A flexible signal transition structure for optical module and high speed connector interconnection

Abstract

This invention discloses a flexible signal switching structure for interconnecting optical modules and high-speed connectors, relating to the field of intelligent connectivity. It includes an optical module connection end, a connector end on the right side of the optical module connection end, a flexible circuit board body mounted on the left end of the connector end, and a gold finger connection end fixedly connected to the right side of the top of the optical module connection end. The gold finger connection end has a pressing plate inside for pressing the flexible circuit board body, and the left end of the pressing plate is rotatably connected to the inner wall of the gold finger connection end. This invention transforms the flexible circuit board body from a signal transmission channel into an intelligent structure that can be monitored in real time. It can not only sense its current physical deformation, vibration, and temperature state, but also predict its future failure risk based on continuous abnormal data trends using a predictive model. This allows maintenance personnel to anticipate the lifespan of the connection components and avoid sudden link interruptions or precipitous performance degradation.

Description

Technical Field

[0001] This invention relates to the field of smart connectivity technology, specifically to a flexible signal switching structure for interconnecting optical modules and high-speed connectors. Background Technology

[0002] Optical interconnect technology, with its advantages of ultra-high bandwidth, low loss, and resistance to electromagnetic interference, has become the backbone transmission solution for modern high-speed information systems. Inside devices, especially switches, routers, and servers, optical modules need to be interconnected with ASICs or SerDes chips on the motherboard via high-speed electrical connectors. However, due to the compact internal space, complex board layout, and operational requirements such as hot-swapping and maintenance, a rigid, direct connection between optical modules and motherboard connectors is often not possible. Therefore, flexible printed circuits (FPCs) with excellent bending performance have become a crucial signal transfer medium. They enable flexible wiring within limited space, connecting optical modules and connectors in different planes or locations, and are one of the core components for building high-density, scalable, high-speed interconnect systems. Their operating environment typically involves high-frequency signals, significant heat generation, and device-level vibrations that may be caused by fans, hard drives, etc.

[0003] In existing technologies, the solution using flexible FPC for connection mainly faces the following core drawbacks:

[0004] Existing connection structures struggle to detect potential degradation of FPCs during long-term service due to repeated insertion / removal, vibration, and thermal cycling. This degradation can lead to fatigue, microcracks, increased contact resistance at the bonding joints, or embrittlement of the solder joints. When connection performance begins to deteriorate due to these physical factors, the impact directly manifests as a subtle decline in high-speed signal quality, such as eye diagram closure, increased jitter, and a higher bit error rate. Traditional systems only detect problems when the bit error rate reaches a critical threshold, triggering an alarm. By this time, the connection may have already failed, making early warning impossible.

[0005] In the face of signal degradation, the existing system’s response is lagging and simplistic, such as directly reducing the link rate or switching redundant paths. This will sacrifice performance and bandwidth, and cannot make precise preventive adjustments for specific failure modes. Summary of the Invention

[0006] (a) Technical problems to be solved

[0007] In view of the above-mentioned shortcomings of the prior art, the present invention provides a flexible signal switching structure for interconnecting optical modules and high-speed connectors, which can effectively solve the problems of the prior art.

[0008] (II) Technical Solution

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] This invention discloses a flexible signal switching structure for interconnecting an optical module and a high-speed connector, including an optical module connection end, a connector end disposed on the right side of the optical module connection end, a flexible circuit board body mounted on the left end of the connector end, a gold finger connection end fixedly connected to the right side of the top of the optical module connection end, a pressing plate for pressing the flexible circuit board body is disposed inside the gold finger connection end, the left end of the pressing plate is rotatably connected to the inner wall of the gold finger connection end, the left side of the flexible circuit board body is electrically connected to the inner wall of the bottom end of the gold finger connection end, and the bottom end of the gold finger connection end is mounted with... Equipped with a temperature sensor, the left side of the gold finger connector is fixedly connected to a hollow shell, and a fan is installed inside the hollow shell. A displacement sensing component is provided on the front side of the flexible circuit board body. The displacement sensing component includes a laser sensor transmitter one, a laser sensor transmitter two, a laser receiver one, and a laser receiver two. The laser sensor transmitter two is installed on the right side of the top of the optical module connector, the laser sensor transmitter one is installed on the top of the laser sensor transmitter two, the laser receiver one is installed on the left side of the top of the connector, and the laser receiver two is installed on the back side of the laser receiver one.

[0011] A control unit is installed at the top of the optical module connection end. The control unit includes a status recognition module and a predictive control module, which are electrically connected. The status recognition module is used to fuse the sensing data of laser sensor emitter one, laser sensor emitter two, vibration sensor, and temperature sensor with the physical properties of the flexible circuit board body to simulate and monitor the deformation, connection status, and environmental conditions of the flexible circuit board body in real time. The predictive control module is used to analyze the abnormal fluctuation trend based on the monitoring data obtained by the status recognition module, match potential faults, predict risks through a prediction model preset by a deep learning algorithm, and drive the data interaction settings of the optical module connection end and the connector end, as well as the start-up settings of the fan, for adaptive control.

[0012] Furthermore, the second laser receiver is used to receive the laser signal emitted by the second laser sensor transmitter, and the first laser receiver is used to receive the laser signal emitted by the first laser sensor transmitter. The first and second laser receivers identify the received laser signals to output the relative displacement data between the optical module connection end and the connector end.

[0013] Furthermore, the status recognition module has sub-modules deployed at its lower level, including a data extraction module, a working condition simulation module, and a connection status monitoring module. The working condition simulation module is electrically connected to the data extraction module and the connection status monitoring module, wherein:

[0014] The data extraction module is used to acquire in real time the three-dimensional displacement data, vibration spectrum data, and temperature data of the contact point between the optical module connection end and the connector end, as well as the contact point between the flexible circuit board body and the gold finger connection end, and to perform standardization processing.

[0015] The working condition simulation module is used to extract standardized sensor data and combine it with pre-stored material mechanical property data and geometric structure data of the flexible circuit board body. Through simulation model, it generates the dynamic flexible bending three-dimensional shape of the flexible circuit board body under the current working condition and the stress distribution in real time. The pre-stored flexible circuit board body property data includes: the deformation coefficient and thermal expansion coefficient of the substrate, the resistance temperature coefficient of the internal conductive circuit, and the thickness and bonding strength parameters of each laminated structure, which are used to accurately calculate the stress concentration points and impedance change trends under bending conditions.

