Optical fiber transmission resource dynamic allocation method and system based on cloud computing

Abstract

The invention relates to the technical field of cloud computing, and solves the technical problems in the prior art that linear and nonlinear characteristics of data are difficult to consider at the same time, a scientific and reasonable weight distribution method is lacked, and service characteristics are not fully considered, so that a prediction result does not conform to the actual demand of an optical fiber. In particular to an optical fiber transmission resource dynamic allocation method and system based on cloud computing, and the method comprises the steps: collecting original network equipment operation data and original service request data from an optical fiber transmission system, and obtaining a multi-source data set through preprocessing. According to historical error inverse proportion weighting, prediction is more accurate, limitation of a single model is avoided, a demand matrix related to optical fiber services is generated through an optical fiber service matrix, actual services are closely fitted, an accurate basis is provided for optical fiber transmission and distribution, and prediction accuracy and practicability of optical fiber service data are improved.

Description

Technical Field

[0001] This invention relates to the field of cloud computing technology, and in particular to a method and system for dynamic allocation of optical fiber transmission resources based on cloud computing. Background Technology

[0002] Cloud computing is a type of distributed computing that refers to breaking down massive data processing programs into countless smaller programs through a network cloud. These smaller programs are then processed and analyzed by a system composed of multiple servers to obtain results and return them to the user. The dynamic allocation method of fiber optic transmission resources is closely integrated with the dynamic performance requirements of cloud services for network bandwidth, latency, etc. It utilizes intelligent algorithms for rapid analysis and calculation, and dynamically adjusts the allocation of fiber optic transmission resources when it detects an increase in demand for a certain business or congestion on a certain link.

[0003] Existing technologies often use a single model to predict fiber optic transmission resource data, which makes it difficult to simultaneously take into account the linear and nonlinear characteristics of the data. This results in significant deviations in the prediction results of fiber optic service data. Furthermore, the lack of a scientific and reasonable weight allocation method makes it impossible to dynamically adjust the weights based on the actual performance of the model. When converting the prediction results into service requirements, the service characteristics are not fully considered, causing the prediction results to be inconsistent with the actual needs of fiber optics. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a cloud computing-based method and system for dynamic allocation of optical fiber transmission resources. It solves the technical problems of existing technologies, such as difficulty in simultaneously considering the linear and nonlinear characteristics of data, lack of scientific and reasonable weight allocation methods, and insufficient consideration of business characteristics, which leads to discrepancies between prediction results and actual optical fiber needs.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a method for dynamic allocation of optical fiber transmission resources based on cloud computing, the steps of which are as follows: Raw network device operation data and raw service request data are collected from the fiber optic transmission system and preprocessed to obtain a multi-source dataset. Perform feature extraction and standardization on multi-source datasets to obtain standardized data; Based on standardized data, a predicted fiber optic service matrix is ​​obtained through demand forecasting. The system acquires real-time network topology and network resource status, and performs dynamic resource allocation based on the predicted fiber service matrix, real-time network topology, and network resource status to obtain a fiber resource allocation scheme. Based on the fiber optic resource allocation scheme, scheme detection and self-healing are performed to obtain a fiber optic network status report.

[0006] Preferably, the multi-source dataset is obtained through preprocessing, including: Distributed probe collection and processing are performed based on raw network device operation data and raw service request data to obtain initial network status data and initial service data. Based on the initial network status data and initial business data, compliance checks are conducted to obtain compliance status data and compliance business data; Time alignment is performed on compliance status data and compliance business data to obtain a multi-source dataset.

[0007] Preferably, feature extraction and standardization processing are performed on multi-source datasets, including: Cleaned data is obtained by cleaning data from multi-source datasets; Feature extraction is performed on the cleaned data to obtain a primary feature vector; Standardized data is obtained by standardizing the primary feature vectors.

[0008] Preferably, demand forecasting is used to process standardized data, including: Based on standardized data, time-series processing is used to obtain fiber optic time-series data; A moving average layer is constructed based on fiber optic time series data, and a linear prediction sequence is obtained through autoregressive moving average processing. The predicted fiber service matrix is ​​obtained by nonlinear processing and dynamic fusion of fiber time series data and linear prediction sequences.

[0009] Preferably, nonlinear processing and dynamic fusion are performed on fiber optic time-series data and linear prediction sequences, including: A long short-term memory (LSM) network layer is constructed to process fiber optic time-series data and obtain nonlinear prediction sequences. An error verification layer is constructed, and the linear and nonlinear prediction sequences are inversely weighted by error to obtain a fused prediction sequence. A matrix generation layer is constructed based on the fused prediction sequence, and the predicted fiber service matrix is ​​obtained.

[0010] Preferably, dynamic resource allocation is performed based on the predicted fiber service matrix, real-time network topology, and network resource status, including: A digital optimization model is obtained by mathematically modeling the predicted fiber optic service matrix, real-time network topology, and network resource status. The chromosome coding layer was constructed based on a digital optimization model to obtain the initial population. Genetic iteration processing is performed on the initial population to obtain the fiber optic resource allocation scheme.

[0011] Preferably, genetic iteration processing is performed based on the initial population, including: An optimized population is obtained by performing population optimization on the initial population. A Q-learning layer is constructed to fine-tune the optimized population, resulting in a tuned population. Feasibility was verified based on the optimized population, and a fiber optic resource allocation scheme was obtained.

[0012] Preferably, the scheme detection and self-healing process based on the fiber optic resource allocation scheme includes: Based on the fiber optic resource allocation scheme and real-time network monitoring data, a real-time multi-dimensional indicator vector is obtained through indicator detection and processing. An isolated forest layer is constructed based on real-time multi-dimensional indicator vectors, and abnormal indicators are obtained through anomaly indicator detection. A historical anomaly detection layer is constructed based on anomaly indicators, and historical anomaly data is obtained through nearest neighbor similarity processing.

