A machine learning-based data line connection failure prediction method and system
By acquiring multidimensional data of the data line through machine learning methods and constructing a normal data distribution using temporal deep neural networks and multi-head attention mechanisms, the problem of the inability to dynamically perceive data line faults in existing technologies is solved, enabling early prediction and multi-level assessment of data line faults.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANXI AGRI UNIV
- Filing Date
- 2026-04-27
- Publication Date
- 2026-07-31
AI Technical Summary
Existing data line connection protection schemes lack dynamic perception and global analysis, making it difficult to capture the gradual degradation process before a fault occurs. They can only passively alarm when parameters exceed preset thresholds, and cannot make advance predictions.
A machine learning-based approach is adopted to acquire electrical, physical, and environmental feature data of the data line. Data preprocessing and modeling are performed using a temporal deep neural network and a multi-head attention mechanism to construct a normal data distribution and to determine faults based on deviations.
It enables early prediction of data cable failures and provides multi-level fault assessments, including health, attention, early warning, and emergency, before a failure occurs, thereby improving the reliability and security of data cable connections.
Smart Images
Figure CN122490118A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data cable fault prediction technology, and more specifically to a data cable connection fault prediction method and system based on machine learning. Background Technology
[0002] Existing data cable connection protection schemes often rely on fixed threshold monitoring of single parameters (such as overcurrent and overtemperature), lacking dynamic perception and global analysis of the data cable's operating status, making them difficult to adapt to complex and ever-changing real-world application scenarios. These methods cannot capture the gradual degradation process before a fault occurs, and can only passively trigger an alarm when the parameter exceeds a preset threshold. Therefore, a data cable connection fault prediction method and system based on machine learning is needed to predict data cable faults in advance based on collected multi-dimensional data. Summary of the Invention
[0003] The purpose of this invention is to provide a data cable connection fault prediction method and system based on machine learning. This prediction method can predict data cable faults in advance based on collected multidimensional data.
[0004] To achieve the above objectives, embodiments of the present invention provide a data cable connection fault prediction method based on machine learning, the prediction method comprising: Acquire the electrical characteristics, physical state, and environmental characteristics of the data cable as raw data. The raw data from the data line is preprocessed; After preprocessing, the preprocessed raw data is fed into a temporal deep neural network to obtain predicted values within a set time period. Based on the electrical characteristics, physical state, and environmental characteristics of the data line under normal conditions as a training set, a normal data distribution is constructed. Fault determination is based on the deviation between the predicted value and the normal data distribution.
[0005] Optionally, the electrical characteristic dimension data includes contact resistance, characteristic impedance, signal attenuation, return loss, and bit error rate; the physical state dimension data includes cumulative insertion and removal times, interface physical deformation detection, and temperature change rate; and the environmental characteristic dimension data includes ambient humidity, temperature, and dust concentration.
[0006] Optionally, after preprocessing, the preprocessed raw data is fed into a temporal deep neural network to obtain predicted values within a set time period, including: The temporal evolution of the preprocessed original data is encoded and modeled using a bidirectional long short-term memory network. A multi-head attention mechanism is introduced to dynamically weight the differential contributions of raw data with different feature dimensions during the degradation process; Based on the feature representation of the original data after temporal feature encoding and attention weighting, the predicted value is output through a fully connected network layer.