[0016] The connection status monitoring module is used to comprehensively determine the real-time tightness of the pressing plate and the connection integrity between the flexible circuit board body and the connector welding end based on the output of the working condition simulation module.

[0017] Furthermore, the process of constructing and running the simulation model in the operating condition simulation module includes:

[0018] Based on the three-dimensional model of the flexible circuit board body, it is discretized into a finite element mesh, and based on its pre-stored material mechanical property data, each element in the mesh is assigned a corresponding deformation load.

[0019] The three-dimensional displacement data obtained in real time by the data extraction module is used as dynamic displacement boundary conditions and loaded onto the nodes in the finite element mesh corresponding to the installation positions of the optical module connection end and the connector end; at the same time, real-time temperature data is used as thermal load boundary conditions and applied to the entire network to obtain thermal strain.

[0020] Within each sampling period, based on the acquired deformation load and thermal strain data, the overall deformation field of the flexible circuit board body under the current displacement and temperature conditions, the stress distribution of each layer structure, and the maximum equivalent stress point and the maximum bending curvature region are defined.

[0021] Based on the obtained stress distribution, combined with the resistance strain coefficient and thermal expansion coefficient of the flexible circuit board body, the real-time changes of parasitic resistance, inductance and capacitance parameters caused by deformation and temperature changes in the key transmission paths on the flexible circuit board body are obtained through a predefined mapping relationship, so as to map the physical deformation state into the trend of electrical parameter changes that affect signal integrity.

[0022] Furthermore, the operation of the connection status monitoring module includes:

[0023] During the initial calibration phase, the feedback of pressing and welding changes corresponding to the inherent vibration spectrum collected by the vibration sensor was recorded under the known good gold finger connection end and pressing plate pressing state, and the connector end and flexible circuit board body welding state, and it was used as the normal state baseline spectrum.

[0024] The system receives broadband vibration data output from vibration sensors in real time, converts it into a real-time vibration spectrum through fast Fourier transform, and extracts amplitude characteristics, spectral energy distribution characteristics, and resonant peak frequency shift characteristics of specific frequency bands.

[0025] The system receives the theoretical bending shape and stress distribution of the flexible circuit board body simulated based on the current displacement and temperature conditions from the working condition simulation module, and defines the expected normal vibration response range of the system under the current working conditions.

[0026] The real-time vibration characteristics were compared with the baseline spectrum under normal conditions and the vibration response range, respectively:

[0027] When the real-time vibration spectrum shows a sustained increase in amplitude above the threshold or a new resonance peak in the high-frequency band, and this feature cannot be matched by the normal vibration response range currently output by the working condition simulation module, it is determined that there is microscopic contact loosening or foreign matter between the gold finger connection end and the pressing plate, and the tight closure state deteriorates.

[0028] When the real-time vibration spectrum shows a change in vibration energy that is strongly correlated with the bending shape of the flexible circuit board body in a specific low-frequency band, but the change pattern deviates from the baseline spectrum of the normal state, it is determined that there are poor solder joints or fatigue damage at the welding end of the flexible circuit board body and the connector end, and the connection integrity is reduced.

[0029] Furthermore, the predictive control module is further equipped with sub-modules, including an abnormal fluctuation analysis module, a preset event matching module, an intelligent prediction module, and an adaptive control module. The intelligent prediction module is electrically connected to the abnormal fluctuation analysis module, the preset event matching module, and the adaptive control module, wherein:

[0030] The abnormal fluctuation analysis module is used to start the analysis process when the state recognition module detects that the flexible circuit board body has undergone bending deformation exceeding the preset threshold or detects abnormal vibration. It determines whether there are abnormal fluctuations in amplitude or timing of the electrical signal transmitted through the flexible circuit board body under the current temperature conditions, and extracts the trend of continuous abnormal fluctuation data.

[0031] The preset event matching module is used to match the abnormal fluctuation trend characteristics extracted by the abnormal fluctuation analysis module with a pre-established abnormal event library. The abnormal event library includes multiple preset abnormal events associated with the flexible circuit board body, pressing connection, and welding connection, as well as their corresponding fluctuation characteristic patterns, and outputs one or more candidate preset abnormal events that may be related. The abnormal event library is pre-constructed through historical fault data mining and accelerated life testing. In addition to containing fluctuation characteristic patterns, each preset abnormal event is also associated with a root cause physical model, typical development stage, and corresponding maintenance strategy code, providing rich contextual information for subsequent prediction.

[0032] The intelligent prediction module is used to build and train a prediction model. The model takes the abnormal fluctuation trend characteristics and the candidate preset abnormal event sequence output by the preset event matching module as input features. After training, it predicts the risk value of the current flexible circuit board body under various preset abnormal event performance indicators within a specified time period in the future.

[0033] The adaptive control module receives the predicted risk value output by the intelligent prediction module. When the risk value exceeds the preset level, it automatically generates control commands that include preventative adjustments to the signal transmission parameters of the optical module connection end and connector end through the flexible circuit board main channel, as well as adaptive changes to the operating power of the associated wind turbine.

[0034] Furthermore, the process by which the abnormal fluctuation analysis module judges abnormal fluctuations includes:

[0035] With the flexible circuit board body in a standard static state and at a standard operating temperature, the high-speed signals transmitted by it are collected and analyzed, and a standard signal quality template is generated and stored. The template includes a standard signal eye diagram profile, an allowable jitter tolerance range, and normal amplitude and rise or fall time parameters.

[0036] Under the current operating conditions, the signal transmitted through the flexible circuit board body is captured in real time, and the signal quality parameters such as the real-time eye diagram shape, total jitter, random jitter, deterministic jitter, signal amplitude and edge rate of the current signal are extracted simultaneously.

[0037] The real-time signal quality parameters extracted synchronously are dynamically compared with the standard template, and the actual deviation values ​​of each parameter are calculated.

[0038] Combining the real-time temperature data obtained by the data extraction module, temperature impact compensation calculations are performed on the deviation values ​​of various signal quality parameters; the compensated deviation values ​​are compared with preset dynamic abnormal thresholds for different temperature levels; when the deviation of one or more key parameters continues to exceed its corresponding threshold, it is determined that there is abnormal fluctuation in the signal.