[0013] This technical solution also provides a system for applying the aforementioned cloud computing-based dynamic allocation method for optical fiber transmission resources, the system comprising: The acquisition module is used to collect raw network device operation data and raw service request data from the fiber optic transmission system, and obtain multi-source datasets through preprocessing. The standardization module is used to perform feature extraction and standardization processing on multi-source datasets to obtain standardized data. The forecasting module is used to obtain a forecasted fiber optic service matrix based on standardized data through demand forecasting processing. The allocation module is used to obtain real-time network topology and network resource status, and to perform dynamic resource allocation based on the predicted fiber service matrix, real-time network topology and network resource status to obtain a fiber resource allocation scheme. The status report module is used to perform scheme detection and self-healing based on the fiber optic resource allocation scheme, and to obtain a fiber optic network status report.

[0014] By employing the above technical solution, the present invention provides a method and system for dynamic allocation of optical fiber transmission resources based on cloud computing, which has at least the following beneficial effects: 1. This invention employs an algorithm combining box plots and sliding windows to accurately identify local anomalies, effectively improving the comprehensiveness and accuracy of anomaly detection and avoiding the omission of abnormal data. The combined use of Pearson correlation coefficient and principal component analysis not only considers the linear correlation between features and removes redundant features, but also mines the potential structure of the data through principal component analysis, extracts the most representative features, reduces data dimensionality, and improves the efficiency of subsequent processing. Through standardization, feature data of different dimensions are unified to a standard normal distribution, eliminating the influence of dimensions and enhancing data comparability.

[0015] 2. This invention dynamically integrates linear and nonlinear predictions, and weights them inversely based on historical errors to make predictions more accurate, avoiding the limitations of a single model. It generates a demand matrix related to fiber optic services through a fiber optic service matrix, closely aligning with actual services and providing a precise basis for fiber optic transmission allocation, thereby improving the prediction accuracy and practicality of fiber optic service data.

[0016] 3. This invention integrates multi-constraint objective programming, genetic algorithms, reinforcement learning, and constraint satisfaction checking to form a comprehensive and refined optimization system. Multi-constraint objective programming provides accurate modeling that fits the actual network requirements. Genetic algorithms have strong global search capabilities and can quickly find near-optimal solution sets. Reinforcement learning further optimizes local solutions and improves the quality of the solution. Constraint satisfaction checking ensures the feasibility of the solution. All steps work closely together to effectively balance global and local considerations, as well as efficiency and quality. This allows for more efficient and accurate determination of fiber optic paths and resource allocation schemes that satisfy multiple constraints and have optimized performance.

[0017] 4. This invention combines the isolated forest anomaly detection algorithm with a case-based reasoning self-healing decision model. The isolated forest algorithm efficiently identifies anomalies in multi-dimensional monitoring indicators, while the case-based reasoning algorithm utilizes a historical handling experience database to match and adjust the optimal handling strategy for the current anomaly. The combination of the two algorithms realizes a complete processing flow from rapid anomaly perception to intelligent decision-making, improving the robustness of the system and the accuracy of fiber optic network adjustment schemes. Attached Figure Description

[0018] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart of the cloud computing-based dynamic allocation method for optical fiber transmission resources according to the present invention. Figure 2 This is a structural block diagram of the cloud computing-based dynamic allocation system for optical fiber transmission resources according to the present invention. Detailed Implementation

[0019] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. This will allow for a full understanding of how the present application uses technical means to solve technical problems and achieve technical effects, and to facilitate its implementation.

[0020] Example 1: Due to the difficulty of simultaneously considering the linear and nonlinear characteristics of data in existing technologies, the lack of a scientifically sound weight allocation method, and the insufficient consideration of business characteristics, the prediction results do not match the actual needs of optical fibers. For further information, please refer to [link to relevant documentation / reference]. Figure 1This embodiment provides a cloud computing-based dynamic allocation method for optical fiber transmission resources, which can scientifically and rationally allocate weights, fully consider the characteristics of optical fiber services, and ensure that the prediction results match the actual demand for optical fibers. The method includes the following steps: S1. Collect raw network equipment operation data and raw service request data from the fiber optic transmission system, and obtain a multi-source dataset through preprocessing. Existing technologies often rely on a single or limited number of collection points for data collection, making it difficult to comprehensively cover the network environment, easily leading to data gaps, incomplete analysis of network operation, and a lack of rigorous and detailed verification mechanisms. Insufficient and in-depth data compliance checks result in a large amount of erroneous and non-standard data entering subsequent processing. To solve these problems, the specific implementation steps are as follows: S11. Based on the original network device operation data and original service request data, distributed probe collection and processing are performed to obtain initial network status data and initial service data. In this step, for the original network device operation data, specific indicator values ​​such as port traffic, device temperature, and device load are obtained in real time through various probes. These values ​​of different devices and different indicators are arranged and integrated in an orderly manner according to the time series. This can be regarded as superimposing and summarizing the different indicator values ​​of various devices along the time dimension to finally form the initial network status data, which includes comprehensive operational status information of network devices at various times. For the original service request data, distributed probes capture key information such as the initiation time, request type, request source, and request target of the service request. This information is arranged and combined according to the chronological order of the service requests, which is equivalent to arranging the service requests... Key information is pieced together in the order of request occurrence to generate an initial business request stream, fully presenting the initiation and flow of business requests. The original business request data includes requests initiated by employees to access the company's internal website at a certain time, recording the request initiation time, request type (e.g., HTTP request), request source (e.g., employee's office computer IP address), and request target (e.g., the company's internal website server IP address). The original network device operation data mainly includes port traffic, device temperature, device load, and device operating status (e.g., normal or faulty) of network devices such as routers, switches, and servers. Distributed probes are specialized devices or software modules deployed in a distributed manner in the network environment for data collection and monitoring, including hardware probes and software probes. This is a commonly used method for data collection and will not be elaborated upon here.