[0007] Optionally, a normal data distribution is constructed based on the electrical characteristic dimension data, physical state dimension data, and environmental characteristic dimension data of the data line under normal conditions as a training set, including: Obtain the multidimensional data in the training set, and set the corresponding threshold radius and quantity threshold; Construct the degradation trajectory points of each dimension of the data in the multidimensional data of the data line under normal conditions, related to the time dimension; Randomly select degenerate trajectory points of the corresponding dimension, and define a range with the selected degenerate trajectory points as the center and the threshold radius as the radius. Select a set of degenerate trajectory points within the defined range whose number of points is greater than the number threshold as the first set, and use the degenerate trajectory point at the center of the circle as the selection point. In the first set, a degenerate trajectory point is randomly selected, and it is determined whether the number of points within a range defined by the selected degenerate trajectory point as the center and the threshold radius as the radius exceeds the number threshold. If the number of points exceeds the threshold, the selected degenerate trajectory points are used as new selection points, and the unclassified degenerate trajectory points within the range centered on the new selection points are added to the first set. If the number threshold is not exceeded and the selected degenerate trajectory point is not in the first set, the selected degenerate trajectory point will be added to the first set. Select the degenerate trajectory points that have not been selected in the first set, and return to the step of determining whether the number of points within a range defined by the selected degenerate trajectory point as the center and the threshold radius as the radius exceeds the number threshold, until all degenerate trajectory points in the first set have been selected; Repeat the operation of randomly selecting degenerate trajectory points that were not selected in the training set and then aggregating them into the first set until all degenerate trajectory points have been selected, and obtain all selected points in the aggregating process; Based on the selected points, a selection curve is fitted according to the time dimension, and the normal data distribution area is constructed using the selection curve as the baseline.
[0008] Optionally, a selection curve is fitted based on the selected points according to the time dimension, and a normal data distribution area is constructed using the selection curve as a baseline, including: obtaining the selection curve and setting the upper and lower tolerance zones of the selection curve, wherein the area between the upper and lower tolerance zones is the normal data distribution area.
[0009] Optionally, fault determination is based on the deviation between the predicted value and the normal data distribution, including: The predicted value and the original data within a predetermined time period preceding the predicted value are collected as comparison data. The comparison data is fitted into a time-series-based comparison curve; Unify the comparison curve and the selection curve in the same coordinate system; The comparison curve and the selection curve are compared, and the curve that is most similar to the comparison curve in the selection curve is selected as the determining curve. Determine the deviation between the predicted value of the comparison curve for each dimension and the upper or lower limit of the normal data distribution area corresponding to the determined curve; The degree of failure of the data line is calculated by comprehensively considering the deviation between the predicted value of the comparison curve for each dimension and the upper or lower limit of the corresponding normal data distribution area.
[0010] Optionally, based on the deviation between the predicted value of the comparison curve for each dimension and the upper or lower limit of the corresponding normal data distribution area, the degree of failure of the data line is comprehensively calculated, including: The connection fault levels of the data cable are defined as follows: Healthy, Attention, Warning, and Emergency. Obtain the deviation value between the comparison curve of each dimension affecting the fault cause of the data line and the upper or lower limit of the corresponding normal data distribution area, and determine the membership degree between the deviation value corresponding to each dimension and the connection fault level to obtain the evaluation matrix. The evaluation matrices are set together to obtain a fuzzy relation matrix; Calculate the weights of each dimension of the data line to obtain a weight matrix; Multiply the weight matrix by the fuzzy relation matrix to obtain the comprehensive evaluation vector; Based on the matching degree between the comprehensive evaluation vector and the connection fault level, the level with the highest matching degree is selected as the connection fault level of the data line, thereby determining the degree of fault of the data line.
[0011] Optionally, based on the deviation between the predicted value of the comparison curve for each dimension and the upper or lower limit of the corresponding normal data distribution area, the degree of failure of the data line is comprehensively calculated, including: Select the curve that is most similar to the comparison curve from the selected curves as the determined curve, and calculate the similarity between the determined curve and the comparison curve; Obtain the slope of the determined curve and the comparison curve at the end of the curve segment; Based on the slope and similarity of the determined curve and the comparison curve at the end of the segment, the slope of the predicted curve is obtained by weighting the slope of the end segment of the determined curve and the comparison curve using the similarity. Based on the data at the end of the comparison curve and the slope of the prediction curve, the predicted data for the next time period is calculated. The degree of failure of the data line in the next time period is calculated based on the deviation between the predicted data for the next time period and the upper or lower limit of the corresponding normal data distribution area.
[0012] On the other hand, the present invention can also provide a data cable connection fault prediction system based on machine learning, the prediction system comprising: The data acquisition module is used to acquire electrical characteristic dimension data, physical state dimension data, and environmental characteristic dimension data of the data cable. The prediction module is used to execute a data line connection fault prediction method based on machine learning, as described above, based on the data obtained by the data acquisition module.