[0039] Furthermore, the trend extraction process of abnormal fluctuation data in the abnormal fluctuation analysis module includes:

[0040] Establish a data buffer to synchronously receive and store the time sequence of quality parameters of each signal that has been determined to be abnormal;

[0041] On the circular buffer, a sliding time window of configurable length is defined; within each window, statistical features, time-domain trend features, frequency-domain transformation features, and event count features are extracted for the sequence of each abnormal parameter.

[0042] Within the same time window, calculate the Pearson correlation coefficient between different anomaly parameters, quantify the correlation strength, and define the correlation between signal anomaly parameters and physical state parameters;

[0043] All extracted features are combined in a preset fixed order to form a trend feature vector that represents the abnormal fluctuation pattern within the current window. The system continuously records the trend feature vectors output by each sliding window and arranges them in chronological order to form a trend evolution sequence.

[0044] Furthermore, the construction process of the prediction model in the intelligent prediction module includes the following steps:

[0045] Step 1: Collect and integrate the dataset for model training;

[0046] Step 2: Preprocess the raw time-series sensing data, including denoising, normalization and alignment; then, extract shallow time-frequency features, deep abstract features and physical guidance features through the following parallel path, and splice and fuse the three types of features in the feature dimension to form a comprehensive feature vector.

[0047] Step 3: Construct the model framework, whose core architecture includes, in sequence, a feature embedding layer, a temporal dependency modeling unit, a spatial attention unit, and an event association and risk prediction layer;

[0048] Step 4: Train the model using an optimization objective that includes a multi-task loss function. During training, use historical datasets and minimize the total loss through backpropagation and an adaptive optimizer until the model converges.

[0049] Step 5: Evaluate the performance of the trained model using independent validation and test sets. The core evaluation metrics include the precision and recall of event association prediction, and the mean absolute error of risk value prediction. After validation, solidify the model parameters and deploy them online.

[0050] Furthermore, the adaptive control module's preventative adjustments to the optical module connection end and connector end include: adjusting the pre-emphasis coefficient of the transmitting end, the equalizer parameters of the receiving end, and the data rate or encoding scheme of the signal link according to a preset scale based on the risk value; the adaptive control module's control instructions for the wind turbine include: when the predicted risk value is strongly correlated with temperature, increasing the wind turbine speed according to the predicted risk value according to a corresponding scale.

[0051] (III) Beneficial Effects

[0052] Compared with the known prior art, the technical solution provided by this invention has the following beneficial effects:

[0053] 1. By integrating vibration sensors, displacement sensing components, temperature sensors, and control units, the flexible circuit board body is transformed from a simple signal transmission channel into an intelligent structure that can be monitored in real time. It can not only sense its current physical deformation, vibration, and temperature status, but also predict its future failure risk based on continuous abnormal data trends through predictive models. This allows maintenance personnel to anticipate the lifespan of the connecting components in advance, significantly improving the reliability and availability of the entire optical interconnect system from the perspective of preventive maintenance.

[0054] 2. By deeply linking the physical layer status monitoring with the electrical layer signal quality analysis, the physical root cause of signal degradation can be located. Based on the predicted risk value, before the signal quality deteriorates significantly, parameters such as pre-emphasis at the signal transmitting end and equalization at the receiving end are proactively adjusted at the optical module connection end or connector end to compensate for the impedance changes or increased losses that will occur at the connection. Thus, during the slow degradation of the physical connection, signal integrity is maintained to the greatest extent, avoiding sudden link interruptions or precipitous performance drops, and ensuring the continuous and smooth operation of high-bandwidth services.

[0055] 3. By enhancing the heat dissipation of the channel in advance when the model predicts a strong correlation between risk and local overheating, the heat dissipation can be enhanced to avoid the accumulation of temperature leading to faster performance degradation. Compared with the traditional global full-load operation heat dissipation, it can more effectively reduce the overall power consumption, thereby reducing the overall maintenance cost of the system from the perspective of the whole life cycle. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0057] Figure 1 This is a schematic diagram of the overall three-dimensional structure of the present invention;

[0058] Figure 2 This is a three-dimensional structural schematic diagram of the present invention from another angle;

[0059] Figure 3 This is a front cross-sectional view of the hollow shell, gold finger connecting end, and pressing plate in this invention.

[0060] Figure 4 This is a schematic diagram of the architecture of the control unit in this invention.

[0061] The labels in the diagram represent: 1. Optical module connection end; 2. Connector end; 3. Flexible circuit board body; 4. Gold finger connection end; 5. Press plate; 6. Hollow shell;

[0062] 7. Displacement sensing assembly; The displacement sensing assembly includes: 71. Laser sensor transmitter one; 72. Laser sensor transmitter two; 73. Laser receiver one; 74. Laser receiver two;

[0063] 8. Control Unit; The control unit includes a status recognition module and a predictive control module. The status recognition module includes: 81. Data extraction module; 82. Operating condition simulation module; 83. Connection status monitoring module; The predictive control module includes: 84. Abnormal fluctuation analysis module; 85. Preset event matching module; 86. Intelligent prediction module; 87. Adaptive control module; 9. Vibration sensor; 10. Fan; 11. Temperature sensor. Detailed Implementation

[0064] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0065] The present invention will be further described below with reference to embodiments.

[0066] Example 1

[0067] This embodiment provides a flexible signal switching structure for interconnecting optical modules and high-speed connectors, such as... Figure 1 - Figure 4As shown, the device includes an optical module connection end 1, a connector end 2 on the right side of the optical module connection end 1, a flexible circuit board body 3 mounted on the left side of the connector end 2, a gold finger connection end 4 fixedly connected to the top right side of the optical module connection end 1, a pressing plate 5 for pressing the flexible circuit board body 3 inside the gold finger connection end 4, the left end of the pressing plate 5 being rotatably connected to the inner wall of the gold finger connection end 4, the left side of the flexible circuit board body 3 being electrically connected to the inner wall of the bottom end of the gold finger connection end 4, a temperature sensor 11 mounted at the bottom end of the gold finger connection end 4, a hollow shell 6 fixedly connected to the left side of the gold finger connection end 4, a fan 10 installed inside the hollow shell 6, and a displacement sensing component 7 on the front side of the flexible circuit board body 3. The displacement sensing component 7 includes a laser sensor transmitter 71, a laser sensor transmitter 72, and a laser receiver. Laser receiver 73 and laser receiver 74 are mounted on the top right side of the optical module connection 1, laser sensor transmitter 72 is mounted on the top of laser sensor transmitter 72, laser receiver 73 is mounted on the top left side of the connector 2, and laser receiver 74 is mounted on the back side of laser receiver 73. Vibration sensor 9 is a wideband piezoelectric ceramic sensor, and temperature sensor 11 is a patch-type digital temperature sensor. Laser receiver 74 is used to receive the laser signal emitted by laser sensor transmitter 72, and laser receiver 73 is used to receive the laser signal emitted by laser sensor transmitter 71. Laser receiver 73 and laser receiver 74 identify the received laser signals to output the relative displacement data between optical module connection 1 and connector 2. The displacement data calculation process is as follows:

[0068] The data output from laser receiver 1 73 is used to calculate the lateral displacement and deflection angle of the optical module connection end 2 relative to the connector end 2 in the X and Y axis planes; the data output from laser receiver 2 74 is used to calculate the displacement and angle change data of the connector end 2 relative to the optical module connection end 1 in the complementary direction; the two sets of relative displacement data are fused and coordinate transformed to synthesize and output the relative displacement vector and attitude angle change data between the optical module connection end 1 and the high-speed connector end 2 in three-dimensional space.

[0069] A control unit 8 is installed at the top of the optical module connection terminal 1. The control unit 8 includes a state recognition module and a predictive control module, which are electrically connected. The state recognition module is used to fuse the sensing data of laser sensor emitter 1 71, laser sensor emitter 2 72, vibration sensor 9, and temperature sensor 11 with the physical properties of the flexible circuit board body 3, and to simulate and monitor the deformation, connection status, and environmental conditions of the flexible circuit board body 3 in real time. Sub-modules are deployed below the state recognition module, including a data extraction module 81, a working condition simulation module 82, and a connection status monitoring module 83. The working condition simulation module 82 is electrically connected to the data extraction module 81 and the connection status monitoring module 83, wherein:

[0070] The data extraction module 81 is used to acquire in real time the three-dimensional displacement data, vibration spectrum data, and temperature data of the contact point between the optical module connection end 1 and the connector end 2, as well as the contact point between the flexible circuit board body 3 and the gold finger connection end 4, and to perform standardization processing.

[0071] The working condition simulation module 82 is used to extract standardized sensor data and combine it with pre-stored material mechanical property data and geometric structure data of the flexible circuit board body 3 to simulate and generate the dynamic flexible bending three-dimensional shape of the flexible circuit board body 3 under the current working condition and stress distribution in real time through the simulation model. The pre-stored property data of the flexible circuit board body 3 includes: the deformation coefficient and thermal expansion coefficient of the substrate, the resistance temperature coefficient of the internal conductive circuit, and the thickness and bonding strength parameters of each laminated structure, which are used to accurately calculate the stress concentration points and impedance change trends under bending conditions.

[0072] The connection status monitoring module 83 is used to comprehensively determine the real-time tightness of the pressing plate 5 and the connection integrity of the flexible circuit board body 3 and the welding end of the connector end 2 based on the output of the working condition simulation module 82. The operation process of the connection status monitoring module 83 includes:

[0073] During the initial calibration phase, the feedback of pressing and welding changes corresponding to the inherent vibration spectrum collected by the vibration sensor 9 is recorded under the known good gold finger connection end 4 and pressing plate 5 pressing state, and connector end 2 and flexible circuit board body 3 welding state, and is used as the normal state baseline spectrum.

[0074] The system receives wideband vibration data output from vibration sensor 9 in real time, converts it into a real-time vibration spectrum through fast Fourier transform, and extracts amplitude characteristics, spectral energy distribution characteristics, and resonant peak frequency shift characteristics of specific frequency bands.

[0075] The system receives the theoretical bending shape and stress distribution of the flexible circuit board body 3, which are simulated based on the current displacement and temperature conditions, from the working condition simulation module 82, and defines the expected normal vibration response range of the system under the current working conditions.

[0076] The real-time vibration characteristics were compared with the baseline spectrum under normal conditions and the vibration response range, respectively:

[0077] When the real-time vibration spectrum shows a sustained increase in amplitude above the threshold or a new resonance peak in the high-frequency band, and this feature cannot be matched by the normal vibration response range currently output by the working condition simulation module 82, it is determined that there is microscopic contact loosening or foreign matter between the gold finger connection end 4 and the pressing plate 5, and the tight closure state deteriorates.

[0078] When the real-time vibration spectrum shows a vibration energy change that is strongly correlated with the bending shape change of the flexible circuit board body 3 in a specific low-frequency band, but its change pattern deviates from the baseline spectrum of the normal state, it is determined that there are poor solder cracks or fatigue damage at the welding end of the flexible circuit board body 3 and the connector end 2, and the connection integrity status is reduced.

[0079] In this embodiment, the vibration spectrum is used as an independent abnormal signal and compared with the expected response of theoretical simulation. This allows for the penetrating identification of early degradation characteristics caused by microscopic contact loosening, foreign object intrusion, or hidden damage to the weld point, greatly improving the accuracy of condition judgment.

[0080] The predictive control module is used to analyze abnormal fluctuation trends based on monitoring data obtained from the state recognition module, match potential faults, predict risks through a pre-set prediction model using a deep learning algorithm, and drive adaptive control of data interaction settings between the optical module connection terminal 1 and the connector terminal 2, as well as the start-up settings of the fan 10. The predictive control module has sub-modules deployed below it, including an abnormal fluctuation analysis module 84, a pre-set event matching module 85, an intelligent prediction module 86, and an adaptive control module 87. The intelligent prediction module 86 is electrically connected to the abnormal fluctuation analysis module 84, the pre-set event matching module 85, and the adaptive control module 87.

[0081] The abnormal fluctuation analysis module 84 is used to initiate an analysis process when the state recognition module detects bending deformation of the flexible circuit board body 3 exceeding a preset threshold or detects abnormal vibration. This process determines whether there are abnormal fluctuations in amplitude or timing of the electrical signal transmitted through the flexible circuit board body 3 under the current temperature conditions, and extracts trends from continuous abnormal fluctuation data. The abnormal fluctuation analysis module 84's abnormal fluctuation judgment process includes:

[0082] With the flexible circuit board body 3 in a standard static state and at a standard operating temperature, the high-speed signal transmitted by it is collected and analyzed, and a standard signal quality template is generated and stored. The template includes a standard signal eye diagram profile, an allowable jitter tolerance range, and normal amplitude and rise or fall time parameters.

