Optical fiber link degradation state prediction method and system for cable television network

Abstract

The invention relates to the field of optical fiber link state management, and particularly discloses an optical fiber link degradation state prediction method and system for a cable television network, and the method comprises the steps: respectively extracting the dynamic evolution characteristics of a network performance index and external context information through a dual-channel time sequence coding structure; then, a cross-modal information fusion mechanism is utilized, so that the model can focus and extract influence information with the most correlation from a complex historical context intelligently according to the current internal state of the link; according to the scheme, a vulnerability gating adaptive adjustment mechanism is further introduced, and the mechanism can dynamically adjust the actual influence of an external context based on the current internal health condition of the link, so that the physical reality that the vulnerable link is more easily influenced by external impact is accurately simulated. In this way, deep modeling of the nonlinear coupling effect of internal and external factors is achieved, and the accuracy and robustness of degradation prediction are improved.

Description

Technical Field

[0001] This application relates to the field of fiber optic link status management, and more specifically, to a method and system for predicting the degradation status of fiber optic links in cable television networks. Background Technology

[0002] As a critical information infrastructure carrying various services such as video, data, and voice, the stability and reliability of the physical layer of cable television networks—the fiber optic link—are paramount. The performance of fiber optic links gradually deteriorates over time due to factors such as component aging, environmental changes, or human interference. Failure to detect and address this in a timely manner can ultimately lead to service interruptions, severely impacting user experience and increasing operators' maintenance costs. Therefore, shifting from a passive fault response model to a proactive predictive maintenance model, and developing a solution capable of accurately predicting the deterioration state of fiber optic links to provide early warnings of potential risks, has significant practical and commercial value for ensuring the service quality of cable television networks, improving operational efficiency, and reducing operating costs.

[0003] Current technologies for monitoring and predicting the status of fiber optic links primarily rely on setting static thresholds. Alarms are generated when monitored network performance indicators (such as optical power and signal-to-noise ratio) exceed these thresholds. This approach can only react when a fault occurs or is imminent, lacking predictive capabilities. With the development of data analysis techniques, some solutions have begun to employ time-series analysis-based models (such as ARIMA) or traditional machine learning models, learning from historical data of network indicators to predict future trends or the probability of degradation. However, these methods generally suffer from fundamental limitations: they typically treat the fiber optic link as a closed system isolated from the outside world, limiting their analysis and modeling data to performance indicators generated by the network equipment itself. This information cocoon-like modeling ignores a crucial fact: fiber optic link performance degradation is often induced or accelerated by external environmental factors or unforeseen events. For example, the performance of internal optical components in outdoor optical node equipment is affected by drastic seasonal temperature changes; external interference such as municipal construction or vehicle collisions around the line can also cause stress damage to the optical cable, leading to instantaneous performance fluctuations or long-term slow degradation. Existing technical solutions lack systematic collection and modeling of these dynamic environmental contexts and discrete event contexts, resulting in a significant decrease in accuracy and robustness of their prediction models when faced with deterioration scenarios driven by external factors. They are unable to capture the complex deterioration process under the coupling of internal and external factors.

[0004] Therefore, an optimized scheme for predicting the degradation status of fiber optic links in cable television networks is desired. Summary of the Invention

[0005] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a method and system for predicting the degradation status of fiber optic links in cable television networks.

[0006] According to one aspect of this application, a method for predicting the degradation status of optical fiber links in a cable television network is provided, comprising: Obtain network metric time series, environment context time series, and event context data streams; The environment context time series and event context data stream are merged using fast variable sequences to obtain the context parameter time series. The network indicator time series and the context parameter time series are encoded using a dual-channel LSTM feature sequence to obtain the network indicator hidden state sequence and the context information hidden state sequence. Cross-modal information fusion is performed on the hidden state sequence of network indicators and the hidden state sequence of context information to obtain the fused context vector; Extract the current hidden state of the network indicators from the hidden state sequence of network indicators; Degradation state prediction is performed based on the current network indicator hidden state and the fused context vector to obtain the degradation prediction probability.

[0007] According to another aspect of this application, a fiber optic link degradation state prediction system for a cable television network is provided, comprising: The time-series data acquisition module is used to acquire network indicator time series, environmental context time series, and event context data streams; The fast variable sequence merging module is used to merge environment context time series and event context data streams to obtain context parameter time series. The time-series coding module is used to perform feature sequence coding based on dual-channel LSTM on the network indicator time series and the context parameter time series to obtain the network indicator hidden state sequence and the context information hidden state sequence. The cross-modal state fusion module is used to perform cross-modal information fusion on the network indicator hidden state sequence and the context information hidden state sequence to obtain the fused context vector; The current network state extraction module is used to extract the current hidden state of network indicators from the hidden state sequence of network indicators; The degradation state prediction module is used to predict the degradation state based on the current hidden state of network indicators and the fused context vector to obtain the degradation prediction probability.