[0013] Through the above technical solution, the present invention provides a data cable connection fault prediction method and system based on machine learning. This method acquires electrical dimension data, physical state dimension data, and environmental dimension data of the data cable as raw data. The raw data is then preprocessed. After preprocessing, the preprocessed raw data is fed into a temporal deep neural network to obtain predicted values for a set time period. A normal data distribution can be constructed using the electrical dimension data, physical state dimension data, and environmental feature dimension data of the data cable under normal conditions as a training set. Fault judgment can be made based on the deviation between the predicted value and the normal data distribution. This allows for the prediction of data cable connection faults. This prediction method can predict data cable faults in advance based on the collected multi-dimensional data.
[0014] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description
[0015] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a data cable connection fault prediction method based on machine learning according to an embodiment of the present invention. Detailed Implementation
[0016] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the scope of the present invention.
[0017] In the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, they do not mean that the applicant has used or necessarily used the solution.
[0018] Figure 1 This is a flowchart of a data cable connection fault prediction method based on machine learning according to an embodiment of the present invention. In this invention, the prediction method may include the following steps: In step S1, the electrical characteristic dimension data, physical state dimension data, and environmental characteristic dimension data of the data line are obtained as raw data.
[0019] In step S2, the raw data of the data line is preprocessed.
[0020] In step S3, after preprocessing is completed, the preprocessed raw data is fed into the temporal deep neural network to obtain the predicted value within the set time period.
[0021] In step S4, a normal data distribution is constructed based on the electrical characteristic dimension data, physical state dimension data, and environmental characteristic dimension data of the data line under normal conditions as a training set.
[0022] In step S5, fault judgment is made based on the deviation between the predicted value and the normal data distribution.
[0023] In this invention, during fault prediction, electrical dimension data, physical state dimension data, and environmental dimension data of the data line can be acquired as raw data. This raw data can then be preprocessed. After preprocessing, the preprocessed raw data can be fed into a temporal deep neural network to obtain predicted values for a set time period. A normal data distribution can be constructed using the electrical dimension data, physical state dimension data, and environmental feature dimension data of the data line under normal conditions as a training set. Fault judgment can be made based on the deviation between the predicted value and the normal data distribution. This allows for fault prediction of the data line connection. This prediction method can predict data line faults in advance based on the collected multidimensional data.
[0024] In one embodiment of the present invention, the electrical characteristic dimension data may include contact resistance, characteristic impedance, signal attenuation, return loss, bit error rate, etc. The physical state dimension data may include cumulative insertion / removal counts, interface physical deformation detection, temperature change rate, etc. The environmental characteristic dimension data may include ambient humidity, temperature, dust concentration, etc.
[0025] In one embodiment of the present invention, the process of obtaining the predicted value may include: In step S6, the temporal evolution of the preprocessed raw data is encoded and modeled using a bidirectional long short-term memory network.
[0026] In step S7, a multi-head attention mechanism is introduced to dynamically weight the differential contributions of the original data of different feature dimensions in the degradation process.
[0027] In step S8, based on the feature representation of the original data after temporal feature encoding and attention weighting, the predicted value is output through a fully connected network layer.
[0028] In this invention, a bidirectional long short-term memory (Bi-LSTM) network can be used to model the temporal evolution of multidimensional features when obtaining predicted values. Bi-LSTM can simultaneously capture the historical degradation trend and current direction of change of connection parameters, making it suitable for the gradual and non-linear temporal process of data line connection degradation. Its bidirectional structure allows the model to learn both the overall degradation trajectory from health to failure and to identify abnormal patterns in short-term fluctuations. A multi-head attention mechanism can then be introduced to dynamically weight the differentiated contributions of different feature dimensions during the degradation process. The causes of data line connection failures are phased; in the early degradation stage, changes in electrical parameters have a higher weight, such as a slow increase in contact resistance; in the mid-to-late stages, the weight of physical state factors increases, such as the cumulative effect of insertion and removal; and environmental factors act as global accelerating factors throughout the entire process. The multi-head attention mechanism automatically learns this dynamic weight allocation strategy, enabling the model to focus on the most discriminative feature dimensions at different degradation stages. Finally, based on the feature representation of the original data after temporal feature encoding and attention weighting, predicted values can be output through a fully connected network layer.