[0083] Under the current operating conditions, the signal transmitted through the flexible circuit board body 3 is captured in real time, and the signal quality parameters such as the real-time eye diagram shape, total jitter, random jitter, deterministic jitter, signal amplitude and edge rate of the current signal are extracted simultaneously.

[0084] The real-time signal quality parameters extracted synchronously are dynamically compared with the standard template, and the actual deviation values ​​of each parameter are calculated.

[0085] Combining the real-time temperature data obtained by the data extraction module 81, temperature impact compensation calculations are performed on the deviation values ​​of various signal quality parameters; the compensated deviation values ​​are compared with preset dynamic abnormal thresholds for different temperature levels; when the deviation of one or more key parameters continues to exceed its corresponding threshold, it is determined that there is abnormal fluctuation in the signal.

[0086] The preset event matching module 85 is used to match the abnormal fluctuation trend features extracted by the abnormal fluctuation analysis module 84 with a pre-established abnormal event library. The abnormal event library includes multiple preset abnormal events and their corresponding fluctuation feature patterns associated with the flexible circuit board body 3, the pressing connection, and the welding connection. It outputs one or more candidate preset abnormal events that may be related. The abnormal event library is pre-built through historical fault data mining and accelerated life test. In addition to the fluctuation feature pattern, each preset abnormal event is also associated with the root cause physical model, typical development stage, and corresponding maintenance strategy code, providing rich contextual information for subsequent prediction.

[0087] The intelligent prediction module 86 is used to construct and train a prediction model. The model takes the abnormal fluctuation trend characteristics and the candidate preset abnormal event sequence output by the preset event matching module 85 as input features. After training, it predicts the risk value of the current flexible circuit board body 3 under various preset abnormal event performance indicators within a specified future time period. The calculation process of the risk value is as follows: For each matched candidate preset abnormal event, the model outputs its probability of occurrence P (0≤P≤1) and the expected severity score within the future prediction period. The severity score is normalized to 0-10 points. According to real-time operating data, a dynamic weight is assigned to each event. This weight is obtained by weighted summation of the event's impact factor on the current signal transmission quality, the maintenance urgency factor, and the historical occurrence frequency factor. The risk value is obtained by weighted calculation of the product of the aforementioned probability of occurrence and severity score.

[0088] The adaptive control module 87 is used to receive the predicted risk value output by the intelligent prediction module 86. When the risk value exceeds the preset level, it automatically generates control instructions including preventive adjustment of the signal transmission parameters of the optical module connection end 1 and connector end 2 through the flexible circuit board body 3 channel, and adaptive change of the operating power of the associated wind turbine 10.

[0089] The adaptive control module 87 performs preventative adjustments to the optical module connection end 1 and connector end 2, including: adjusting the pre-emphasis coefficient of the transmitting end, the equalizer parameters of the receiving end, and the data rate or coding scheme of the signal link according to a preset scale based on the risk value, in order to pre-compensate for signal integrity damage that may be introduced due to connection degradation; the adaptive control module 87 provides control instructions to the fan 10, including: when the predicted risk value is strongly correlated with temperature, increasing the fan speed of 10 according to the predicted risk value according to a corresponding scale; the process of analyzing whether the predicted risk value is strongly correlated with temperature is: calculating the risk prediction value sequence of various abnormal events in historical data and the corresponding local temperature. By statistically correlating key thermal characteristic sequences such as temperature change gradient, high temperature duration, and temperature anomaly trigger threshold, a risk-thermal correlation database is established. During online prediction, when the prediction model outputs a high-risk warning, the set of candidate abnormal events associated with the current risk value is compared in real time. If the events in the set are marked in the database as being mainly caused by thermal stress, thermal fatigue, or performance degradation due to high temperature, then the predicted risk value is determined to be strongly correlated with temperature. Simultaneously, the temperature data collected by the current temperature sensor 11 is analyzed to see if there is a trend of abnormality or if it is about to exceed the preset threshold, as an auxiliary verification to ensure the accuracy and timeliness of control decisions.

[0090] Example 2

[0091] This embodiment provides a trend extraction process for abnormal fluctuation data, including:

[0092] A first-in-first-out (FIFO) circular data buffer is established to synchronously receive and store the time sequence of each signal quality parameter that has been determined to be abnormal. The signal quality parameters include at least eye diagram closure, total jitter, deterministic jitter, and signal amplitude attenuation. At the same time, the buffer also stores sensor data at the same timestamp, including bending deformation amplitude, vibration main frequency amplitude, and connection point temperature.

[0093] A configurable sliding time window is defined on the circular buffer. Within each window, statistical features, time-domain trend features, frequency-domain transformation features, and event count features are extracted for the sequence of each anomalous parameter. Statistical features include calculating the mean, variance, skewness, and kurtosis of the data within the window to describe the concentration, dispersion, and asymmetry of anomalous fluctuations. Time-domain trend features involve linearly fitting the data within the window using the least squares method to calculate its slope; simultaneously, the autocorrelation coefficient of the data is calculated to assess its periodicity or persistence. Frequency-domain transformation features involve performing a fast Fourier transform on the jitter or amplitude sequence within the window to extract the main frequency components and their power spectral density, quantifying the coupling relationship with mechanical vibration frequencies. Event count features involve statistically analyzing the frequency and percentage of anomalous events where the parameter exceeds different severity thresholds within the window.

[0094] Within the same time window, calculate the Pearson correlation coefficient between different abnormal parameters, quantify the correlation strength, and define the correlation between signal abnormal parameters and physical state parameters. For example, analyze whether the deterioration of eye diagram closure is synchronously enhanced with a specific bending angle change or vibration spectrum component.

[0095] All extracted features are combined in a preset fixed order to form a trend feature vector that represents the abnormal fluctuation pattern within the current window. The system continuously records the trend feature vector output by each sliding window and arranges them in chronological order to form a trend evolution sequence.

[0096] By analyzing the trend evolution sequence, the main trend component that best represents its direction of change is extracted. By detecting the directional stability and acceleration of change of the main trend component in the trend evolution sequence, it is determined whether the abnormal fluctuation is intensifying, stabilizing or easing, and a compressed trend signature is generated for subsequent matching and prediction.