[0008] Compared with existing technologies, this application provides a method and system for predicting the degradation status of optical fiber links in cable television networks. First, it uses a dual-channel time-series coding structure to extract the dynamic evolution characteristics of network performance indicators and external context information. Then, utilizing a cross-modal information fusion mechanism, the model can intelligently focus on and extract the most relevant impact information from complex historical contexts based on the current internal state of the link. The scheme further introduces an adaptive adjustment mechanism for vulnerability gating. This mechanism can dynamically adjust the actual influence of the external context based on the current internal health of the link, thereby accurately simulating the physical reality that vulnerable links are more susceptible to external shocks. Through this complete link of perception-fusion-modulation-prediction, this scheme achieves deep modeling of the nonlinear coupling effects of internal and external factors, effectively overcoming the shortcomings pointed out in the background technology, and thus significantly improving the accuracy and robustness of degradation prediction. Attached Figure Description

[0009] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 This is a flowchart of a fiber optic link degradation state prediction method for cable television networks according to an embodiment of this application; Figure 2 This is a schematic diagram of data flow in a fiber optic link degradation state prediction method for cable television networks according to an embodiment of this application. Figure 3 This is a flowchart illustrating the cross-modal information fusion of network indicator hidden state sequences and context information hidden state sequences to obtain a fused context vector, according to an embodiment of the optical fiber link degradation state prediction method for cable television networks based on this application. Figure 4 This is a flowchart illustrating a method for predicting the degradation state of an optical fiber link in a cable television network according to an embodiment of this application. The method predicts the degradation state based on the current hidden state of network indicators and the fused context vector to obtain the degradation prediction probability. Figure 5 This is a block diagram of an optical fiber link degradation state prediction system for a cable television network according to an embodiment of this application. Detailed Implementation

[0011] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0012] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not specifically singular and may include plural forms. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0013] While this application makes various references to certain modules of the systems according to embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The modules described are merely illustrative, and different aspects of the systems and methods may use different modules.

[0014] Flowcharts are used in this application to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0015] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0016] To address the technical problem of insufficient prediction accuracy in existing prediction methods that focus only on internal network indicators while neglecting the influence of external environment and events, this application proposes a method for predicting the degradation status of fiber optic links in cable television networks. The specific implementation process begins with simultaneously acquiring time series of network indicators representing the internal performance of the link, as well as external context data including continuous environmental information such as temperature and humidity, and discrete events such as construction and severe weather. First, to handle heterogeneous external information, the scheme unifies the environmental and event data into a regular context parameter time series through alignment, structuring, and vectorization operations. Next, a dual-channel long short-term memory network architecture is used to encode the network indicator sequence and context parameter sequence in parallel to capture their inherent dynamic evolution patterns and generate their respective hidden state sequences. Based on this, the scheme introduces a crucial cross-modal information fusion step. Using the current hidden state of the network indicators as the query, an attention mechanism is used to search through historical hidden state sequences of context information, thereby intelligently extracting the external influences most relevant to the current link state and converging them into a fused context vector. Furthermore, to simulate the modulating effect of link vulnerability on the degree of impact from external shocks in the real physical world, the scheme designs a link vulnerability gating unit. This unit generates a gating vector using the current hidden state of network metrics and adaptively adjusts the influence of this gating vector on the fused context vector. Finally, the hidden state representing the current intrinsic state of the link is aggregated with this dynamically adjusted context vector and fed into a classifier, thus outputting a degradation prediction probability that accurately reflects the complex interaction between internal and external factors.

[0017] The technical solution of this application proposes a method for predicting the degradation status of optical fiber links in cable television networks. Figure 1 This is a flowchart of a fiber optic link degradation state prediction method for cable television networks according to an embodiment of this application. Figure 2 This is a schematic diagram of data flow in a fiber optic link degradation state prediction method for cable television networks according to an embodiment of this application. Figure 1 and Figure 2As shown, the fiber optic link degradation state prediction method for cable television networks according to an embodiment of this application includes the following steps: S100, acquiring network indicator time series, environmental context time series, and event context data stream; S200, performing fast variable sequence merging on the environmental context time series and event context data stream to obtain a context parameter time series; S300, performing feature sequence encoding based on dual-channel LSTM on the network indicator time series and context parameter time series to obtain a network indicator hidden state sequence and a context information hidden state sequence; S400, performing cross-modal information fusion on the network indicator hidden state sequence and the context information hidden state sequence to obtain a fused context vector; S500, extracting the current network indicator hidden state from the network indicator hidden state sequence; S600, performing degradation state prediction based on the current network indicator hidden state and the fused context vector to obtain a degradation prediction probability.

