A Deep Learning-Based Diagnostic and Analysis Method for IoT Data Transmission Links

Abstract

This invention provides a deep learning-based diagnostic analysis method for IoT data transmission links, belonging to the field of transmission link diagnostic technology. Specific steps include: dividing the transmission into multiple levels of segments and multiple continuous, non-overlapping windows; constructing feature vectors and inputting them into a trained deep learning model to obtain the state of each window; locating the first abnormal window of each transmission segment, determining the initial segment and the earliest abnormal window, obtaining the next-level abnormal associated windows and abnormal overlapping windows, calculating the abnormal propagation delay time and counting the total number of abnormal windows; and correcting the total number of abnormal windows by combining window correction coefficients and abnormal overlap ratios, locking the abnormal time range and achieving abnormal point location. This invention overcomes the shortcomings of time-dimensional dynamic diagnosis, solves the problem of ambiguous abnormal time-series boundary location, significantly improves the time accuracy and timeliness of link anomaly diagnosis, and effectively solves the problem of inter-segment confusion in scenarios of multi-segment synchronous anomalies and cross-segment propagation.

Description

Technical Field

[0001] This invention relates to the field of transmission link diagnostic technology, specifically to a method for diagnostic analysis of IoT data transmission links based on deep learning. Background Technology

[0002] With the rapid penetration of IoT technology, complex application scenarios such as industrial IoT device interconnection, smart city data transmission, and multi-terminal communication in smart homes continue to expand. Massive numbers of terminal nodes rely on multi-link concurrent modes for data transmission, making the stability and reliability of data transmission links increasingly crucial for ensuring efficient system operation. However, IoT links generally suffer from inherent characteristics such as wide node distribution, complex and variable transmission environments, and frequent and diverse interference factors. Once transmission anomalies such as data packet loss, signal attenuation, or sharp increases in latency occur, they can easily trigger a chain reaction of problems, including industrial production interruptions, urban service disruptions, and home control command failures. Against this backdrop, achieving accurate identification and rapid location of abnormal nodes has become a critical need that urgently needs to be addressed in the field of IoT operation and maintenance.

[0003] In the prior art, CN115412430A discloses a method, apparatus, electronic device, and readable storage medium for locating abnormal nodes, comprising the following steps: acquiring data quality parameters of data transmitted between a data receiver and a data provider; when the data quality parameters indicate the presence of a data transmission anomaly, acquiring network quality parameters of nodes traversed by the data in the data transmission link according to the access mode corresponding to the data receiver; determining the abnormal node based on the difference between the network quality parameters of the traversed nodes; and identifying the source node in the data transmission link that caused the data transmission anomaly, starting from the abnormal node. This location method can improve the efficiency of anomaly location to a certain extent, and further identifying the source node causing the data transmission anomaly starting from the abnormal node can improve the accuracy of anomaly location to a certain extent.

[0004] However, the following shortcomings still exist: the existing technology lacks a dynamic diagnostic mechanism in the time dimension, which makes it impossible to accurately lock the abnormal time boundary and establish the transmission time correlation of abnormalities between multi-level transmission segments. The lack of time correlation makes it impossible to build an effective abnormal transmission inference model by means of abnormal overlap window filtering. In the case of multi-segment synchronous abnormalities and cross-segment transmission scenarios, it is easy to confuse the initial abnormal segment with the transmission abnormal segment, which in turn leads to misjudgment of the source node.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for diagnosing and analyzing IoT data transmission links based on deep learning, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A deep learning-based diagnostic analysis method for IoT data transmission links includes the following steps:

[0009] S1. Divide the data transmission link from the sensor to the gateway node in the IoT data transmission system into continuous multi-level transmission segments, and determine the link diagnosis interval formed by multiple continuous non-overlapping windows based on the current link diagnosis time.

[0010] S2. Collect scene environment data and multi-source data of all transmission segments in each window to construct the feature vector of each transmission segment in each window, and input it into the trained deep learning model to obtain the status of each transmission segment in each window, which is normal or abnormal.