[0097] This embodiment transforms continuous raw signal fluctuation data into a quantifiable trend description containing evolutionary information, providing high-quality input features for accurate matching with preset abnormal event patterns and model prediction.

[0098] Example 3

[0099] This embodiment provides a process for constructing a prediction model, including the following steps:

[0100] Step 1: Collect and integrate the dataset for model training. The dataset includes: displacement, vibration, and temperature sensing data during the entire process of normal conditions and various preset abnormal events; obtain simulation results of physical quantities corresponding to the sensing data, such as stress distribution, deformation modes, and theoretical impedance changes of the flexible circuit board body 3; based on the abnormal event library of the preset event matching module 85, label each time series data segment with its corresponding preset abnormal event type, event start point, development stage, and severity level; the dataset should cover at least 10,000 hours of continuous or segmented runtime time series data, and the sampling frequency should be no less than 10 kHz to ensure that details of vibration and transient electrical signal interference can be captured. The data should cover the operating temperature range from -10°C to 85°C, different bending radii, and comprehensive working conditions under swept-frequency vibration and random vibration.

[0101] Step 2: Preprocess the raw time-series sensing data, including denoising, normalization, and alignment. Then, extract shallow time-frequency features, deep abstract features, and physical guidance features through the following parallel paths: Shallow time-frequency features: extract the signal's standard deviation, root mean square, peak-to-peak value, and frequency domain energy distribution and dominant frequency obtained through Fast Fourier Transform and Wavelet Transform; Deep abstract features: automatically extract deep abstract features related to anomalous patterns from the raw or slightly processed time-series signals using a pre-trained convolutional neural network; Physical guidance features: use the physical simulation label data obtained in Step 1, such as maximum stress point values, local bending curvature, and theoretical impedance offset, as auxiliary features, directly concatenating and fusing them with the above shallow and deep features along the feature dimension; a guided feature concatenation approach is adopted, allowing the model to simultaneously learn the correlation between data-driven features and physical rule features during training, thereby enhancing the physical interpretability and extrapolation ability of the model's prediction results; the three types of features are concatenated and fused along the feature dimension to form a comprehensive feature vector;

[0102] Step 3: Construct the model framework, whose core architecture includes, in sequence, a feature embedding layer, a temporal dependency modeling unit, a spatial attention unit, and an event association and risk prediction layer, wherein:

[0103] Feature embedding layer: maps the fused comprehensive feature vector to a high-dimensional embedding space; Temporal dependency modeling unit: uses a bidirectional long short-term memory network to model the embedded feature sequence in order to capture the temporal dependency of abnormal fluctuation trends;

[0104] Spatial Attention Unit: Following the temporal dependency modeling unit, an attention mechanism module is introduced; this module automatically calculates the importance weights of different sensor data streams to the current prediction task, highlighting information about local spatial locations most relevant to potential faults;

[0105] Event Association and Risk Prediction Layer: Taking the contextual representation output from the aforementioned steps and weighted by spatiotemporal attention as input, it performs multi-task learning simultaneously through a fully connected network: Task 1 is multi-label abnormal event association prediction, outputting the probability of several preset abnormal events associated with the current state; Task 2 is risk value regression prediction, based on the output of Task 1 and the contextual representation, it regresses and predicts the specific risk value under various preset abnormal event performance indicators within a specified future period.

[0106] Step 4: Train the model using an optimization objective that includes a multi-task loss function. The loss function is defined as: Total Loss = α * Event Association Cross-Entropy Loss + β * Risk Value Mean Squared Error Loss + γ * Model Regularization Term; where α, β, and γ are hyperparameters that control the trade-off between classification accuracy, regression accuracy, and model complexity, respectively. The training process uses historical datasets and minimizes the total loss through backpropagation and an adaptive optimizer until the model converges.

[0107] Step 5: Evaluate the performance of the trained model using independent validation and test sets. Core evaluation metrics include: precision and recall for event association prediction, and mean absolute error for risk value prediction. These core evaluation metrics must meet preset performance thresholds: precision for event association prediction must be greater than 90%, and recall must be greater than 85% to balance false positives and false negatives; mean absolute error for risk value prediction must be less than 0.1 on a risk scale of 0-1 to ensure the accuracy of risk warnings. After validation, solidify the model parameters and deploy them online.

[0108] This embodiment deeply integrates real-time sensing data with the attribute data of the flexible circuit board body 3, and uses a spatiotemporal attention deep learning architecture. It can not only accurately identify the abnormal patterns of the current connection, but also explore the deep correlation between multi-dimensional temporal features and future failure risks. It can predict the development trajectory and risk value of multiple potential failure modes in advance, thereby providing a decision basis for dynamically adjusting signal parameters and heat dissipation strategies, realizing true preventive maintenance, and improving the long-term reliability and performance stability of the high-speed interconnect system.

[0109] In operation, this invention provides signal transmission between the optical module connection end 1 and the connector end 2 via the flexible circuit board body 3. The flexible circuit board body 3 is fixed by the pressing plate 5. The gold finger contact area on the left end of the flexible circuit board body 3 contacts the gold finger connection end 4 and is secured by flipping the pressing plate 5. Temperature data inside the gold finger connection end 4 is collected by the temperature sensor 11. Cooling is provided to the gold finger connection end 4 by the rotation of the fan 10. Vibration data of the gold finger connection end 4 is collected by the vibration sensor 9. When the optical module connection end 1 and the connector end 2 are displaced, a signal is emitted from the laser sensor transmitter 71 to the laser receiver 73 to obtain feedback, or a signal is emitted from the laser sensor transmitter 72 to the laser receiver 74 to obtain feedback. This ensures that even if there is a height difference between the optical module connection end 1 and the connector end 2, laser data can still be obtained, thereby acquiring the relative displacement data of the optical module connection end 1 and the connector end 2. The control unit 8 analyzes the sensor data to generate a signal reception and transmission scheme for the optical module connection end 1 and the connector end 2, as well as an adjustment scheme for the operating power of the fan 10.