[0018] Specifically, in step S100, network indicator time series, environmental context time series, and event context data stream are acquired. It should be understood that relying solely on the network performance indicators of the fiber optic link itself cannot fully characterize the complex and nonlinear degradation process caused by changes in the external environment and sudden events, leading to an incomplete model input feature space and thus limiting the accuracy and generalization ability of the prediction model. Therefore, in the technical solution of this application, by acquiring network indicator time series, environmental context time series, and event context data stream, a complete multimodal data foundation covering the intrinsic state evolution and exogenous driving factors of the link is constructed. This provides the necessary data support for subsequent models to reveal and learn the deep coupling relationship between the internal state of the link and external influences, thereby laying the foundation for accurate and robust degradation state prediction.

[0019] More specifically, in a specific example of this application, the process of obtaining network indicator time series includes: periodically polling devices such as optical nodes or optical receivers deployed in the network using standard network management protocols, such as Simple Network Management Protocol (SNMP). This polling operation can collect key indicators reflecting the intrinsic performance status of the link, including but not limited to optical receiver input optical power, radio frequency signal-to-noise ratio, and bit error rate, and aggregate these indicator values ​​with high-precision timestamps to form a dynamic sequence characterizing the evolution of the internal state of the link over time. The implementation method for obtaining environmental context time series is as follows: IoT terminal devices integrating temperature, humidity, and vibration sensors are installed at or near the deployment locations of key fiber optic network equipment, such as outdoor optical node cabinets. These terminal devices periodically collect physical environment data and report the data to a unified IoT data platform via a wireless communication network. The system in this solution obtains continuous environmental parameters of a specified device within a corresponding time period by calling the platform's application programming interface (API), thus forming an environmental context time series. The execution method for obtaining event context data streams includes: the system actively connecting to multiple third-party information service systems through the application programming interface (API). For example, it connects to meteorological service systems to obtain early warnings of severe weather such as typhoons and blizzards; it connects to municipal management information systems to obtain planned events such as road construction and pipeline excavation within specific areas; and it connects to the operator's own operation and maintenance work order system to obtain planned maintenance or network cutover events. The system integrates these non-periodic event information from different sources into a discrete data stream containing attributes such as event time, type, and scope of impact.

[0020] Specifically, in step S200, the environmental context time series and the event context data stream are merged using fast variable sequences to obtain the context parameter time series. It should be understood that environmental context data and event context data obtained from different information systems differ significantly in data format, sampling frequency, and time base. For example, environmental data is continuous time series, while event data is discrete and unstructured, making it impossible to directly process them uniformly using a time series model. Therefore, in the technical solution of this application, the environmental context time series and the event context data stream are further merged using fast variable sequences to obtain the context parameter time series. This transforms and integrates two different sources and varying properties of external influence information into a unified, time-aligned, and numerical time series data structure. This provides a well-organized and processable context input matrix for the subsequent dual-channel coding model, enabling the model to simultaneously analyze the combined impact of gradual environmental changes and sudden external events on the fiber optic link.

[0021] More specifically, in this embodiment of the application, merging the environmental context time series and the event context data stream using fast variable sequences to obtain the context parameter time series includes: aligning the environmental context time series based on a reference time grid to obtain an aligned environmental context time series; structuring and vectorizing the discrete event data in the event context data stream based on the reference time grid to obtain a vectorized event time series; and merging the aligned environmental context time series and the vectorized event time series to obtain the context parameter time series.

[0022] In a specific example of this application, the process of aligning environmental context time series data based on a reference time grid involves first determining a unified reference time grid whose time points are consistent with those of the network index time series. Then, for environmental context time series with sampling frequencies inconsistent with this grid, such as temperature data reported hourly, linear interpolation or forward / backward padding methods are used to calculate or fill in the corresponding environmental parameter values ​​at each time point of the reference time grid, thereby generating an aligned environmental context time series that is strictly time-aligned with the network index series.

[0023] Next, the process of structuring and vectorizing the discrete event data stream based on the reference time grid includes the following steps: For each discrete event, such as a construction event occurring at a certain time point, the event's timestamp is first mapped to the nearest time point on the reference time grid. Then, multiple attributes of the event, such as event type, severity level, and text description, are numerically encoded. For example, the event type is one-hot encoded, the severity level is normalized, and the text description is converted into a fixed-dimensional vector using a word embedding model. These encoded values ​​are concatenated into an event feature vector and placed at the corresponding time point on the time grid. Other time points in the grid that are not where the event occurred are filled with zero vectors, thus forming a vectorized event time series.

[0024] Finally, the sequence merging of the aligned environmental context time series and the vectorized event time series involves concatenating the environmental vector and event vector at each time point of the reference time grid along the feature dimension, forming a unified context vector with a wider dimension that includes all external context information at that moment. The unified context vectors from all time points are then stacked chronologically to form a complete context parameter time series that can be used as input to the model. Specifically, in this embodiment, merging the aligned environmental context time series and the vectorized event time series to obtain the context parameter time series includes: concatenating the features of each corresponding set of environmental vectors and event vectors in the aligned environmental context time series and the vectorized event time series to obtain the context parameter time series.

