experience has not yet been accumulated, and the value of each path has not been clearly judged.

[0115] Among them, the status It can be represented by a multi-dimensional vector, , Network topology information is used to determine the physical connectivity of the network. It includes the connection relationship between nodes, link length, and optical power attenuation. It is composed of three N×N connection relationships (Connectivity), link length (LinkLength), and optical power attenuation (Attenuation) to form a three-dimensional matrix. Indicates the dynamic weight of each logical resource pool, used to guide path search, and is a one-dimensional array. Represents the health prediction value of each intelligent fiber optic wiring robot, which is a one-dimensional array. Indicates the fiber core occupancy status. Fiber cores are numbered according to "resource pool-robot-fiber core" and marked with the fiber core occupancy status. This allows you to quickly locate available fiber cores in the resource pool. It is a two-dimensional array.

[0116] In the Q-value calculation formula for each path, action a represents the selection of a feasible fiber connection path from the source node to the destination node, that is, the selection of a specific fiber core combination to establish an optical path connection. For example, selecting fiber core 1 from node A, passing through fiber core 5 at intermediate node B, and finally connecting to fiber core 8 at destination node C is an action. In a certain state, an action (i.e., a path) is selected from the set of feasible paths using the following greedy strategy: a path is randomly selected from the set of feasible paths with probability ε as the action; and the path with the highest Q-value is selected as the action with probability 1-ε:

[0117]

[0118] in, is the action selected at time t; is the exploration rate, and its value range is [0, 1]. In the initial stage of learning, Set a larger value (such as 0.3) so that there are more opportunities to explore different actions; as learning progresses, Gradually decreases, and is more inclined to choose the action that is currently considered optimal; The A-star algorithm calculates a set of feasible paths that meet optical power requirements based on network topology and service requests (starting and ending nodes, optical power requirements, etc.). In complex fiber networks, the A-star algorithm uses a heuristic function to estimate the cost from the current node to the target node, thereby quickly screening feasible paths. In state The Q value of executing action a under state The expected long-term cumulative reward of selecting action a.

[0119] Assume the current state is , a path is selected from the set of feasible paths After executing the path selection action, you will get instant rewards based on whether the connection is successfully established and whether the optical power meets the requirements. Then, the new Q value is calculated based on the Q value update formula. For example, if the connection is successfully established and the optical power meets the requirements, is positive (such as + 1); if the connection fails or the optical power does not meet the standard, is negative (such as -1). is the learning rate, which controls the influence of new reward information on the Q value update; is a discount factor that measures the importance of future rewards. In the new state The maximum Q value among all possible actions (i.e., all feasible fiber core connection paths) is determined by continuous iteration. Through continuous updates, the Q value gradually reflects the value of each path under different conditions. As experience accumulates, the master node evaluates the value of paths based on the updated Q value and prioritizes paths with high Q values ​​to achieve better decision-making results.

[0120] Among them, the instant reward after the action is executed is calculated based on the set reward strategy, and the reward strategy can be set according to one or more indicators of connection quality, fiber core health, business goals, and path efficiency. For example, according to the selected action, a fiber jumper instruction is issued to the designated intelligent fiber optic wiring robot, and the intelligent fiber optic wiring robot performs the fiber core connection operation. After the connection is completed, if a connection that meets the optical power index requirements is successfully established, an instant reward of +1 is given; if the connection fails, an instant reward of -1 is given; if the connection is successfully established but exceeds the optical power index threshold, a smaller negative reward (such as -0.5) is given. The specific setting depends on the actual application, and this application does not limit or fix it.

[0121] The above process of acquiring state information, selecting actions, executing actions, calculating rewards, and updating policies is repeated. As the number of iterations increases, the policy is continuously learned and optimized, gradually finding a more optimal fiber core path selection method. By combining the A-star algorithm with Q-value reinforcement learning, the action selection stage can more effectively utilize network topology information to find the optimal fiber core path for service requests. This overcomes the limitations of traditional path selection methods, eliminating blind searches or reliance on a single factor. Furthermore, the learning capabilities of reinforcement learning are leveraged to continuously optimize policies, improve SLA (Service-Level Agreement) compliance and resource utilization, and adapt to dynamic changes in network status.