[0021] S12. Based on the initial network state data and initial business data, a compliance check is performed to obtain compliance state data and compliance business data. In this step, the initial network state stream and initial business request stream are first parsed according to protocols. Based on the various protocol specifications followed by network communication, the data stream is broken down into data units with clear meaning. For example, a complex string of codes is split into multiple independent and meaningful segments according to the data length A. Then, a compliance check is performed, comparing each data unit's parameters against established data format standards. For example, whether the data length meets the specified range can be understood as comparing the actual data length with the specified minimum and maximum values. Within the range formed by the minimum and maximum values, compliance is ensured. Data type matching is also checked; for example, if a parameter is specified to be an integer, checking if it is an integer can be seen as comparing the data type identifier with the specified integer type identifier. If they match, compliance is ensured. The existence and reasonable value of key fields are also checked, such as the request type field in the business request flow. Checking if it is in the predefined request type set can be seen as comparing the actual request type value with each element in the predefined set. If it exists, compliance is ensured. Data units that do not meet the requirements are marked and corrected. After processing all data units, they are reassembled into a data flow, ultimately resulting in a compliant network state flow and a compliant business request flow.

[0022] S13. Time-align the compliance status data and compliance business data to obtain a multi-source dataset. In this step, a fixed-duration time window is first set, for example, 5 seconds. Using the start time of the time window as a baseline, data within that time window is filtered from both the compliance status data and the compliance business data. For the compliance status data, the status data collected by all network devices at different times within the window is initially organized according to dimensions such as device type and collection metrics to obtain status processing data. For the compliance business data, the relevant information of all business requests within the window is initially classified according to dimensions such as request type and request source to obtain business processing data. Then, the initially organized and classified status processing data is compared with the business processing data. This invention aggregates data by using time windows as a unified identifier. For example, data from different sources but with similar times are merged according to time windows. Ultimately, all network status data and business request data within the same time window are combined to form a merged multi-source dataset. This invention employs a distributed probe collection method, which can be flexibly deployed at various key nodes in the network to comprehensively and accurately capture raw network device operation data and raw business request data. This avoids data omissions that may occur due to a single collection point. Through compliance checks, erroneous and non-standard data are effectively eliminated, greatly improving data quality. By synchronously aggregating data through time windows, data from different sources but with similar times are accurately integrated, providing a unified and complete data foundation for subsequent analysis and mining.

[0023] S2. Perform feature extraction and standardization on the multi-source dataset to obtain standardized data. Existing technologies are difficult to adapt to the diversity and dynamic changes in fiber optic data distribution, resulting in inaccurate anomaly detection. Relying solely on simple correlation analysis cannot fully consider the influence relationships between features and the potential correlations between data. The extracted features may lack representativeness and fail to effectively reflect data changes. To solve these problems, the specific implementation steps are as follows: S21. Based on the multi-source dataset, data cleaning is performed to obtain cleaned data. In this step, for each type of data indicator in the data pool, such as network traffic, device load, and business request response time, the quartiles are calculated using the box plot principle, namely the first quartile, median, and third quartile. Then, the interquartile range is calculated as the third quartile minus the first quartile. According to the rules of the box plot, the judgment range for outliers is set. The lower limit is obtained by subtracting 1.5 from the first quartile and multiplying by the interquartile range. The upper limit is obtained by adding 1.5 to the third quartile. If... Data values ​​between the upper and lower limits are considered normal, while those outside these limits are considered outliers. A sliding window mechanism is then introduced, using a specific time interval or data size as the window size (e.g., M data points as the window size). A sliding window is established on the data sequence with a sliding step size of N. Box plots are then applied to the data within each window to detect outliers. This approach considers local characteristics and temporal variations in the data, avoiding the omission of local anomalies by global statistical methods. Finally, the detected outliers are replaced with the mean of the adjacent data within the window, resulting in cleaned multi-source data.

[0024] S22. Feature extraction is performed on the cleaned data to obtain primary feature vectors. In this step, the Pearson correlation coefficient of the cleaned data is calculated pairwise to measure the linear correlation between features. Features can be understood as the dimensionality or key information variables of different categories of data in the cleaned data. By setting a correlation coefficient threshold, such as 0.75, features with high correlation coefficients are filtered to remove redundant features and retain key features with independent information, thereby reducing data dimensionality. Then, a covariance matrix is ​​constructed on the filtered feature data. The elements of the covariance matrix are composed of the covariance of different feature data. The covariance is obtained by calculating the expectation of the product of the difference between two feature data and their respective means. Then, the eigenvalues ​​and eigenvectors of the covariance matrix are solved. The feature vectors are sorted in descending order of eigenvalues, and the first few main feature vectors are selected. These feature vectors can retain the information of the original data to the greatest extent. Finally, the filtered key feature data are projected onto the selected main feature vectors to obtain primary feature vectors. The calculation formula of the covariance matrix is ​​a commonly used feature calculation method, which will not be elaborated here.

[0025] S23. Based on the primary feature vectors, standardized data is obtained through standardization. In this step, a feature dimension is first selected, and the mean of all sample feature values ​​under that dimension is calculated. For example, the transmission rate and signal strength at three different times in two fiber optic channels represent three different dimensions of the two samples. This mean represents the average level of the data in that feature dimension. Next, the standard deviation of all sample feature values ​​under that dimension is calculated. The standard deviation reflects the dispersion of the data in that feature dimension relative to the mean. Then, the mean is subtracted from each original feature value under that dimension, and then divided by the standard deviation. Through this operation, a linear transformation is performed on each feature value, ultimately obtaining the standardized data. The eigenvalues, among which the formulas for calculating the mean and standard deviation are commonly used statistical formulas and will not be elaborated here. This invention employs an algorithm combining box plots and sliding windows to accurately identify local anomalies, effectively improving the comprehensiveness and accuracy of anomaly detection and avoiding the omission of abnormal data. The combined use of Pearson correlation coefficient and principal component analysis not only considers the linear correlation between features and removes redundant features, but also mines the potential structure of the data through principal component analysis, extracts the most representative features, reduces data dimensionality, and improves the efficiency of subsequent processing. Through standardization, feature data of different dimensions are unified to a standard normal distribution, eliminating the influence of dimensions and enhancing data comparability.