[0025] Specifically, in step S300, the network indicator time series and the context parameter time series are encoded using a dual-channel LSTM-based feature sequence to obtain the network indicator hidden state sequence and the context information hidden state sequence. It should be understood that since the network indicator time series and the context parameter time series represent two types of information with drastically different physical meanings and dynamic characteristics—the former reflecting the endogenous state evolution of the link, and the latter characterizing the driving factors of the external environment—simply merging them and processing them with a single model would confuse their unique temporal dependencies, making it difficult to effectively separate and learn their independent contributions to the degradation process. Therefore, in the technical solution of this application, the network indicator time series and the context parameter time series are further encoded using a dual-channel LSTM-based feature sequence to obtain the network indicator hidden state sequence and the context information hidden state sequence. This allows for the independent extraction of their respective high-order dynamic features over time from the data of the two modalities through parallel, dedicated temporal encoding channels. This generates two sets of feature sequences containing historical memory and highly condensed information, providing a clearly structured and feature-rich input for subsequent meaningful cross-modal information fusion and interactive modeling.

[0026] More specifically, in a concrete example of this application, this step first inputs the network indicator time series generated in the preceding steps into a separately configured network indicator encoder, which is a neural network composed of one or more Long Short-Term Memory (LSTM) layers stacked together. The network indicator encoder processes the network indicator vector sequence sequentially over time steps. At each time step, the LSTM unit combines the current network indicator vector with the hidden state and unit state from the previous time step, and updates and outputs the current hidden state through its internal forget gate, input gate, and output gate structure. This process is repeated until the entire input sequence is processed. The hidden states from all time steps constitute the network indicator hidden state sequence, which at each time point encapsulates the network performance evolution history from the beginning of the sequence to the current time step. Furthermore, this step performs the encoding of the context parameter time series in parallel. The context parameter time series is input into another independent context information encoder, which is also based on an LSTM architecture. The network structure and parameters of this context information encoder are independent of the network indicator encoder and can be specifically configured according to the characteristics of the context data. It processes the sequence of context parameter vectors step by step in the same way, and outputs a hidden state of context information at each time point that captures the pattern of changes in the external environment and events from the beginning of the sequence to the current time. The hidden states of context information at all time points are combined to form the sequence of hidden states of context information.

[0027] Specifically, in step S400, cross-modal information fusion is performed on the network indicator hidden state sequence and the context information hidden state sequence to obtain a fused context vector. It should be understood that since the network indicator hidden state sequence and the context information hidden state sequence obtained through separate encoding are two independent, uncorrelated high-order feature sequences, direct subsequent processing would fail to reflect the differences in the impact of the external environment on the current link state at different historical moments, i.e., it fails to establish a dynamic correlation between the current link state and historical external influences. Therefore, in the technical solution of this application, cross-modal information fusion is further performed on the network indicator hidden state sequence and the context information hidden state sequence to obtain a fused context vector. This establishes an attention mechanism driven by endogenous states, using the current intrinsic state of the link as a query guide to dynamically and selectively focus on and aggregate the most relevant information from the entire historical flow of external influences. This generates a highly condensed, non-static context representation strongly correlated with the current state, providing an effective external factor vector that has been intelligently filtered and refined for subsequent accurate prediction.

[0028] Figure 3This is a flowchart illustrating cross-modal information fusion of network indicator hidden state sequences and context information hidden state sequences according to embodiments of this application to obtain a fused context vector. Figure 3 As shown, step S400 includes: S410, extracting the current network indicator hidden state from the network indicator hidden state sequence; S420, performing a linear transformation on the current network indicator hidden state to obtain the network indicator query vector; S430, performing a linear transformation on the context information hidden state sequence to obtain the context information key matrix and the context information value matrix; S440, inputting the network indicator query vector, the context information key matrix, and the context information value matrix into the converter module to obtain the fused context vector.

[0029] More specifically, in a concrete example of this application, this step is first implemented by extracting the current network indicator hidden state from the sequence of network indicator hidden states. This current network indicator hidden state is the hidden state vector of the last time step in the sequence, because it mathematically encapsulates all the network performance evolution information from the beginning of the sequence to the current time step.

[0030] Then, the process of performing a linear transformation on the current network indicator hidden state to obtain the network indicator query vector, and simultaneously performing a linear transformation on the entire context information hidden state sequence to obtain the context information key matrix and the context information value matrix, is as follows: the current network indicator hidden state is projected into a query vector through three independent fully connected neural network layers with learnable parameters, and each hidden state vector in the context information hidden state sequence is projected into a key vector and a value vector, respectively. All key vectors and value vectors constitute the context information key matrix and the context information value matrix, respectively.