[0122] The physical fiber cores are abstracted into logical resource pools, and dynamic weighted partition scheduling is used to optimize resource allocation. This overcomes the physical limitations of traditional resource allocation, eliminating the constraints imposed by a single robot's management range. By calculating dynamic weights for each logical resource pool based on real-time load and fiber core health, service requests can be precisely assigned to the optimal robot management area, achieving load balancing and overload avoidance. Compared with traditional fixed resource allocation models, this significantly improves fiber resource utilization efficiency. In high-density equipment room scenarios, it effectively resolves the issue of rigid resource allocation and improves overall O&M efficiency. Furthermore, reinforcement learning is combined with the A-star algorithm to generate scheduling paths that meet optical power and service level requirements. This approach overcomes the limitations of traditional path selection methods, eliminating blind search or reliance on a single factor. Path selection strategies are continuously optimized with experience, improving SLA compliance and resource utilization. This approach can better adapt to dynamic network conditions and achieve efficient service scheduling in complex fiber network environments.

[0123] Furthermore, when changes in the network, equipment, or link status are detected, or when fluctuations in the volume of service requests exceed a set threshold, the dynamic weights are recalculated and the scheduling path is determined.

[0124] (1) When the network status changes, such as a sudden failure in a robot management area, the dynamic weights of all logical resource pools are immediately recalculated. During the recalculation process, the fiber cores in the fault area are marked as unavailable, and the weights of the logical resource pools in which they are located are reduced. At the same time, the feasible path sets of all business requests are re-evaluated. For business requests originally assigned to the logical resource pools related to the fault area, the A-star algorithm is re-used to calculate the feasible paths based on the updated network topology and weight information. During the calculation process, the fiber cores in the fault area are avoided, and the Q value table is updated according to the new feasible path set.

[0125] (2) When the business request volume fluctuates significantly, the weight coefficients of load and health are dynamically adjusted according to the real-time business request queue length and the load of each logical resource pool. 、 If the request volume is high, the load factor is weighted more heavily in the weight calculation, prioritizing logical resource pools with more idle cores. If the request volume is low, the health factor is weighted more heavily, prioritizing areas with high core health. The action selection strategy is then adjusted based on the updated dynamic weights. When selecting paths, more attention is paid to logical resource pools with significant weight fluctuations, prioritizing feasible paths from these areas to improve resource allocation efficiency.

[0126] When multiple service requests are received, they are sorted by priority to form a priority queue. When multiple high-priority service requests compete for the same resource, the system first checks for available alternative resources (such as spare fiber cores or idle fiber cores in other logical resource pools). If alternative resources are available, the competing service request is allocated to the alternative resource. If no alternative resources exist, the following strategy is adopted based on the service's wait time and resource requirements: Prioritize resource allocation for services with longer wait times and relatively lower resource requirements; for services with similar wait times, the system further subdivides resources according to service priority, prioritizing the needs of higher-priority services.

[0127] (3) When a robot fails, stop allocating new service requests to the robot and mark the fiber core it manages as unavailable. Recalculate the dynamic weights of all logical resource pools and reduce the weight of the logical resource pool where the robot is located. For the fiber jump service being executed on the robot, if other robots have spare idle fiber cores, migrate the service to the corresponding robot, replan the fiber core connection path, and update the service connection relationship. If other robots do not have spare idle fiber cores, reschedule the service based on the service priority and importance. For high-priority services, try to find alternative paths in other robot management areas; for low-priority services, temporarily wait or process according to the interruption time allowed by the service. At the same time, notify the maintenance personnel to repair the faulty robot. After the robot is repaired, it will be reintegrated into the resource management system and normal service scheduling will be restored.

[0128] (4) When a fiber core failure is detected, the backup fiber core is immediately searched for based on the robot management area and logical resource pool information where the faulty fiber core is located. If there is a backup fiber core in the robot management area, an instruction is issued to the corresponding robot to switch the service of the faulty fiber core to the backup fiber core. At the same time, the fiber core occupancy status and service connection relationship are updated to ensure service continuity. If there is no backup fiber core in the area, the master control node recalculates the dynamic weight and searches for available fiber cores in other logical resource pools. The A-star algorithm is used to replan the service path to avoid the area where the faulty fiber core is located, and a new fiber jump instruction is issued to the relevant robots. After the service switch is completed, the faulty fiber core is marked and maintenance personnel are arranged to inspect it.