[0026] S3. Based on standardized data, a predicted fiber optic service matrix is ​​obtained through demand forecasting. Existing technologies often use a single model for prediction, which makes it difficult to simultaneously consider the linear and nonlinear characteristics of the data. This leads to significant deviations in the predicted fiber optic service data, a lack of scientific and reasonable weight allocation methods, and an inability to dynamically adjust weights based on the actual performance of the model. Furthermore, when converting the prediction results into service requirements, the service characteristics are not fully considered, resulting in a discrepancy between the prediction results and the actual fiber optic demand. To address these issues, the specific implementation steps are as follows: S31. Based on the standardized data, time-series processing is performed to obtain fiber optic time-series data. In this step, a window of size P is first set on the standardized data. The size of the window can be determined according to the time granularity of the data requirements and the characteristics of the data. Starting from the beginning of the standardized data, the data covered by the window is taken as a subsequence. Feature data in the subsequence is extracted, and these feature data are weighted and summed according to certain weights. The weights can be preset according to the degree of influence of each feature on the business requirements. For example, the weight coefficient is set based on the size of the Pearson correlation coefficient. If the Pearson correlation coefficients of the features are the same, the weights are equal, resulting in a comprehensive value. This comprehensive value represents the quantitative value of fiber optic demand within the time window period. Then, the window slides forward by one step according to the time sequence, which is determined according to the data time interval, for example, setting the step size to R. The above weighted summation operation is repeated until the window slides to the end of the standardized data, and finally the fiber optic time-series data is obtained.

[0027] S32. Construct a moving average layer based on fiber optic time series data, and obtain a linear prediction sequence through autoregressive moving average processing. This step uses an autoregressive integral moving average model. First, the fiber optic time series data is tested for stationarity. If the data is not stationary, it is made stationary through differencing. The degree of differencing is recorded as a specific value. Stationarity testing is a commonly used method for detecting the stationarity of time series data, which will not be elaborated here. After stationarization, the autoregressive part, i.e., the linear combination of the current sequence value and the sequence values ​​of several past times, is used to characterize the autocorrelation of the data. The specific steps include subtracting the sum of the past time values ​​multiplied by their corresponding autoregressive coefficients from the current time value. Then, the moving average part is considered, using a linear combination of white noise sequences to describe the impact of random fluctuations on the current sequence value, i.e., the current... The time value is added to the sum of the white noise values ​​of each past time point multiplied by the corresponding moving average coefficient. The autoregressive part is combined with the differentially processed sequence, and then an equation is constructed with the moving average part. By solving the parameters in this equation, such as the autoregressive coefficient, the moving average coefficient, and the number of differences, the future time values ​​of the fiber optic time series data are linearly predicted based on these parameters and historical data. For example, by using inverse difference operations, the difference prediction values ​​are accumulated to the previous data to obtain a linear prediction result sequence of the future daily transmission data volume of the original fiber optic time series data. For example, if the difference data volume and estimated parameters of the first two days are known, the predicted value after difference on the third day is calculated, and then added to the original data volume of the second day to obtain the predicted value of the original fiber optic transmission data volume on the third day, and finally a linear prediction sequence is obtained.

[0028] S33. Based on the optical fiber time series data and the linear prediction sequence, nonlinear processing and dynamic fusion are performed to obtain the predicted optical fiber service matrix. S331. Construct a Long Short-Term Memory (LSTM) network layer to process fiber optic time-series data and obtain a nonlinear prediction sequence. In this step, an LSTM network layer is constructed based on an LSTM model. For each fiber optic time-series data, the input gate first concatenates the hidden state from the previous time step with the input data from the current time step. After linear transformation and processing with the sigmoid function, a value between 0 and 1 is obtained. This value determines the proportion of new information flowing into the cell state at the current time step. Then, the forget gate processes the concatenated data in the same way, outputting a value between 0 and 1, which is used to determine the proportion of information retained in the cell state from the previous time step. Finally, based on the input gate and the forget gate... The cell state is updated based on the results. First, the cell state from the previous time step is filtered and retained through a forget gate. Then, new information filtered by the input gate is added. The new information is obtained by linearly transforming the concatenation of the hidden state from the previous time step and the current input, and then processing it through the tanh function. This gives the cell state at the current time step. Finally, the output gate processes the concatenated data to obtain a value between 0 and 1, which is multiplied by the result of the tanh function processing of the cell state at the current time step to obtain the hidden state at the current time step. This hidden state can be used as the prediction output at the current time step or passed to the next time step to continue participating in the calculation. The above process is repeated until all input data is processed, and finally, a nonlinear prediction sequence is obtained.

[0029] S332. Construct an error validation layer to inversely weight the linear and nonlinear prediction sequences to obtain a fused prediction sequence. In this step, firstly, calculate the mean absolute error (MAO) of the moving average layer and the long short-term memory network layer on the validation set. The formula for calculating MAO is a commonly used formula and will not be elaborated here. This error reflects the degree of deviation between the model's predicted value and the true value; the smaller the error, the more accurate the model's prediction. Then, determine the weights of the two prediction sequences based on these two MAOs. The weight of the linear prediction sequence is the ratio of the MAO of the nonlinear prediction sequence to the sum of the MAOs of the two models. The weight of the nonlinear prediction sequence is the ratio of the MAO of the linear prediction sequence to the sum of the MAOs of the two models. Thus, the model with the smaller error will be given a larger weight. Finally, multiply the linear prediction sequence by its weight, and the nonlinear prediction sequence by its weight, then add the two products to obtain the fused prediction sequence.