[0031] Finally, the network metric query vector, context information key matrix, and context information value matrix are input into the converter module to obtain the fused context vector. Inside the converter module, the dot product similarity between the network metric query vector and each key vector in the context information key matrix is ​​first calculated. The obtained similarity score is scaled and then normalized into a set of attention weights using a Softmax function. This set of weights clearly quantifies the importance of each historical context state to the current network state. Finally, this set of attention weights is used as coefficients to perform a weighted summation of the value vectors in the context information value matrix. The final summation result is the fused context vector. That is, in this embodiment, inputting the network metric query vector, context information key matrix, and context information value matrix into the converter module to obtain the fused context vector includes: the converter module processes the network metric query vector, context information key matrix, and context information value matrix using the following formula to obtain the fused context vector: , in, For network metric query vectors, For context information key matrix, For context information value matrix, Let be the dimension of each row vector in the context information key matrix. for function, This is the fused context vector.

[0032] Specifically, in step S500, the current network indicator hidden state is extracted from the network indicator hidden state sequence. It should be understood that since the network indicator hidden state sequence generated in the preceding steps is a set of sequences containing states from multiple historical time points, and subsequent decision-making processes such as cross-modal fusion and degradation prediction require a comprehensive and single state representation that can represent the link up to the current moment. Therefore, in the technical solution of this application, the current network indicator hidden state is further extracted from the network indicator hidden state sequence to accurately separate the final state vector encapsulating all historical information from the complete temporal evolution history. This provides a highly information-dense and representative endogenous state benchmark for subsequent attention query generation and final feature aggregation, ensuring that subsequent calculations are based on the most comprehensive understanding of the current state of the link.

[0033] More specifically, in a concrete example of this application, after one channel of the dual-channel LSTM, namely the network indicator encoder, processes the entire network indicator time series of length T, it outputs a network indicator hidden state sequence of the same length T. This sequence is structurally a two-dimensional tensor with a dimension of T multiplied by the hidden layer dimension. Extracting the current network indicator hidden state involves indexing this two-dimensional tensor along the time dimension, selecting and retrieving the network indicator hidden state vector corresponding to index T-1, which is the last time step. Under the recurrent neural network's operating mechanism, this vector recursively contains all the information from time step 1 to time step T-1; therefore, it is defined as the current network indicator hidden state representing the current overall state of the link.

[0034] Specifically, in step S600, degradation state prediction is performed based on the current hidden state of network indicators and the fused context vector to obtain the degradation prediction probability. It should be understood that directly concatenating and fusing the feature vector representing the internal state of the link with the context vector representing the influence of the external environment in the scenario of fiber optic link degradation prediction is a technical solution with inherent limitations. Specifically, this solution fails to explicitly model the nonlinear and asymmetric interaction between the internal state and external influences, i.e., the state vulnerability modulation effect. In physical reality, the actual degree of harm to the link from external environmental shocks highly depends on the link's own health status. For example, a high-performance new link may exhibit extremely strong robustness to slight temperature fluctuations or construction vibrations; however, a vulnerable link with aged components and in a sub-healthy state will have its performance degradation probability drastically amplified when subjected to external shocks of equal intensity. Simple splicing operations treat the internal and external factors of a link as two independent sets of features that equally influence the final result. This is a static and overly simplistic fusion method that fails to capture the key physical processes by which internal factors (link vulnerability) dynamically adjust and determine the actual destructive force of external factors (environmental shocks), thus limiting the upper limit of the accuracy and interpretability of the prediction model. To address the above technical deficiencies, a preferred embodiment of this solution introduces an adaptive context fusion mechanism based on state vulnerability gating. This mechanism no longer simply juxtaposes internal and external factors, but introduces a gating network that dynamically generates a vulnerability gating signal based on the internal state of the link to adaptively adjust the influence of external context features, thereby achieving an asymmetric feature fusion that is more consistent with physical causality. That is, in the technical solution of this application, degradation state prediction is further performed based on the current hidden state of network indicators and the fused context vector to obtain the degradation prediction probability, thereby constructing an adaptive gating mechanism driven by the inherent vulnerability of the link to dynamically adjust the effective influence of external context information, thereby achieving an asymmetric feature interaction that is more consistent with physical reality. This allows the model to go beyond simple feature aggregation, deeply capturing the coupling effect of internal and external factors, thus outputting more accurate and robust prediction results in complex and deteriorating scenarios, and enhancing the interpretability of the model.

[0035] Figure 4 This is a flowchart illustrating how to predict a degradation state based on the current hidden state of network metrics and the fused context vector, according to an embodiment of this application, to obtain a degradation prediction probability. (See attached flowchart.) Figure 4As shown, step S600 includes: S610, performing link vulnerability gating on the current hidden state of network indicators to obtain a vulnerability gating vector; S620, based on the vulnerability gating vector, adaptively adjusting the context influence of the fused context vector to obtain a state-adjusted context vector; S630, performing feature aggregation on the state-adjusted context vector and the current hidden state of network indicators to obtain an aggregated feature vector; S640, inputting the aggregated feature vector into a classifier to obtain the degradation prediction probability.