[0129] The application proposes an intelligent fiber optic distribution collaborative scheduling method that integrates federated learning, dynamic weighted scheduling, and reinforcement learning to create a three-layer collaborative architecture. This method automatically makes resource allocation and path selection decisions based on diverse business needs, minimizing manual intervention and human error, and enabling intelligent operations and maintenance management. Furthermore, the dynamic weighted partitioning mechanism and reinforcement learning algorithm allow for flexible policy adjustments based on network status changes, adapting to dynamic network changes.

[0130] Based on the same inventive concept, an intelligent fiber optic distribution collaborative scheduling device is also provided in an embodiment of the present application. Since the principle of solving the problem by the device in the embodiment of the present application is similar to the above-mentioned intelligent fiber optic distribution collaborative scheduling method in the embodiment of the present application, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be repeated.

[0131] As the instruction manual Figure 4 As shown, an embodiment of the present application further provides an intelligent optical fiber wiring collaborative scheduling device, which is applied to a cluster composed of multiple intelligent optical fiber wiring robots, and the device includes:

[0132] A federated learning health prediction module 401 is configured to train a global LSTM model for predicting the health status of the intelligent fiber optic wiring robot based on federated learning, obtain global model parameters, and embed the global model parameters into the local LSTM model of each intelligent fiber optic wiring robot to evaluate the health of each intelligent fiber optic wiring robot;

[0133] Dynamic weight partitioning module 402, for abstracting the cores of different intelligent fiber optic patching robots into logical resource pools, and calculating the dynamic weight of each logical resource pool according to the load of the logical resource pool and the health of the intelligent fiber optic patching robots contained therein;

[0134] The reinforcement learning path matching module 403 is configured to determine an optimal resource range based on the dynamic weights, and determine a scheduling path that meets service requirements from the optimal resource range based on a search algorithm and a reinforcement learning algorithm.

[0135] In one embodiment, the apparatus further comprises:

[0136] The adaptive operation and maintenance module is used to recalculate dynamic weights and determine scheduling paths when changes in network, equipment, or link status are detected, or when fluctuations in the volume of business requests exceed set thresholds.

[0137] In one embodiment, the federated learning health prediction module 401 obtains global model parameters based on federated learning training of a global LSTM model for predicting the health status of an intelligent fiber optic distribution robot, including: initializing the global LSTM model and sending the initialized global model parameters to the local LSTM model of each intelligent fiber optic distribution robot; preprocessing the device data collected by the intelligent fiber optic distribution robot, constructing time series data of different modalities, and training the local LSTM model based on the time series data; using the local training parameters obtained from each intelligent fiber optic distribution robot to perform parameter alignment compensation and parameter aggregation strategy on the global LSTM model to obtain updated global model parameters.

[0138] In one embodiment, the device data includes optical power loss, the number of plugging and unplugging of optical fiber connectors, the mileage of the lead screw and environmental data. The federated learning health prediction module 401 pre-processes the device data collected by the intelligent optical fiber distribution robot to construct time series data of different modalities, including: counting the number of plugging and unplugging in each window based on a sliding window mechanism, and dividing it into low-frequency plugging and unplugging scenarios and high-frequency plugging and unplugging scenarios according to a preset threshold value; wherein, in the low-frequency plugging and unplugging scenario, the optical power loss is normalized by extreme value; in the high-frequency plugging and unplugging scenario, the optical power loss is normalized by standard deviation; based on the number of plugging and unplugging, the mileage of the lead screw and the time attenuation effect, the cumulative fatigue is calculated to measure the cumulative stress and loss suffered by the equipment during use; the timestamp of each plugging and unplugging operation is obtained, and it is converted into a first time code and a second time code according to the periodicity of the plugging and unplugging of the intelligent optical fiber distribution robot; the normalized optical power loss, cumulative fatigue, number of plugging and unplugging, first time code, second time code and environmental data are aligned according to time steps to form a two-dimensional input tensor as time series data of different modalities.