[0029] In one embodiment of the present invention, the process of constructing a normal data distribution may include: In step S9, multidimensional data in the training set is obtained, and the corresponding threshold radius and quantity threshold are set.
[0030] In step S10, the degradation trajectory points of each dimension of the data in the multidimensional data of the data line under normal conditions are constructed in relation to the time dimension.
[0031] In step S11, degenerate trajectory points of the corresponding dimension are randomly selected, and a range is defined with the selected degenerate trajectory points as the center and the threshold radius as the radius. The set of degenerate trajectory points with a number of points greater than the number threshold within the defined range is selected as the first set, and the degenerate trajectory point at the center of the circle is selected as the selection point.
[0032] In step S12, degenerate trajectory points are randomly selected from the first set, and it is determined whether the number of points within a range defined by the selected degenerate trajectory point as the center and the threshold radius as the radius exceeds the number threshold.
[0033] In step S13, if the number threshold is exceeded, the selected degenerate trajectory point is taken as a new selection point, and the degenerate trajectory points that are not classified within the range defined by the new selection point are added to the first set.
[0034] In step S14, if the number threshold is not exceeded and the selected degenerate trajectory point is not in the first set, the selected degenerate trajectory point is added to the first set.
[0035] In step S15, select the degenerate trajectory points that were not selected in the first set, and return to step S12 until all degenerate trajectory points in the first set are selected.
[0036] In step S16, the operation of randomly selecting degenerate trajectory points that were not selected in the training set and then aggregating them into the first set is repeated until all degenerate trajectory points are selected, and all selected points in the aggregating process are obtained.
[0037] In step S17, a selection curve is fitted based on the selected points according to the time dimension, and the normal data distribution area is constructed using the selection curve as the baseline.
[0038] In this invention, when constructing the declarative data distribution, multidimensional data in the training set can be obtained, and corresponding threshold radii and quantity thresholds can be set. Then, degenerate trajectory points related to the time dimension of each dimension of the multidimensional data in the normal state of the data line can be constructed. After obtaining the degenerate trajectory points, degenerate trajectory points of the corresponding dimension can be randomly selected, and a range can be defined with the selected degenerate trajectory point as the center and the threshold radius as the radius. The set of degenerate trajectory points with a number of points greater than the quantity threshold within the defined range can be filtered as the first set. The degenerate trajectory point at the center of the selected circle can be used as the selection point. Degenerate trajectory points can be randomly selected in the first set, and it can be determined whether the number of points within the range defined with the selected degenerate trajectory point as the center and the threshold radius as the radius exceeds the quantity threshold. If the quantity threshold is exceeded, the selected degenerate trajectory point can be used as a new selection point, and the unclassified degenerate trajectory points within the range defined with the new selection point as the center can be added to the first set. If the quantity threshold is not exceeded, and the selected degenerate trajectory point is not in the first set, the selected degenerate trajectory point can be added to the first set. Select any degenerate trajectory points that were not selected in the first set and return to step S12, until all degenerate trajectory points in the first set have been selected. Repeat the operation of randomly selecting degenerate trajectory points that were not selected in the training set and then aggregating them into the first set, until all degenerate trajectory points have been selected, and obtain all selected points in the aggregating process. After obtaining the selected points, a selection curve can be fitted based on the selected points according to the time dimension, and the normal data distribution area can be constructed using the selection curve as the baseline.
[0039] In one embodiment of the present invention, when constructing a normal data distribution area, a selection curve can be obtained, and upper and lower tolerance zones of the selection curve can be set. Then, the area between the upper and lower tolerance zones can be the normal data distribution area.
[0040] In one embodiment of the present invention, the fault determination process may include: In step S18, the predicted value and the original data within a set time period prior to the predicted value are obtained and collected as comparison data.
[0041] In step S19, the comparison data is fitted into a comparison curve based on the time series.
[0042] In step S20, the comparison curve and the selection curve are unified in the same coordinate system.
[0043] In step S21, the comparison curve and the selected curve are compared, and the curve that is most similar to the comparison curve is selected as the determined curve.