[0036] Accordingly, in step S610, link vulnerability gating is applied to the current hidden state of network indicators to obtain a vulnerability gating vector. It should be understood that a mechanism is needed to transform the abstract, high-dimensional internal health state of an optical fiber link into a quantifiable, dimension-corresponding explicit representation of its sensitivity to various external environmental shocks, thereby providing a dynamic, state-dependent control basis for subsequent modeling of internal and external factors. Therefore, in the technical solution of this application, link vulnerability gating is further applied to the current hidden state of network indicators to obtain a vulnerability gating vector, thereby constructing an intelligent mapping from the internal state of the link to its vulnerability to external shocks. In this way, instead of passively receiving external information, it actively generates a dynamic adjustment signal representing its sensitivity or defense against different external influences, providing a crucial control basis for subsequent adaptive fusion.

[0037] In a specific example of this application, this step is implemented by condensing the feature vector representing the current intrinsic health state of the fiber optic link. The current hidden state of the network metric is input into a gating unit consisting of a fully connected layer and a sigmoid activation function. This gating unit takes the current hidden state as input, performs a linear transformation on it using an internally learnable weight matrix and bias vector, and then inputs the transformed result into the sigmoid activation function. Since the sigmoid function constrains any input value to between 0 and 1, the gating unit ultimately outputs a vulnerability gating vector with the same dimension as the external context vector. Specifically, link vulnerability gating is applied to the current hidden state of network indicators using the following formula: , in, This represents the current hidden state of network metrics that indicate the internal state of the link. and These are the weight matrix and bias vector of the gated unit neural network layer, respectively, which are the parameters that the model needs to learn. The Sigmoid activation function constrains the output value to between 0 and 1. During training, the model autonomously learns physically meaningful mappings, thus constructing an intelligent mapping from the internal state of the link to its vulnerability to external shocks. For example, when the current network metric is hidden... When the dimension representing the aging of the laser in the optical module is high, the model will cause the output vulnerability-gated vector to be affected. In the model, the gating value of the dimension corresponding to the temperature effect approaches 1; similarly, when the value of the dimension representing the small reflection and unstable performance of the fiber optic connector in the hidden state is high, the model will also make the gating value of the dimension corresponding to the construction vibration effect in the gating vector approach 1.

[0038] Accordingly, in step S620, based on the vulnerability gating vector, the fused context vector is adaptively adjusted to reflect the contextual influence to obtain a state-adjusted context vector. It should be understood that since the fused context vector generated in the preceding steps only represents the universal and objective impact of the external environment and events, without combining this impact with the specific health status or defense capabilities of the fiber optic link itself, it cannot directly reflect the true and effective destructive force of external shocks on that specific link. Therefore, in the technical solution of this application, the fused context vector is further adaptively adjusted based on the vulnerability gating vector to obtain a state-adjusted context vector. This allows the link vulnerability signal generated in the previous step to actually act on the external environmental impact data, achieving meaningful interaction between the two. This accurately simulates the state vulnerability modulation effect, transforming the original, universal external environmental context information into a personalized, effective impact vector filtered and scaled by the current link's specific state, making information fusion more physically realistic and causally logical.

[0039] In a specific example of this application, this step is implemented by performing an element-wise multiplication operation, i.e., a Hadamard product, on the vulnerability gating vector generated in the previous step, representing the link's sensitivity to various external shocks, and the fused context vector, representing the combined impact of the external environment and events. Specifically, the context influence of the fused context vector is adaptively adjusted using the following formula: , in, This is the original external influence vector to be modulated, i.e., the fused context vector; This represents the vulnerability-gated vector; ⊙ represents the Hadamard product operation; This is the context vector after state adjustment. This operation precisely achieves the state vulnerability modulation effect: if the vulnerability gate vector... If a value of an element in the vector is close to 0, it indicates that the link has strong robustness in that dimension. Therefore, even after fusion, the context vector will remain robust. If the corresponding external shock signal is strong, its ultimate effective impact will be significantly weakened; conversely, if the gate value is close to 1, it indicates that the link is extremely vulnerable in this dimension, and the external shock signal will be transmitted with almost no attenuation, posing an effective threat. In this way, the model transforms the universal external impact into a personalized and effective impact on the specific link at present.

[0040] Accordingly, in steps S630 and S640, the context vector after state adjustment and the current hidden state of network indicators are aggregated to obtain an aggregated feature vector, and the aggregated feature vector is input into a classifier to obtain the degradation prediction probability. It should be understood that since the preceding steps have generated feature vectors representing the intrinsic state of the link and effective external influence vectors modulated by the link's own state, these two parts of information need to be integrated to form a comprehensive feature expression for the final decision. Therefore, in the technical solution of this application, the context vector after state adjustment and the current hidden state of network indicators are further aggregated to obtain an aggregated feature vector, and the aggregated feature vector is input into a classifier to obtain the degradation prediction probability. This integrates the internal state information after deep interaction and filtering with the effective external influence information, and makes an accurate degradation judgment accordingly. In this way, by aggregating the basic health status of the link and the external pressures that can actually threaten it under this condition, a feature expression with a clearer causal relationship and stronger predictive ability can be formed, and the final decision can be made based on this high-quality, high-information-content fused feature.