[0139] In one embodiment, the reinforcement learning path matching module 403 determines the optimal resource range based on the dynamic weight, including: determining the priority of the logical resource pool according to the size of the dynamic weight, and determining the priority of each intelligent fiber optic distribution robot in the logical resource pool according to the health of the intelligent fiber optic distribution robot and the number of idle fiber cores; selecting a set number of logical resource pools and the intelligent fiber optic distribution robots contained therein in order of priority as the optimal resource range.

[0140] In one embodiment, the reinforcement learning path matching module 403 determines a scheduling path that meets the business requirements from the optimal resource range based on a search algorithm and a reinforcement learning algorithm, including: using a search algorithm based on network topology information to calculate a set of feasible paths from a source node to a destination node from the optimal resource range; wherein, if there is no path that meets the business requirements, the selected logical resource pool and the number of intelligent fiber optic wiring robots contained therein are expanded in order of priority, and the optimal resource range is re-determined until a path that meets the business requirements is found; obtaining the current state of the optical fiber network, and using a greedy strategy to select a path from the feasible path set as an action Execution; wherein, if the generated random number is less than the set exploration rate, a path is randomly selected from the set of feasible paths as the action to be executed; if the generated random number is greater than the set exploration rate, the path with the largest expected cumulative reward is selected from the set of feasible paths as the action to be executed; the immediate reward after the action is executed is calculated based on the set reward strategy, and the expected cumulative reward after the action is executed is calculated based on the immediate reward, the set learning rate and discount factor, the expected cumulative reward when the action is not executed, and the largest expected cumulative reward among all feasible paths; based on the calculated expected cumulative reward, the path is iteratively optimized to determine the scheduling path that meets the business requirements. The search algorithm adopts the A-star algorithm, and the state of the optical fiber network includes network topology information, the dynamic weight of the logical resource pool, and the health of the intelligent optical fiber distribution robot; the network topology information includes the connection relationship between nodes, link length, and optical power attenuation.

[0141] The intelligent fiber optic wiring collaborative scheduling device described in this application is applied to a cluster composed of multiple intelligent fiber optic wiring robots. The global LSTM model used to predict the health status of the intelligent fiber optic wiring robots is trained based on federated learning through a federated learning health prediction module to obtain global model parameters, and the global model parameters are loaded into the local LSTM model of each intelligent fiber optic wiring robot to evaluate the health of each intelligent fiber optic wiring robot; the fiber cores of different intelligent fiber optic wiring robots are abstracted into logical resource pools through a dynamic weight partitioning module, and the dynamic weights of each logical resource pool are calculated based on the load of each logical resource pool and the health of the intelligent fiber optic wiring robots contained therein; the optimal resource range is determined based on the dynamic weights through a reinforcement learning path matching module, and a scheduling path that meets business requirements is determined from the optimal resource range based on a search algorithm and a reinforcement learning algorithm. Thus, a three-layer collaborative architecture is constructed by integrating federated learning, dynamic weight scheduling, and reinforcement learning. It can automatically make resource allocation and path selection decisions according to different business needs, reduce manual intervention, reduce human errors, and realize intelligent operation and maintenance management.

[0142] Based on the same concept of the present invention, as shown in the attached specification Figure 5As shown, an embodiment of the present application provides a structure of an electronic device 500, which includes: at least one processor 501, at least one network interface 504 or other user interface 503, a memory 505, and at least one communication bus 502. The communication bus 502 is used to implement connection and communication between these components. The electronic device 500 optionally includes a user interface 503, including a display (e.g., a touch screen, LCD, CRT, holographic imaging (Holographic) or projector (Projector), etc.), a keyboard or a pointing device (e.g., a mouse, trackball (trackball), touchpad or touch screen, etc.).

[0143] The memory 505 may include a read-only memory and a random access memory, and provides instructions and data to the processor 501. A portion of the memory 505 may also include a non-volatile random access memory (NVRAM).

[0144] In some embodiments, the memory 505 stores the following elements, executable modules, or data structures, or a subset or extended set thereof:

[0145] Operating system 5051, including various system programs for implementing various basic services and processing hardware-based tasks;

[0146] The application module 5052 includes various application programs, such as a launcher, a media player, and a browser, and is used to implement various application services.