[0044] In step S22, the deviation between the predicted value of the comparison curve for each dimension and the upper or lower limit of the normal data distribution area corresponding to the curve is determined.
[0045] In step S23, the degree of failure of the data line is calculated by comprehensively considering the deviation between the predicted value of the comparison curve for each dimension and the upper or lower limit of the corresponding normal data distribution area.
[0046] In this invention, during fault determination, predicted values and raw data within a predetermined time period preceding the predicted values can be acquired and aggregated into comparison data. This comparison data can be fitted into a time-series-based comparison curve. After obtaining the comparison curve, it and the selected curve can be unified in the same coordinate system. The comparison curve and the selected curve can then be compared, and the curve most similar to the comparison curve in the selected curve can be selected as the determining curve. The deviation between the predicted value of the comparison curve in each dimension and the upper or lower limit of the corresponding normal data distribution area can be determined. Based on the deviation between the predicted value of the comparison curve in each dimension and the corresponding upper or lower limit of the normal data distribution area, the degree of fault in the data line can be comprehensively calculated.
[0047] In one embodiment of the present invention, the process for determining the degree of fault may include: In step S24, the connection fault levels of the data cable are constructed, including: healthy, attention, warning, and emergency.
[0048] In step S25, the deviation value between the comparison curve of each dimension affecting the fault cause of the data line and the upper or lower limit of the corresponding normal data distribution area is obtained, and the membership degree between the deviation value corresponding to each dimension and the connection fault level is determined to obtain the evaluation matrix.
[0049] In step S26, the evaluation matrices are set together to obtain a fuzzy relation matrix.
[0050] In step S27, the weights of each dimension of the data line are calculated to obtain the weight matrix.
[0051] In step S28, the weight matrix is multiplied by the fuzzy relation matrix to obtain the comprehensive evaluation vector.
[0052] In step S29, based on the matching degree between the comprehensive evaluation vector and the connection fault level, the level with the highest matching degree is selected as the connection fault level of the data line, thereby determining the degree of fault of the data line.
[0053] In this invention, when determining the degree of fault, a connection fault level for the data line can first be constructed, including: healthy, attention, warning, and emergency. Then, the deviation values between the predicted values of the comparison curves for each dimension affecting the fault causes of the data line and the upper or lower limits of the corresponding normal data distribution area can be obtained. The membership degree between the deviation values corresponding to each dimension and the connection fault level can be determined, thus obtaining an evaluation matrix. The evaluation matrices can then be aggregated to obtain a fuzzy relation matrix. After obtaining the fuzzy relation matrix, the weights of each dimension of the data line can be calculated to obtain a weight matrix. The weights of each dimension can be calculated based on the importance and proportion of each dimension, and then normalized. The weight matrix can be multiplied by the fuzzy relation matrix to obtain a comprehensive evaluation vector. Based on the matching degree between the comprehensive evaluation vector and the connection fault level, the level with the highest matching degree can be selected as the connection fault level of the data line, thereby determining the degree of fault of the data line.
[0054] In one embodiment of the present invention, the process for determining the degree of fault may further include: In step S30, the curve that is most similar to the comparison curve is selected as the determination curve, and the similarity between the determination curve and the comparison curve is calculated.
[0055] In step S31, the slope of the determined curve and the comparison curve at the end segment is obtained.
[0056] In step S32, based on the slope of the final segment of the determined curve and the comparison curve and their similarity, the slope of the final segment of the determined curve and the comparison curve is weighted by the similarity to obtain the slope of the predicted curve.
[0057] In step S33, based on the data at the end of the comparison curve and the slope of the prediction curve, the predicted data for the next time period is calculated.
[0058] In step S34, the degree of failure of the data line in the next time period is calculated based on the deviation between the predicted data of the next time period and the upper or lower limit of the corresponding normal data distribution area.