[0110] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A flexible signal switching structure for interconnecting optical modules and high-speed connectors, characterized in that, The device includes an optical module connection end (1), a connector end (2) on the right side of the optical module connection end (1), a flexible circuit board body (3) on the left side of the connector end (2), a gold finger connection end (4) fixedly connected to the right side of the top of the optical module connection end (1), a pressing plate (5) for pressing the flexible circuit board body (3) is provided inside the gold finger connection end (4), the left side of the pressing plate (5) is rotatably connected to the inner wall of the gold finger connection end (4), the left side of the flexible circuit board body (3) is electrically connected to the inner wall of the bottom end of the gold finger connection end (4), a temperature sensor (11) is installed at the bottom end of the gold finger connection end (4), and the left side of the gold finger connection end (4) is fixedly connected to the inner wall of the bottom end of the gold finger connection end (4). The flexible circuit board body (3) has a hollow shell (6), and a fan (10) is installed inside the hollow shell (6). A displacement sensing component (7) is provided on the front side of the flexible circuit board body (3). The displacement sensing component (7) includes a laser sensor transmitter (71), a laser sensor transmitter (72), a laser receiver (73), and a laser receiver (74). The laser sensor transmitter (72) is installed on the right side of the top of the optical module connection end (1). The laser sensor transmitter (71) is installed on the top of the laser sensor transmitter (72). The laser receiver (73) is installed on the left side of the top of the connector end (2). The laser receiver (74) is installed on the back side of the laser receiver (73). The top of the optical module connection end (1) is equipped with a control unit (8). The control unit (8) includes a status recognition module and a prediction control module. The status recognition module and the prediction control module are electrically connected. The status recognition module is used to fuse the sensing data of laser sensor transmitter one (71), laser sensor transmitter two (72), vibration sensor (9) and temperature sensor (11) with the physical properties of the flexible circuit board body (3) to simulate and monitor the deformation, connection status and environmental conditions of the flexible circuit board body (3) in real time. The prediction control module is used to analyze the abnormal fluctuation trend based on the monitoring data obtained by the status recognition module, match potential faults, predict risks through the prediction model preset by the deep learning algorithm, and drive the data interaction settings of the optical module connection end (1) and the connector end (2) and the start-up settings of the fan (10) for adaptive control.

2. The flexible signal switching structure for interconnecting optical modules and high-speed connectors according to claim 1, characterized in that, The second laser receiver (74) is used to receive the laser signal emitted by the second laser sensor transmitter (72), and the first laser receiver (73) is used to receive the laser signal emitted by the first laser sensor transmitter (71). The first laser receiver (73) and the second laser receiver (74) identify the received laser signal to output the relative displacement data between the optical module connection end (1) and the connector end (2). The displacement data calculation process is as follows: The data output from the first laser receiver (73) is used to calculate the lateral displacement and deflection angle of the optical module connection end (1) relative to the connector end (2) in the X and Y axis planes; the data output from the second laser receiver (74) is used to calculate the displacement and angle change data of the connector end (2) relative to the optical module connection end (1) in the complementary direction; the two sets of relative displacement data are fused and coordinate transformed to synthesize and output the relative displacement vector and attitude angle change data between the optical module connection end (1) and the high-speed connector end (2) in three-dimensional space.

3. The flexible signal switching structure for interconnecting optical modules and high-speed connectors according to claim 1, characterized in that, The status recognition module has sub-modules deployed below it, including a data extraction module (81), a working condition simulation module (82), and a connection status monitoring module (83). The working condition simulation module (82) is electrically connected to the data extraction module (81) and the connection status monitoring module (83), wherein: The data extraction module (81) is used to acquire in real time the three-dimensional displacement data, vibration spectrum data and temperature data of the contact point between the optical module connection end (1) and the connector end (2) and the contact point between the flexible circuit board body (3) and the gold finger connection end (4), and to perform standardization processing. The working condition simulation module (82) is used to extract the standardized sensor data and combine it with the pre-stored material mechanical property data and geometric structure data of the flexible circuit board body (3) to simulate and generate the dynamic flexible bending three-dimensional shape and stress distribution of the flexible circuit board body (3) under the current working condition in real time through the simulation model. The connection status monitoring module (83) is used to comprehensively determine the real-time tightness of the pressing plate (5) and the connection integrity status between the flexible circuit board body (3) and the connector end (2) based on the output of the working condition simulation module (82).

4. The flexible signal switching structure for interconnecting optical modules and high-speed connectors according to claim 3, characterized in that, The process of constructing and running the simulation model in the operating condition simulation module (82) includes: Based on the three-dimensional model of the flexible circuit board body (3), it is discretized into a finite element mesh, and based on its pre-stored material mechanical property data, each element in the mesh is assigned a corresponding deformation load; The three-dimensional displacement data obtained in real time by the data extraction module (81) is used as the dynamic displacement boundary condition and loaded onto the nodes in the finite element mesh corresponding to the installation positions of the optical module connection end (1) and the connector end (2); at the same time, the real-time temperature data is used as the thermal load boundary condition and applied to the entire network to obtain thermal strain. Within each sampling period, based on the acquired deformation load and thermal strain data, the overall deformation field of the flexible circuit board body (3) under the current displacement and temperature conditions, the stress distribution of each layer structure, and the maximum equivalent stress point and the maximum bending curvature region are defined. Based on the obtained stress distribution, combined with the resistance strain coefficient and thermal expansion coefficient of the flexible circuit board body (3), the real-time changes in parasitic resistance, inductance and capacitance parameters caused by deformation and temperature changes on the key transmission path of the flexible circuit board body (3) are obtained through a predefined mapping relationship.

5. A flexible signal switching structure for interconnecting optical modules and high-speed connectors according to claim 3, characterized in that, The operation of the connection status monitoring module (83) includes: During the initial calibration phase, the feedback of pressing and welding changes corresponding to the inherent vibration spectrum collected by the vibration sensor (9) is recorded under the pressing state of the known good gold finger connection end (4) and the pressing plate (5), and the welding state of the connector end (2) and the flexible circuit board body (3), and is used as the baseline spectrum of the normal state. The wideband vibration data output by the vibration sensor (9) is received in real time, and converted into a real-time vibration spectrum through fast Fourier transform. The amplitude characteristics, spectral energy distribution characteristics and resonant peak frequency shift characteristics of a specific frequency band are extracted. The theoretical bending shape and stress distribution of the flexible circuit board body (3) simulated based on the current displacement and temperature conditions are received from the working condition simulation module (82), and the expected normal vibration response range of the system under the current working condition is defined. The real-time vibration characteristics were compared with the baseline spectrum under normal conditions and the vibration response range, respectively: When the real-time vibration spectrum shows a sustained increase in amplitude above the threshold or a new resonance peak in the high-frequency band, it is determined that there is microscopic contact loosening or foreign matter between the gold finger connection end (4) and the pressing plate (5), and the tight closure state deteriorates. When the real-time vibration spectrum shows a vibration energy change that is strongly correlated with the bending shape change of the flexible circuit board body (3) in a specific low-frequency band, but its change pattern deviates from the baseline spectrum of the normal state, it is determined that there is a false soldering crack or fatigue damage at the welding end of the flexible circuit board body (3) and the connector end (2).