[0147] In the embodiment of the present application, by calling the program or instructions stored in the memory 505, the processor 501 is used to execute the steps of an intelligent optical fiber distribution collaborative scheduling method.

[0148] The present application also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the computer program executes steps in an intelligent optical fiber distribution collaborative scheduling method.

[0149] Specifically, the storage medium can be a general storage medium, such as a mobile disk, a hard disk, etc. When the computer program on the storage medium is run, it can realize the dynamic scheduling of optical fiber resources within the cluster through the three-layer collaborative architecture of federated learning prediction-dynamic weight scheduling-reinforcement learning decision.

[0150] In the embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interface, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0151] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0152] In addition, each functional unit in the embodiments provided in the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0153] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the existing technology, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program code.

[0154] Finally, it should be noted that the above embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The scope of protection of the present application is not limited thereto. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above embodiments within the technical scope disclosed in the present application, or replace some of the technical features therein with equivalents. However, these modifications, changes, or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application. They should all be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A method for collaborative scheduling of intelligent optical fiber distribution, characterized in that: Applied to a cluster consisting of multiple intelligent fiber optic wiring robots, the method includes the following steps: Based on federated learning, a global LSTM model for predicting the health status of intelligent fiber optic wiring robots is trained to obtain global model parameters. These global model parameters are then loaded into the local LSTM model of each intelligent fiber optic wiring robot to evaluate the health of each intelligent fiber optic wiring robot. The fiber cores of different intelligent fiber optic patching robots are abstracted into logical resource pools, and their dynamic weights are calculated based on the load of each logical resource pool and the health of the intelligent fiber optic patching robots contained therein; An optimal resource range is determined based on the dynamic weight, and a scheduling path that meets the business requirements is determined from the optimal resource range based on a search algorithm and a reinforcement learning algorithm; wherein, a search algorithm is used to calculate a set of feasible paths from a source node to a destination node from the optimal resource range based on network topology information; wherein, if there is no path that meets the business requirements, the selected logical resource pool and the number of intelligent fiber optic wiring robots contained therein are expanded in order of priority, and the optimal resource range is re-determined until a path that meets the business requirements is found; the current state of the optical fiber network is obtained, and a path is selected from the set of feasible paths using a greedy strategy as an action execution. row; wherein, if the generated random number is less than the set exploration rate, a path is randomly selected from the set of feasible paths as the action to be executed; if the generated random number is greater than the set exploration rate, the path with the largest expected cumulative reward value is selected from the set of feasible paths as the action to be executed; the immediate reward after the action is executed is calculated based on the set reward strategy, and the expected cumulative reward after the action is executed is calculated based on the immediate reward, the set learning rate and discount factor, the expected cumulative reward when the action is not executed, and the largest expected cumulative reward among all feasible paths; the path is iteratively optimized based on the calculated expected cumulative reward to determine the scheduling path that meets the business requirements.

2. The intelligent optical fiber distribution collaborative scheduling method according to claim 1, characterized in that: The method further comprises the following steps: When changes in the network, equipment, or link status are detected, or when fluctuations in the volume of service requests exceed a set threshold, the dynamic weights are recalculated and the scheduling path is determined.

3. The intelligent optical fiber distribution collaborative scheduling method according to claim 1, characterized in that: The method of training a global LSTM model for predicting the health status of an intelligent optical fiber distribution robot based on federated learning to obtain global model parameters includes the following steps: Initializing the global LSTM model and sending the initialized global model parameters to the local LSTM models of each intelligent fiber optic wiring robot; Preprocessing the device data collected by the intelligent fiber optic wiring robot to construct time series data of different modalities, and training a local LSTM model based on the time series data; The local training parameters obtained by each intelligent fiber optic wiring robot are used to perform parameter alignment compensation and parameter aggregation strategy on the global LSTM model to obtain the updated global model parameters.