[0041] In a specific example of this application, the implementation method for feature aggregation of the state-adjusted context vector and the current network indicator hidden state is as follows: the original internal state vector representing the current intrinsic state of the link, i.e., the current network indicator hidden state, is used as the basis for feature aggregation. This is related to the adaptively modulated context vector representing the effective external influence, i.e., the state-adjusted context vector. This involves concatenating features along their respective dimensions. This operation results in a highly condensed final fused feature vector, also known as the aggregated feature vector. It simultaneously includes both the internal factors of the link and the effective external factors after adjustment by the internal factors. Subsequently, the feature vectors will be aggregated. The implementation method for obtaining the degradation prediction probability by inputting a classifier is as follows: The final fused feature vector generated in the previous step is fed into a prediction classifier composed of fully connected layers. This classifier performs a linear transformation on the final fused feature vector using its internally learned weight matrix and bias vector, and then feeds the result into a Sigmoid activation function. This function constrains the final output value to the interval between 0 and 1, thereby obtaining the final degradation prediction probability. This probability value can be further mapped to high, medium, and low risk levels, providing decision support for network operation and maintenance. This can be expressed by the following formula: , , in, This represents a vector concatenation operation. To degrade the predicted probability, the final output probability value is between 0 and 1. and These are the weight matrix and bias vector of the prediction classifier layer, respectively. The concatenation operation here occurs after the information has undergone effective interaction and filtering. That is, it aggregates the basic health status of the link and the external pressures that can actually threaten it under this status, thus forming a feature expression with a clearer causal relationship and stronger predictive ability.

[0042] By introducing state vulnerability gating through the above preferred embodiment, the original context information is adaptively modulated, ultimately achieving deep modeling of the nonlinear interaction between the internal state of the link and the influence of the external environment. This overcomes the symmetry and static nature of information processing inherent in simple splicing and fusion, making the feature fusion process more consistent with the causal logic of the physical world. This mechanism allows the model to dynamically and adaptively assess the actual threat level of different external shocks under different link health states, thereby significantly improving the accuracy and robustness of predictions, especially when dealing with degradation scenarios where link state and environmental factors are complexly intertwined. Furthermore, by analyzing the numerical values ​​of the gating vector, the interpretability of the model can be enhanced, providing network operators with deep insights into why the current link is particularly sensitive to certain environmental changes, achieving an intelligent upgrade from predicting what to explain why.

[0043] In summary, the fiber optic link degradation prediction method for cable television networks according to the embodiments of this application is explained. First, it extracts the dynamic evolution characteristics of network performance indicators and external context information through a dual-channel time-series coding structure. Then, using a cross-modal information fusion mechanism, the model can intelligently focus on and extract the most relevant impact information from complex historical contexts based on the current internal state of the link. The scheme further introduces an adaptive adjustment mechanism for vulnerability gating, which can dynamically adjust the actual influence of the external context based on the current internal health of the link, thereby accurately simulating the physical reality that vulnerable links are more susceptible to external shocks. Through this complete perception-fusion-modulation-prediction link, this scheme achieves deep modeling of the nonlinear coupling effect of internal and external factors, effectively overcoming the shortcomings pointed out in the background art, and thus significantly improving the accuracy and robustness of degradation prediction.

[0044] Furthermore, a fiber optic link degradation prediction system for cable television networks is also provided.

[0045] Figure 5 This is a block diagram of an optical fiber link degradation state prediction system for a cable television network according to an embodiment of this application. Figure 5 As shown, the fiber optic link degradation state prediction system 100 for cable television networks according to an embodiment of this application includes: a time-series data acquisition module 110, used to acquire network indicator time series, environmental context time series, and event context data stream; a fast variable sequence merging module 120, used to merge the environmental context time series and event context data stream using fast variable sequences to obtain a context parameter time series; a time-series encoding module 130, used to perform feature sequence encoding based on dual-channel LSTM on the network indicator time series and context parameter time series to obtain a network indicator hidden state sequence and a context information hidden state sequence; a cross-modal state fusion module 140, used to perform cross-modal information fusion on the network indicator hidden state sequence and the context information hidden state sequence to obtain a fused context vector; a current network state extraction module 150, used to extract the current network indicator hidden state from the network indicator hidden state sequence; and a degradation state prediction module 160, used to predict the degradation state based on the current network indicator hidden state and the fused context vector to obtain a degradation prediction probability.

[0046] As described above, the fiber optic link degradation prediction system 100 for cable television networks according to embodiments of this application can be implemented in various wireless terminals, such as servers with fiber optic link degradation prediction algorithms for cable television networks. In one possible implementation, the fiber optic link degradation prediction system 100 for cable television networks according to embodiments of this application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the fiber optic link degradation prediction system 100 for cable television networks can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the fiber optic link degradation prediction system 100 for cable television networks can also be one of many hardware modules of the wireless terminal.