[0059] In this invention, if, after predicting the predicted value for a given time period and performing fault assessment, the next stage of data line fault prediction is needed, the curve most similar to the comparison curve can be selected as the determining curve. The similarity between the determining curve and the comparison curve is calculated, and the slopes of both curves at their final segments can be obtained. Based on the slopes and similarity of the final segments of the determining and comparison curves, the slopes of the final segments can be weighted using the similarity to obtain the predicted curve slope. Based on the data at the end of the comparison curve and the predicted curve slope, the predicted data for the next time period can be calculated. Based on the deviation between the predicted data for the next time period and the upper or lower limit of the corresponding normal data distribution area, the degree of data line fault in the next time period can be comprehensively calculated. Therefore, this method allows for the prediction of data line fault conditions over a future period as needed.
[0060] On the other hand, the present invention can also provide a data cable connection fault prediction system based on machine learning. The prediction system includes a data acquisition module and a prediction module. The data acquisition module can be used to acquire electrical characteristic dimension data, physical state dimension data, and environmental characteristic dimension data of the data cable. The prediction module can be used to execute a machine learning-based data cable connection fault prediction method as described above based on the data acquired by the data acquisition module.
[0061] Through the above technical solution, the present invention provides a data cable connection fault prediction method and system based on machine learning. This method acquires electrical dimension data, physical state dimension data, and environmental dimension data of the data cable as raw data. The raw data is then preprocessed. After preprocessing, the preprocessed raw data is fed into a temporal deep neural network to obtain predicted values for a set time period. A normal data distribution can be constructed using the electrical dimension data, physical state dimension data, and environmental feature dimension data of the data cable under normal conditions as a training set. Fault judgment can be made based on the deviation between the predicted value and the normal data distribution. This allows for the prediction of data cable connection faults. This prediction method can predict data cable faults in advance based on the collected multi-dimensional data.
[0062] Those skilled in the art will understand that embodiments of this application can be provided as methods, approaches, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0063] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (methods), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0064] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0065] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0066] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.
[0067] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.
[0068] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.
[0069] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.
[0070] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.
Claims
1. A data cable connection fault prediction method based on machine learning, characterized in that, The prediction method includes: Acquire the electrical characteristics, physical state, and environmental characteristics of the data cable as raw data. The raw data from the data line is preprocessed; After preprocessing, the preprocessed raw data is fed into a temporal deep neural network to obtain predicted values within a set time period. Based on the electrical characteristics, physical state, and environmental characteristics of the data line under normal conditions as a training set, a normal data distribution is constructed. Fault determination is based on the deviation between the predicted value and the normal data distribution.
2. The prediction method according to claim 1, characterized in that, The electrical characteristic dimension data includes contact resistance, characteristic impedance, signal attenuation, return loss, and bit error rate; Physical state data includes cumulative insertion / removal counts, interface physical deformation detection, and temperature change rate; Environmental characteristic dimensions include: ambient humidity, temperature, and dust concentration.
3. The prediction method according to claim 1, characterized in that, After preprocessing, the preprocessed raw data is fed into a temporal deep neural network to obtain predicted values for a set time period, including: The temporal evolution of the preprocessed original data is encoded and modeled using a bidirectional long short-term memory network. A multi-head attention mechanism is introduced to dynamically weight the differential contributions of raw data with different feature dimensions during the degradation process; Based on the feature representation of the original data after temporal feature encoding and attention weighting, the predicted value is output through a fully connected network layer.
4. The prediction method according to claim 1, characterized in that, Based on the electrical characteristic dimension data, physical state dimension data, and environmental characteristic dimension data of the data line under normal conditions as a training set, a normal data distribution is constructed, including: Obtain the multidimensional data in the training set, and set the corresponding threshold radius and quantity threshold; Construct the degradation trajectory points of each dimension of the data in the multidimensional data of the data line under normal conditions, related to the time dimension; Randomly select degenerate trajectory points of the corresponding dimension, and define a range with the selected degenerate trajectory points as the center and the threshold radius as the radius. Select a set of degenerate trajectory points within the defined range whose number of points is greater than the number threshold as the first set, and use the degenerate trajectory point at the center of the circle as the selection point. In the first set, a degenerate trajectory point is randomly selected, and it is determined whether the number of points within a range defined by the selected degenerate trajectory point as the center and the threshold radius as the radius exceeds the number threshold. If the number of points exceeds the threshold, the selected degenerate trajectory points are used as new selection points, and the unclassified degenerate trajectory points within the range centered on the new selection points are added to the first set. If the number threshold is not exceeded and the selected degenerate trajectory point is not in the first set, the selected degenerate trajectory point will be added to the first set. Select the degenerate trajectory points that have not been selected in the first set, and return to the step of determining whether the number of points within a range defined by the selected degenerate trajectory point as the center and the threshold radius as the radius exceeds the number threshold, until all degenerate trajectory points in the first set have been selected; Repeat the operation of randomly selecting degenerate trajectory points that were not selected in the training set and then aggregating them into the first set until all degenerate trajectory points have been selected, and obtain all selected points in the aggregating process; Based on the selected points, a selection curve is fitted according to the time dimension, and the normal data distribution area is constructed using the selection curve as the baseline.