[0011] S3. Locate the window when the state of each transmission segment first becomes abnormal, take the earliest abnormal transmission segment as the initial segment and locate the earliest abnormal window, analyze the multi-source data of the initial segment and the next level transmission segment to determine the abnormal association window of the next level transmission segment, combine the window when the state of the initial segment becomes abnormal to obtain the abnormal overlapping window, calculate the abnormal propagation delay time and count the total number of abnormal windows of the initial segment and the next level transmission segment.

[0012] S4. Determine the corresponding window correction coefficient based on the abnormal propagation delay time. Calculate the abnormal overlap ratio based on the number of abnormal overlapping windows and the number of windows when the initial segment is in an abnormal state. Using the total number of abnormal windows as a baseline, combine the abnormal overlap ratio and the window correction coefficient to obtain the corrected total number of abnormal windows, thereby locking in the time range of the abnormality and completing the location of the abnormal point in the transmission link.

[0013] Furthermore, the scene environment data includes the temperature, humidity, and electromagnetic interference intensity of the environment in which the transmission segment is located; the multi-source data includes the link transmission rate, signal attenuation amplitude, packet loss rate, and bit error rate of the transmission segment.

[0014] Furthermore, the data transmission link is divided into multiple consecutive transmission segments, with the specific logic as follows:

[0015] Using the signal transfer nodes and device interfaces in the data transmission link as the dividing points, the data flow is divided into the first-level transmission segment, the second-level transmission segment, ..., the nth-level transmission segment, according to the transmission path from the sensor end to the gateway end.

[0016] The first-level transmission segment is the transmission path from the sensor to the first boundary point, the second-level transmission segment is the transmission path from the first boundary point to the second boundary point, and so on, with the nth-level transmission segment being the transmission path from the last boundary point to the gateway.

[0017] The link diagnostic interval is determined by multiple consecutive, non-overlapping windows, and the specific logic is as follows:

[0018] Starting from the current link diagnosis time, trace back a preset time period to form a link diagnosis interval, and evenly divide it into multiple continuous non-overlapping windows.

[0019] Furthermore, the window at which each transmission segment first becomes abnormal is located. The transmission segment that first becomes abnormal is taken as the initial segment, and the earliest abnormal window is located. The specific logic is as follows:

[0020] The window at which the first abnormal state occurs in each transmission segment is taken as its starting abnormal window. The time sequence of the starting abnormal windows of each transmission segment is compared, and the transmission segment with the earliest time is selected as the initial segment. The starting abnormal window of the initial segment is defined as the earliest abnormal window.

[0021] If two or more transmission segments have the same earliest start abnormal window time, the transmission segment closest to the sensor end will be determined as the initial segment.

[0022] Furthermore, the abnormal association window of the next-level transmission segment is determined, and combined with the window of the initial segment that has an abnormal status, an abnormal overlap window is obtained. The specific logic is as follows:

[0023] The earliest anomaly window is used as the baseline window, and multi-source data within the baseline window of the initial segment are extracted to construct the baseline vector.

[0024] Extract multi-source data from each window of the next-level transmission segment to generate comparison vectors one by one. Calculate the cosine similarity between each comparison vector and the reference vector. Select windows with a cosine similarity greater than or equal to a preset similarity threshold that are not located before the reference window as abnormal association windows of the next-level transmission segment.

[0025] An intersection analysis is performed on the abnormal windows of the initial segment and the abnormal associated windows of the next-level transmission segment to identify the abnormal overlapping windows and count their number.

[0026] Furthermore, the abnormal propagation delay time is calculated and the total number of abnormal windows between the initial segment and the next-level transmission segment is counted. The specific logic is as follows:

[0027] Extract the start timestamps of all anomaly-associated windows in the next-level transmission segment, filter out the anomaly-associated window with the earliest time, and subtract the start timestamp of the earliest anomaly window from the start timestamp of the anomaly-associated window. The resulting time difference is the anomaly propagation delay time.