[0047] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for predicting the degradation status of optical fiber links in cable television networks, characterized in that, include: Obtain network metric time series, environment context time series, and event context data streams; The environment context time series and event context data stream are merged using fast variable sequences to obtain the context parameter time series. The network indicator time series and the context parameter time series are encoded using a dual-channel LSTM feature sequence to obtain the network indicator hidden state sequence and the context information hidden state sequence. Cross-modal information fusion is performed on the hidden state sequence of network indicators and the hidden state sequence of context information to obtain the fused context vector; Extract the current hidden state of the network indicators from the hidden state sequence of network indicators; Degradation state prediction is performed based on the current network indicator hidden state and the fused context vector to obtain the degradation prediction probability.

2. The method for predicting the degradation status of optical fiber links in cable television networks according to claim 1, characterized in that, The environment context time series and event context data stream are merged using fast variable sequences to obtain the context parameter time series, including: Based on a reference time grid, the environmental context time series is aligned to obtain an aligned environmental context time series. Based on the reference time grid, the discrete event data of the event context data stream is structured and vectorized to obtain a vectorized event time series; The aligned environmental context time series and the vectorized event time series are merged to obtain the context parameter time series.

3. The method for predicting the degradation status of optical fiber links in cable television networks according to claim 2, characterized in that, The process of merging the aligned environment context time series and the vectorized event time series to obtain the context parameter time series includes: performing feature concatenation on each group of corresponding environment vectors and event vectors in the aligned environment context time series and the vectorized event time series to obtain the context parameter time series.

4. The method for predicting the degradation status of optical fiber links in cable television networks according to claim 1, characterized in that, Cross-modal information fusion is performed on the hidden state sequences of network metrics and contextual information to obtain the fused context vector, including: Extract the current hidden state of the network indicators from the hidden state sequence of network indicators; A linear transformation is performed on the current hidden state of network metrics to obtain the network metric query vector. A linear transformation is performed on the hidden state sequence of context information to obtain the context information key matrix and the context information value matrix; Input the network metric query vector, context information key matrix, and context information value matrix into the converter module to obtain the fused context vector.

5. The method for predicting the degradation status of optical fiber links in cable television networks according to claim 4, characterized in that, The network metric query vector, context information key matrix, and context information value matrix are input into the converter module to obtain the fused context vector. This includes: the converter module processes the network metric query vector, context information key matrix, and context information value matrix using the following formula to obtain the fused context vector: in, This is a query vector for network metrics. For context information key matrix, This is a matrix of context information values. Let be the dimension of each row vector in the context information key matrix. for function, This is the fused context vector.

6. The method for predicting the degradation status of optical fiber links in cable television networks according to claim 1, characterized in that, Degradation state prediction is performed based on the current network metric hidden state and the fused context vector to obtain the degradation prediction probability, including: Link vulnerability gating is applied to the current hidden state of network indicators to obtain a vulnerability gating vector; Based on the vulnerability gating vector, the context influence of the fused context vector is adaptively adjusted to obtain the state-adjusted context vector; The context vector after state adjustment and the current network metric hidden state are used to perform feature aggregation to obtain an aggregated feature vector; The aggregated feature vector is input into the classifier to obtain the degradation prediction probability.

7. The method for predicting the degradation status of optical fiber links in cable television networks according to claim 6, characterized in that, To obtain a vulnerability gating vector, the current hidden state of network indicators is subjected to link vulnerability gating, including: applying link vulnerability gating to the current hidden state of network indicators using the following formula: in, The current network metrics are hidden. and These are the weight matrix and bias vector of the gated unit neural network layer, respectively. It is the Sigmoid activation function. This is the vulnerability gating vector.

8. The method for predicting the degradation status of optical fiber links in cable television networks according to claim 6, characterized in that, Based on the vulnerability gating vector, the context influence of the fused context vector is adaptively adjusted to obtain the state-adjusted context vector, including: adaptively adjusting the context influence of the fused context vector using the following formula: in, This is the fused context vector. This is the vulnerability-gated vector, where ⊙ represents the Hadamard product operation. This is the context vector after state adjustment.

9. A fiber optic link degradation prediction system for cable television networks, characterized in that, include: The time-series data acquisition module is used to acquire network indicator time series, environmental context time series, and event context data streams; The fast variable sequence merging module is used to merge environment context time series and event context data streams to obtain context parameter time series. The time-series coding module is used to perform feature sequence coding based on dual-channel LSTM on the network indicator time series and the context parameter time series to obtain the network indicator hidden state sequence and the context information hidden state sequence. The cross-modal state fusion module is used to perform cross-modal information fusion on the network indicator hidden state sequence and the context information hidden state sequence to obtain the fused context vector; The current network state extraction module is used to extract the current hidden state of network indicators from the hidden state sequence of network indicators; The degradation state prediction module is used to predict the degradation state based on the current hidden state of network indicators and the fused context vector to obtain the degradation prediction probability.

Images