5. The prediction method according to claim 4, characterized in that, Based on the selected points, a selection curve is fitted according to the time dimension, and a normal data distribution area is constructed using the selection curve as a baseline. This includes: obtaining the selection curve and setting the upper and lower tolerance zones of the selection curve, wherein the area between the upper and lower tolerance zones is the normal data distribution area.
6. The prediction method according to claim 4, characterized in that, Fault determination based on the deviation between the predicted value and the normal data distribution includes: The predicted value and the original data within a predetermined time period preceding the predicted value are collected as comparison data. The comparison data is fitted into a time-series-based comparison curve; Unify the comparison curve and the selection curve in the same coordinate system; The comparison curve and the selection curve are compared, and the curve that is most similar to the comparison curve in the selection curve is selected as the determining curve. Determine the deviation between the predicted value of the comparison curve for each dimension and the upper or lower limit of the normal data distribution area corresponding to the determined curve; The degree of failure of the data line is calculated by comprehensively considering the deviation between the predicted value of the comparison curve for each dimension and the upper or lower limit of the corresponding normal data distribution area.
7. The prediction method according to claim 6, characterized in that, Based on the deviation between the predicted value of the comparison curve for each dimension and the upper or lower limit of the corresponding normal data distribution area, the degree of failure of the data line is calculated comprehensively, including: The connection fault levels of the data cable are defined as follows: Healthy, Attention, Warning, and Emergency. Obtain the deviation value between the comparison curve of each dimension affecting the fault cause of the data line and the upper or lower limit of the corresponding normal data distribution area, and determine the membership degree between the deviation value corresponding to each dimension and the connection fault level to obtain the evaluation matrix. The evaluation matrices are set together to obtain a fuzzy relation matrix; Calculate the weights of each dimension of the data line to obtain a weight matrix; Multiply the weight matrix by the fuzzy relation matrix to obtain the comprehensive evaluation vector; Based on the matching degree between the comprehensive evaluation vector and the connection fault level, the level with the highest matching degree is selected as the connection fault level of the data line, thereby determining the degree of fault of the data line.
8. The prediction method according to claim 7, characterized in that, Based on the deviation between the predicted value of the comparison curve for each dimension and the upper or lower limit of the corresponding normal data distribution area, the degree of failure of the data line is calculated comprehensively, including: Select the curve that is most similar to the comparison curve from the selected curves as the determined curve, and calculate the similarity between the determined curve and the comparison curve; Obtain the slope of the determined curve and the comparison curve at the end of the curve segment; Based on the slope and similarity of the determined curve and the comparison curve at the end of the segment, the slope of the predicted curve is obtained by weighting the slope of the end segment of the determined curve and the comparison curve using the similarity. Based on the data at the end of the comparison curve and the slope of the prediction curve, the predicted data for the next time period is calculated. The degree of failure of the data line in the next time period is calculated based on the deviation between the predicted data for the next time period and the upper or lower limit of the corresponding normal data distribution area.
9. A data cable connection fault prediction system based on machine learning, characterized in that, The prediction system includes: The data acquisition module is used to acquire electrical characteristic dimension data, physical state dimension data, and environmental characteristic dimension data of the data cable. The prediction module is used to execute a data line connection fault prediction method based on any one of claims 1-8, according to the data obtained by the data acquisition module.