[0030] S333. Construct a matrix generation layer based on the fused prediction sequence and obtain the predicted fiber service matrix. First, based on the original data of fiber transmission services, clarify the characteristics of different service types and their corresponding resource requirements. For example, video services typically require high bandwidth but are relatively insensitive to latency. Real-time interactive services, such as online games, have strict latency limits and also have certain bandwidth requirements. According to the preset mapping rules between services and resource requirements, split and transform each predicted value in the fused prediction sequence according to the service type. For bandwidth requirements, multiply the basic bandwidth requirement of the service by an adjustment coefficient based on the predicted value to obtain the predicted bandwidth. For example, set the adjustment coefficient to 'a'. Regarding the latency limit, the corresponding value is directly selected from the preset latency limit standard according to the service type, or fine-tuned within a certain range based on the predicted value. The service identifier, the calculated predicted bandwidth, and the predicted latency limit are organized and combined to form a predicted fiber service matrix containing columns such as service identifier, predicted bandwidth, and predicted latency limit. This invention dynamically integrates linear and nonlinear predictions, and uses inverse weighting based on historical errors to make the prediction more accurate, avoiding the limitations of a single model. The fiber service matrix generates a demand matrix related to fiber services, closely matching actual services, providing accurate basis for fiber transmission allocation, and improving the prediction accuracy and practicality of fiber service data.

[0031] S4. Obtain real-time network topology and network resource status. Based on the predicted fiber service matrix, real-time network topology, and network resource status, perform dynamic resource allocation to obtain a fiber resource allocation scheme. Existing technologies mostly use single statistical or shallow machine learning models for prediction, which have poor adaptability to complex and non-stationary service traffic patterns and large prediction errors. To solve the above problems, the specific implementation steps are as follows: S41. Mathematical modeling is performed on the predicted fiber service matrix, real-time network topology, and network resource status to obtain a digital optimization model. The real-time network topology is obtained through network management protocols such as Simple Network Management Protocol (SNMP). Network resource status includes bandwidth status and computing status. Bandwidth status is obtained by monitoring the actual traffic of each link in real time through network devices, and the remaining bandwidth is obtained by comparing it with the maximum capacity of the link. Computing status can be reflected by deploying monitoring agents on servers and other devices to collect data such as CPU utilization and memory utilization in real time. In this step, a multi-constraint objective programming method is used to construct a mathematical optimization model to minimize the total bandwidth congestion rate. The objective is measured by summing the congested traffic caused by insufficient bandwidth for each service and calculating its ratio to the total required traffic. Constraints include bandwidth constraints (each service's allocated bandwidth cannot be less than its predicted bandwidth requirement), latency constraints (the actual latency of service transmission cannot exceed its predicted latency limit), and link capacity constraints (the total bandwidth allocated to all services on each link cannot exceed the actual capacity of that link). By integrating these objectives and constraints, mathematical operations are used to transform data such as service requirements, network topology, and resource status into a digital optimization model containing an objective function and a series of constraint equations. This digital optimization model is mathematically constructed using commonly used fiber-related constraints, which will not be elaborated upon here.

[0032] S42. Construct a chromosome coding layer based on a digital optimization model to obtain the initial population. This step first clarifies the service path information in the fiber optic network. Each chromosome represents a candidate service resource allocation scheme. The transmission path of each service from the source node to the destination node is taken as a gene segment. All service paths are combined to form a complete chromosome. Multiple chromosomes are formed by randomly generating multiple different service path combinations. The collection of these chromosomes constitutes the initial population, i.e., a set of candidate allocation schemes. This process does not involve complex formula calculations; it mainly determines each chromosome, i.e., the specific content of the candidate scheme, by randomly selecting paths based on the network topology and service requirements. For example, suppose there is a simple network topology containing 3 nodes A, B, and C, and 2... Links AB and BC have two services: Service 1 has source node A and destination node C, and Service 2 has source node B and destination node C. For Service 1, the only possible path is A to B to C; for Service 2, the only possible path is B to C. Using a path-based chromosome coding method, an initial population is randomly generated. The first candidate allocation scheme is in chromosome 1, where Service 1 is assigned the path A to B to C, and Service 2 is assigned the path B to C. The second candidate allocation scheme is in chromosome 2, where Service 1 is assigned the path A to B to C, and Service 2 is also assigned the path B to C. In this example, Service 2 has limited path choices. In more complex networks, there are more choices. By randomly generating multiple such different combinations, such as 10, 10 chromosomes are obtained, which are the initial populations of candidate allocation schemes.

[0033] S43. Perform genetic iteration processing based on the initial population to obtain the fiber optic resource allocation scheme; S431. Based on the initial population, an optimized population is obtained through population optimization. In this step, the fitness function is first calculated as 1 divided by 1 plus the objective function value corresponding to each individual in the population. The fitness value is obtained through the fitness function, which transforms the objective function value into a fitness value. The smaller the objective function value, the larger the fitness value, indicating a better individual. The objective function can be minimizing the fiber optic transmission response time. In practice, the objective function is flexibly set according to the needs of fiber optic transmission. Then, a roulette wheel selection method is used to determine the probability of an individual being selected based on the proportion of its fitness value to the total fitness value of the population. Individuals with higher fitness values ​​have a greater probability of being selected from the initial population. A subset of individuals are introduced into the next generation of the population. Then, a single-point crossover operation is performed. Two individuals are randomly selected from the initial population, and a crossover point is randomly determined, which is a data point in the two data sequences. The parts of the two individuals after the crossover point are swapped to generate new individuals, increasing the diversity of the population. Finally, a random position mutation operation is performed, where certain gene positions in the individuals are randomly selected and their values ​​are changed to further explore the search space. The number of iterations is set to K. After K rounds of selection, crossover, and mutation iterations, the optimized population is obtained. Among them, the roulette wheel selection method, single-point crossover operation, and random position mutation operation are commonly used selection, crossover, and mutation operations in genetic algorithms, which will not be elaborated here.