6. The flexible signal switching structure for interconnecting optical modules and high-speed connectors according to claim 1, characterized in that, The predictive control module has sub-modules deployed at its lower level, including an abnormal fluctuation analysis module (84), a preset event matching module (85), an intelligent prediction module (86), and an adaptive control module (87). The intelligent prediction module (86) is electrically connected to the abnormal fluctuation analysis module (84), the preset event matching module (85), and the adaptive control module (87), wherein: The abnormal fluctuation analysis module (84) is used to start the analysis process when the state recognition module detects that the flexible circuit board body (3) has a bending deformation exceeding the preset threshold or detects abnormal vibration, to determine whether there are abnormal fluctuations in amplitude or timing of the electrical signal transmitted through the flexible circuit board body (3) under the current temperature conditions, and to extract the trend of continuous abnormal fluctuation data. The preset event matching module (85) is used to match the abnormal fluctuation trend characteristics extracted by the abnormal fluctuation analysis module (84) with the pre-established abnormal event library. The abnormal event library includes multiple preset abnormal events and their corresponding fluctuation characteristic patterns associated with the flexible circuit board body (3), the pressing connection, and the welding connection. It outputs one or more candidate preset abnormal events that may be associated. The intelligent prediction module (86) is used to build and train the prediction model. The model takes the abnormal fluctuation trend characteristics and the candidate preset abnormal event sequence output by the preset event matching module (85) as input features. After training, it predicts the risk value of the current flexible circuit board body (3) under various preset abnormal event performance indicators within a specified time period in the future. The adaptive control module (87) is used to receive the predicted risk value output by the intelligent prediction module (86). When the risk value exceeds the preset level, it automatically generates control instructions including preventive adjustment of the signal transmission parameters of the optical module connection end (1) and connector end (2) through the flexible circuit board body (3) channel, and adaptive change of the operating power of the associated wind turbine (10).

7. A flexible signal switching structure for interconnecting an optical module and a high-speed connector according to claim 6, characterized in that, The abnormal fluctuation analysis module (84) performs the abnormal fluctuation judgment process, including: When the flexible circuit board body (3) is in a standard static state and at a standard operating temperature, the high-speed signal transmitted by it is collected and analyzed, and a standard signal quality template is generated and stored. The template includes a standard signal eye diagram profile, an allowable jitter tolerance range, and normal amplitude and rise or fall time parameters. Under the current working conditions, the signal transmitted through the flexible circuit board body (3) is captured in real time, and the real-time eye diagram shape, total jitter, random jitter, deterministic jitter, signal amplitude and edge rate of the current signal are extracted simultaneously. The real-time signal quality parameters extracted synchronously are dynamically compared with the standard template, and the actual deviation values ​​of each parameter are calculated. Combined with the real-time temperature data obtained by the data extraction module (81), temperature influence compensation calculation is performed on the deviation values ​​of various signal quality parameters; the compensated deviation values ​​are compared with the preset dynamic abnormal thresholds for different temperature levels. When the deviation of one or more key parameters continues to exceed its corresponding threshold, it is determined that there is abnormal fluctuation in the signal.

8. A flexible signal switching structure for interconnecting an optical module and a high-speed connector according to claim 6, characterized in that, The trend extraction process of abnormal fluctuation data in the abnormal fluctuation analysis module (84) includes: Establish a data buffer to synchronously receive and store the time sequence of quality parameters of each signal that has been determined to be abnormal; Define a sliding time window of configurable length on the data buffer; within each window, extract statistical features, time-domain trend features, frequency-domain transformation features, and event count features for the sequence of each outlier parameter. Within the same time window, calculate the Pearson correlation coefficient between different anomaly parameters, quantify the correlation strength, and define the correlation between signal anomaly parameters and physical state parameters; All extracted features are combined in a preset fixed order to form a trend feature vector that represents the abnormal fluctuation pattern within the current window. The system continuously records the trend feature vectors output by each sliding window and arranges them in chronological order to form a trend evolution sequence.

9. A flexible signal switching structure for interconnecting an optical module and a high-speed connector according to claim 6, characterized in that, The construction process of the prediction model in the intelligent prediction module (86) includes the following steps: Step 1: Collect and integrate the dataset for model training; The dataset includes: displacement, vibration and temperature sensing data of the entire process of normal state and various preset abnormal events; Obtain the physical quantity simulation results of stress distribution, deformation mode and theoretical impedance change of the flexible circuit board body (3) corresponding to the sensing data; Based on the abnormal event library of the preset event matching module (85), mark the corresponding preset abnormal event type, event start point, development stage and severity level for each time series data segment; Step 2: Preprocess the raw time-series sensing data, including denoising, normalization and alignment; then, extract shallow time-frequency features, deep abstract features and physical guidance features through the following parallel path; the three types of features are spliced ​​and fused in the feature dimension to form a comprehensive feature vector; Step 3: Construct the model framework, whose core architecture includes, in sequence, a feature embedding layer, a temporal dependency modeling unit, a spatial attention unit, and an event association and risk prediction layer; Step 4: Train the model using an optimization objective that includes a multi-task loss function. During training, use historical datasets and minimize the total loss through backpropagation and an adaptive optimizer until the model converges. Step 5: Evaluate the performance of the trained model using independent validation and test sets. After validation, solidify the model parameters and deploy it online.

10. A flexible signal switching structure for interconnecting an optical module and a high-speed connector according to claim 6, characterized in that, The adaptive control module (87) makes preventative adjustments to the optical module connection end (1) and connector end (2) according to the risk value and a preset scale, including: adjusting the pre-emphasis coefficient of the transmitting end, the equalizer parameters of the receiving end, and the data rate or encoding scheme of the signal link according to the risk value; the adaptive control module (87) makes control instructions to the fan (10) including: when the predicted risk value is strongly correlated with the temperature, increasing the fan (10) speed according to the predicted risk value and a corresponding scale.

Images