4. The intelligent optical fiber distribution collaborative scheduling method according to claim 3, characterized in that: in, The device data includes optical power loss, number of plugging and unplugging of optical fiber connectors, lead screw mileage and environmental data. The device data collected by the intelligent optical fiber wiring robot is preprocessed to construct time series data of different modes, including the following steps: The number of plug-in / plug-out times within each window is counted based on a sliding window mechanism, and is divided into low-frequency plug-in / plug-out scenarios and high-frequency plug-in / plug-out scenarios according to a preset threshold value. In the low-frequency plug-in / plug-out scenario, the optical power loss is normalized using extreme value; in the high-frequency plug-in / plug-out scenario, the optical power loss is normalized using standard deviation. The cumulative fatigue is calculated based on the number of plug-in and pull-out times, the screw mileage and the time decay effect, which is used to measure the cumulative stress and loss that the equipment is subjected to during use; Obtaining the timestamp of each plugging and unplugging operation, and converting it into a first time code and a second time code according to the periodicity of plugging and unplugging of the intelligent optical fiber distribution robot; The normalized optical power loss, cumulative fatigue, number of plug-in and unplugging times, first time code, second time code, and environmental data are aligned by time step to form a two-dimensional input tensor as time series data of different modalities.

5. The intelligent optical fiber distribution collaborative scheduling method according to claim 1, characterized in that: Determining the optimal resource range based on the dynamic weight includes the following steps: Determining the priority of the logical resource pool according to the size of the dynamic weight, and determining the priority of each intelligent optical fiber distribution robot in the logical resource pool according to the health of the intelligent optical fiber distribution robot and the number of idle fiber cores; A set number of logical resource pools and the intelligent fiber optic wiring robots contained therein are selected in order of priority as the optimal resource range.

6. The intelligent optical fiber distribution collaborative scheduling method according to claim 5, characterized in that: in, The search algorithm adopts the A-star algorithm, and the state of the optical fiber network includes network topology information, dynamic weights of the logical resource pool, and the health of the intelligent optical fiber distribution robot; the network topology information includes the connection relationship between nodes, link length, and optical power attenuation.

7. An intelligent optical fiber wiring collaborative scheduling device, characterized in that: Applied to a cluster consisting of multiple intelligent fiber optic wiring robots, the device includes: A federated learning health prediction module is used to train a global LSTM model for predicting the health status of intelligent fiber optic wiring robots based on federated learning, obtain global model parameters, and implement the global model parameters into the local LSTM model of each intelligent fiber optic wiring robot to evaluate the health of each intelligent fiber optic wiring robot; A dynamic weight partitioning module is used to abstract the fiber cores of different intelligent fiber optic patching robots into logical resource pools, and calculate the dynamic weight of each logical resource pool according to the load of the logical resource pool and the health of the intelligent fiber optic patching robots contained therein; A reinforcement learning path matching module is used to determine the optimal resource range based on the dynamic weight, and determine the scheduling path that meets the business requirements from the optimal resource range based on the search algorithm and the reinforcement learning algorithm; wherein, based on the network topology information, a search algorithm is used to calculate the set of feasible paths from the source node to the destination node from the optimal resource range; wherein, if there is no path that meets the business requirements, the selected logical resource pool and the number of intelligent fiber optic wiring robots contained therein are expanded in order of priority, and the optimal resource range is re-determined until a path that meets the business requirements is found; the current state of the optical fiber network is obtained, and a path is selected from the set of feasible paths using a greedy strategy. path as the action to be executed; wherein, if the generated random number is less than the set exploration rate, a path is randomly selected from the set of feasible paths to be executed as the action; if the generated random number is greater than the set exploration rate, a path with the largest expected cumulative reward value is selected from the set of feasible paths to be executed as the action; the immediate reward after the action is executed is calculated based on the set reward strategy, and the expected cumulative reward value after the action is executed is calculated based on the immediate reward, the set learning rate and discount factor, the expected cumulative reward value when the action is not executed, and the largest expected cumulative reward value among all feasible paths; the path is iteratively optimized based on the calculated expected cumulative reward value to determine the scheduling path that meets the business requirements.

8. An electronic device, characterized in that: include: A processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor and the memory communicate via the bus. When the machine-readable instructions are executed by the processor, the steps of the intelligent optical fiber distribution collaborative scheduling method according to any one of claims 1 to 6 are performed.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, executes the steps of the intelligent optical fiber distribution collaborative scheduling method according to any one of claims 1 to 6.

Images