[0034] S432. Construct a Q-learning layer to fine-tune the optimization population, resulting in a tuned population. In this step, the real-time network load distribution is defined as the real-time network state. Fine-tuning operations on candidate solutions in the optimization population, such as path switching, are treated as actions. After taking an action, the reward value is determined based on the improvement of the objective function before and after the implementation of the solution. Using the Q-learning algorithm, the Q-value is iteratively updated according to the Q-value update formula. The Q-value update formula is: the new Q-value equals the learning rate multiplied by the sum of the maximum Q-values ​​among all possible actions in the next state, minus the difference of the old Q-values, and the product is added to the old Q-value. The learning rate controls the degree of updating of the old Q-value with new information, and the discount factor measures the degree of influence of future rewards on the current decision. The learning rate can be set to 0.1, and the discount factor can be set to 0.9. By continuously exploring different actions in the state space, the state with the larger Q-value is selected as the optimal action. After K rounds of iteration, the allocation scheme adjustment strategy that can make the objective function reach the optimal is finally determined based on the Q-value. The best allocation scheme tuned by reinforcement learning is output and integrated into the tuned population.

[0035] S433. Feasibility verification is performed based on the optimized population, and a fiber optic resource allocation scheme is obtained. This step first clarifies various constraints existing in the fiber optic network, such as the upper limit of fiber bandwidth capacity, the maximum number of connections at a specific node, and the tolerance range of services for transmission delay. Then, the optimal allocation scheme corresponding to the optimized population is checked one by one to see if it meets these constraints. If any are not met, the scheme is corrected according to preset adjustment rules. The adjustment process may involve replanning paths or reallocating bandwidth. After repeated checks and adjustments, the scheme is continued until all constraints are met. Finally, an optimized fiber optic path and resource allocation scheme that meets the actual network operation requirements is output, including specific path information and the bandwidth allocated to each path. For example, if a service needs to transmit from node A to node D, the optimal allocation scheme obtained through reinforcement learning optimization is: path 1 is ABCD, allocated bandwidth 20Gbps, and path 2 is AEFD, allocated bandwidth 15Gbps. However, the upper limit of fiber bandwidth capacity in this network is 25Gbps. Obviously, the bandwidth allocation of path 1 exceeds the constraints. In this case, the constraint satisfaction check method is used for adjustment. The process involves adjusting the bandwidth allocation of a path, identifying path 3 as AGHD, with an additional bandwidth limit of 10Gbps. Based on a bandwidth capacity limit of 25Gbps, the adjusted scheme is as follows: path 1 is allocated 15Gbps, path 2 15Gbps, and path 3 5Gbps. At this point, the bandwidth allocation for all paths satisfies the fiber bandwidth capacity limit constraint. The optimized fiber resource allocation scheme is output, including fiber path, bandwidth allocation, wavelength allocation, and time slot allocation, among other fiber transmission information. This invention integrates multi-constraint objective programming, genetic algorithms, reinforcement learning, and constraint satisfaction checking to form a comprehensive and refined optimization system. Multi-constraint objective programming provides accurate modeling that aligns with actual network requirements. Genetic algorithms offer strong global search capabilities, quickly finding near-optimal solutions. Reinforcement learning further refines local optimization, improving scheme quality. Constraint satisfaction checking ensures scheme feasibility. The close collaboration of each step effectively balances global and local considerations, efficiency and quality. Compared to existing single technologies or simple combinations, this invention more efficiently and accurately yields multi-constraint-compliant and performance-optimized fiber path and resource allocation schemes.

[0036] S5. Based on the fiber optic resource allocation scheme, perform scheme detection and self-healing processing to obtain a fiber optic network status report. Existing network monitoring and fault handling often rely on threshold alarms and predefined scripts, which are slow to respond, lack flexibility, and cannot cope with unknown and complex anomalies. For example, while the packet loss rate slightly increases, the routing convergence time slightly increases. Individually, these two factors may not exceed the threshold, but the combined effect causes video conferencing to lag. To solve the above problems, the specific implementation steps are as follows: S51. Based on the fiber optic resource allocation scheme and real-time network monitoring data, a real-time multi-dimensional indicator vector is obtained through indicator detection and processing. This step allocates fiber optic resources based on the fiber optic resource allocation scheme and obtains data under real-time network monitoring, including real-time bandwidth utilization, transmission latency, and packet loss rate. First, based on the fiber optic resource allocation scheme, the types of key performance indicators that need to be monitored are identified, such as bandwidth utilization, transmission latency, and packet loss rate. Raw data related to these indicators is continuously extracted from the real-time network monitoring data. For each indicator, the raw data is processed according to its specific calculation logic. For example, for bandwidth utilization, the actual bandwidth used at a certain moment is divided by the total bandwidth allocated to that path; for transmission latency, it is calculated by recording the time difference between the sending and receiving ends of data packets. By calculating and integrating various types of raw data according to corresponding rules, a real-time multi-dimensional indicator vector containing multiple key performance indicator values ​​is finally output.

[0037] S52. Construct an isolated forest layer based on real-time multi-dimensional indicator vectors and obtain anomaly indicators through anomaly indicator detection. In this step, firstly, a feature dimension is randomly selected in the feature space formed by the real-time multi-dimensional indicator vectors, and then a split value is randomly determined on that dimension to construct a binary tree structure. The sample data is gradually isolated into different child nodes. Because the distribution of anomalies differs greatly from that of normal data, they can usually be isolated with fewer splits, that is, the path length from the root node to the leaf node is shorter. By calculating the average path length of each sample on multiple isolated trees, and calculating the anomaly score according to a specific formula, the anomaly score is compared with the set threshold to determine whether the sample is anomaly. Finally, the anomaly detection flag and specific anomaly indicators are output. The formula for calculating the anomaly score is the commonly used formula for calculating the anomaly score in isolated forests, which will not be elaborated here.

[0038] S53. Construct a historical anomaly detection layer based on anomaly indicators, and obtain historical anomaly data through nearest neighbor similarity processing. This step first filters out all data marked as anomaly from the historical database, extracts the anomaly feature data of these data, and calculates the similarity between the features corresponding to the current input anomaly indicators and the corresponding features in the historical anomaly data one by one. When calculating the similarity, the Euclidean distance formula is used for quantification. By squaring, summing, and square rooting the difference between the current anomaly features and the features of the historical anomaly data, the similarity value is obtained. The historical anomaly data is sorted according to the similarity value, and the top S cases with the highest similarity are selected as the retrieval results and the historical anomaly data is output.

[0039] S54. Based on historical anomaly data, an adjusted solution is obtained through integration using a similarity scheme. This step first extracts and analyzes the handling strategies employed for each similar historical anomaly, identifying key operational steps and parameter settings within these strategies. Then, considering constraints such as the current network's real-time status, resource availability, and service priorities, the extracted handling strategies are adaptively adjusted. This adjustment may involve changing the order of certain operational steps, modifying or adding / removing parameter values, and adding necessary operational steps based on the current network characteristics. By integrating the handling strategies of multiple similar cases and adaptively adjusting them, several feasible self-healing strategies are formed, ultimately outputting the adjusted solution. For example, the handling strategy for Case 1 is: first, adjust the bandwidth of some low-priority services from 5Mbps to 2Mbps, and then allocate the freed-up 3Mbps bandwidth to high-priority services; the handling strategy for Case 2 is: directly shut down some non-critical services, releasing 8Mbps bandwidth to ensure core services; The handling strategy in Example 3 is: activate the backup link and switch some service traffic to the backup link, which has a bandwidth of 6Mbps. The current network status is: resources are relatively tight, and it is not possible to directly shut down too many services, but there is a backup link with a bandwidth of 4Mbps that can be activated. At the same time, the bandwidth range that low-priority services can adjust is limited. Based on these constraints, the above strategy is adaptively adjusted as follows: Combine the strategies of Example 1 and Example 3 to generate a new strategy 1: adjust the bandwidth of low-priority services from 5Mbps to 3Mbps, release 2Mbps of bandwidth, and activate the backup link to switch some service traffic to the backup link, thus comprehensively protecting high-priority services; Combine the strategies of Example 2 and Example 3 to generate a new strategy 2: instead of directly shutting down non-critical services, activate the backup link to switch some traffic of core services to the backup link, and simultaneously rate-limit non-critical services, restricting their bandwidth usage to no more than 2Mbps; the final output candidate self-healing strategy set includes the two feasible adjusted solutions, new strategy 1 and new strategy 2.

[0040] S55. Based on the adjusted scheme, a fiber optic network status report is generated through simulation evaluation. This step first constructs a simplified network simulation environment based on the various operation steps and parameter settings in the adjusted scheme, combined with key information such as network topology, traffic distribution, and device performance in the real-time network status. In this simulation environment, the network is virtually operated according to the adjusted scheme, simulating the execution of various self-healing strategies, such as bandwidth adjustment, route switching, and device restart. During the simulation, key network performance indicators, such as bandwidth utilization, transmission latency, and packet loss rate, are monitored in real time. The simulation results are compared with preset reasonable thresholds to determine the network health status. A network health status report is generated based on the results. If the simulation results do not meet the requirements for normal network operation, a self-healing adjustment command is generated. For example, in the current real-time network state, a critical link has an original bandwidth of 100Mbps, a bandwidth utilization rate of 0.7 before adjustment, and a bandwidth increase of 20Mbps after adjustment. Simultaneously, it is estimated that the new traffic in the real-time network will be 15Mbps. The original bandwidth plus the bandwidth increase is calculated to obtain the changed bandwidth value. The bandwidth utilization rate is obtained by multiplying the changed bandwidth value by the original bandwidth utilization rate, adding the new traffic value, and dividing by the changed bandwidth value. A bandwidth utilization threshold is then set based on historical data. The value is Z, for example, the bandwidth utilization threshold is 0.8. In actual calculations, if the actual bandwidth utilization is greater than or equal to the bandwidth utilization threshold, the network status is at risk, and a potential risk detection result is generated. Conversely, if the actual bandwidth utilization is less than or equal to the threshold, the network status is normal, and a normal network result is generated. The fiber optic network status report includes the content of the adjusted plan, as well as key data related to fiber optic network transmission, such as bandwidth utilization, transmission latency, packet loss rate, equipment operating status, and memory utilization. This invention uses the isolated forest anomaly detection algorithm and the case-based reasoning self-healing decision model. The isolated forest efficiently identifies anomalies in multi-dimensional monitoring indicators, while the case-based reasoning uses a historical handling experience database to match and adjust the optimal handling strategy for the current anomaly. The combination of the two realizes a complete processing flow from rapid anomaly perception to intelligent decision-making, improving the robustness of the system and the accuracy of the fiber optic network adjustment plan.

[0041] Example 2: Due to the difficulty of simultaneously considering the linear and nonlinear characteristics of data in existing technologies, the lack of a scientifically sound weight allocation method, and the insufficient consideration of business characteristics, the prediction results do not match the actual needs of optical fibers. For further information, please refer to [link to relevant documentation]. Figure 2 The diagram shown is a structural block diagram of the cloud computing-based dynamic allocation system for optical fiber transmission resources provided in this embodiment. The system includes an acquisition module, a standardization module, a prediction module, an allocation module, and a status reporting module. The acquisition module is used to collect raw network device operation data and raw service request data from the fiber optic transmission system, and obtain multi-source datasets through preprocessing. The standardization module is used to perform feature extraction and standardization processing on multi-source datasets to obtain standardized data. The forecasting module is used to obtain a forecasted fiber optic service matrix based on standardized data through demand forecasting processing. The allocation module is used to obtain real-time network topology and network resource status, and to perform dynamic resource allocation based on the predicted fiber service matrix, real-time network topology and network resource status to obtain a fiber resource allocation scheme. The status report module is used to perform scheme detection and self-healing based on the fiber optic resource allocation scheme, and to obtain a fiber optic network status report.

[0042] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, this application can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code, including but not limited to disk storage, CD-ROM, optical storage, etc.

[0043] The above embodiments provide a detailed description of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for dynamic allocation of fiber optic transmission resources based on cloud computing, characterized in that, The method involves the following steps: collecting raw network device operation data and raw service request data from the fiber optic transmission system, and obtaining a multi-source dataset through preprocessing; Perform feature extraction and standardization on multi-source datasets to obtain standardized data; Based on standardized data, a predicted fiber optic service matrix is ​​obtained through demand forecasting. The system acquires real-time network topology and network resource status, and performs dynamic resource allocation based on the predicted fiber service matrix, real-time network topology, and network resource status to obtain a fiber resource allocation scheme. Based on the fiber optic resource allocation scheme, scheme detection and self-healing are performed to obtain a fiber optic network status report.

2. The method for dynamic allocation of fiber optic transmission resources based on cloud computing according to claim 1, characterized in that, The preprocessing process yields a multi-source dataset, including: distributed probe collection and processing based on raw network device operation data and raw service request data to obtain initial network status data and initial service data. Based on the initial network status data and initial business data, compliance checks are conducted to obtain compliance status data and compliance business data; Time alignment is performed on compliance status data and compliance business data to obtain a multi-source dataset.

3. The method for dynamic allocation of fiber optic transmission resources based on cloud computing according to claim 1, characterized in that, Feature extraction and standardization of multi-source datasets include: obtaining cleaned data from multi-source datasets through data cleaning. Feature extraction is performed on the cleaned data to obtain a primary feature vector; Standardized data is obtained by standardizing the primary feature vectors.

4. The method for dynamic allocation of fiber optic transmission resources based on cloud computing according to claim 1, characterized in that, Demand forecasting based on standardized data includes: obtaining fiber optic time-series data through time-series processing based on standardized data; A moving average layer is constructed based on fiber optic time series data, and a linear prediction sequence is obtained through autoregressive moving average processing. The predicted fiber service matrix is ​​obtained by nonlinear processing and dynamic fusion of fiber time series data and linear prediction sequences.

5. The method for dynamic allocation of optical fiber transmission resources based on cloud computing according to claim 4, characterized in that, Nonlinear processing and dynamic fusion of fiber optic time-series data and linear prediction sequences are performed, including: constructing a long short-term memory network layer, processing the fiber optic time-series data using the long short-term memory network, and obtaining nonlinear prediction sequences; An error verification layer is constructed, and the linear and nonlinear prediction sequences are inversely weighted by error to obtain a fused prediction sequence. A matrix generation layer is constructed based on the fused prediction sequence, and the predicted fiber service matrix is ​​obtained.

6. The method for dynamic allocation of optical fiber transmission resources based on cloud computing according to claim 1, characterized in that, Dynamic resource allocation based on predicted fiber service matrix, real-time network topology and network resource status includes: mathematical modeling of predicted fiber service matrix, real-time network topology and network resource status to obtain digital optimization model; The chromosome coding layer was constructed based on a digital optimization model to obtain the initial population. Genetic iteration processing is performed on the initial population to obtain the fiber optic resource allocation scheme.

7. The method for dynamic allocation of optical fiber transmission resources based on cloud computing according to claim 6, characterized in that, Genetic iteration processing is performed based on the initial population, including: obtaining an optimized population through population optimization based on the initial population; A Q-learning layer is constructed to fine-tune the optimized population, resulting in a tuned population. Feasibility was verified based on the optimized population, and a fiber optic resource allocation scheme was obtained.

8. The method for dynamic allocation of optical fiber transmission resources based on cloud computing according to claim 1, characterized in that, Scheme detection and self-healing based on fiber optic resource allocation schemes include: obtaining real-time multi-dimensional indicator vectors through indicator detection and processing based on fiber optic resource allocation schemes and real-time network monitoring data. An isolated forest layer is constructed based on real-time multi-dimensional indicator vectors, and abnormal indicators are obtained through anomaly indicator detection. A historical anomaly detection layer is constructed based on anomaly indicators, and historical anomaly data is obtained through nearest neighbor similarity processing.

9. A system applied to the cloud computing-based dynamic allocation method for optical fiber transmission resources as described in any one of claims 1-8, characterized in that, The system includes: The acquisition module is used to collect raw network device operation data and raw service request data from the fiber optic transmission system, and obtain multi-source datasets through preprocessing. The standardization module is used to perform feature extraction and standardization processing on multi-source datasets to obtain standardized data. The forecasting module is used to obtain a forecasted fiber optic service matrix based on standardized data through demand forecasting processing. The allocation module is used to obtain real-time network topology and network resource status, and to perform dynamic resource allocation based on the predicted fiber service matrix, real-time network topology and network resource status to obtain a fiber resource allocation scheme. The status report module is used to perform scheme detection and self-healing based on the fiber optic resource allocation scheme, and to obtain a fiber optic network status report.

Images