[0028] Perform a union analysis on the abnormal windows of the initial segment and the abnormal associated windows of the next-level transmission segment to determine the total number of abnormal windows.

[0029] Furthermore, the corresponding window correction coefficient is determined based on the anomaly propagation delay time, and the specific logic is as follows:

[0030] The delay scenario category is determined based on the abnormal propagation delay time, and a corresponding window correction coefficient is assigned to each delay scenario category, according to the following rules:

[0031] If the abnormal propagation delay time is less than or equal to the zero-delay threshold, the delay scenario category is determined to be zero-delay, and the assigned window correction coefficient is... ;

[0032] If the abnormal propagation delay time is less than the short-long delay boundary threshold, and the abnormal propagation delay time is greater than the no-delay threshold, the delay scenario is classified as short delay, and the assigned window correction coefficient is... ;

[0033] If the abnormal propagation delay time is greater than or equal to the short-long delay boundary threshold, the delay scenario is classified as long delay, and the assigned window correction coefficient is... ;

[0034] Among them, the no-delay threshold is less than the short-long delay boundary threshold, and .

[0035] Based on the above rules, the abnormal propagation delay time is matched and calculated to determine the delay scenario category to which it belongs, and then the corresponding window correction coefficient is obtained.

[0036] Furthermore, based on the number of abnormal overlapping windows and the number of windows with an abnormal initial segment state, the abnormal overlap ratio is calculated using the following formula:

[0037]

[0038] in, This indicates an abnormal overlap ratio. This represents the number of abnormally overlapping windows. This represents the number of windows in the initial segment that are in an abnormal state.

[0039] The total number of abnormal windows after correction is obtained by combining the abnormal overlap ratio and the window correction coefficient. The specific logic is as follows:

[0040]

[0041] in, This represents the total number of abnormal windows after correction. This represents the total number of abnormal windows. This is the window correction factor for the current delay scenario category.

[0042] To achieve the above objectives, the present invention also provides the following technical solution:

[0043] A deep learning-based IoT data transmission link diagnostic analysis system, the system being used to execute any of the deep learning-based IoT data transmission link diagnostic analysis methods described above, comprising:

[0044] The segmentation module is used to divide the data transmission link from the sensor to the gateway node in the IoT data transmission system into continuous multi-level transmission segments, and to determine the link diagnosis interval formed by multiple continuous non-overlapping windows based on the current link diagnosis time.

[0045] The state simulation module is used to collect scene environment data and multi-source data of all transmission segments in each window to construct the feature vector of each transmission segment in each window, and input it into the trained deep learning model to obtain the state of each transmission segment in each window, which is normal or abnormal.

[0046] The window statistics module is used to locate the window when the state of each transmission segment first becomes abnormal. The earliest transmission segment to become abnormal is taken as the initial segment and the earliest abnormal window is located. The multi-source data of the initial segment and the next level transmission segment are analyzed to determine the abnormal associated window of the next level transmission segment. The abnormal overlapping window is obtained by combining the window when the initial segment becomes abnormal. The abnormal propagation delay time is calculated and the total number of abnormal windows of the initial segment and the next level transmission segment is counted.

[0047] The time positioning module is used to determine the corresponding window correction coefficient based on the abnormal propagation delay time. Based on the number of abnormal overlapping windows and the number of windows when the initial segment is in an abnormal state, the abnormal overlap ratio is calculated. Using the total number of abnormal windows as the baseline value, the total number of abnormal windows is obtained after correction by combining the abnormal overlap ratio and the window correction coefficient, so as to lock the time range of the abnormality and complete the location of the abnormal point in the transmission link.

[0048] Compared with the prior art, the beneficial effects of the present invention are:

[0049] This invention provides a precise carrier for anomaly time-series identification by defining the data transmission link from the sensor to the gateway node in the Internet of Things data transmission system and determining the link diagnosis interval by multiple consecutive non-overlapping windows based on the current link diagnosis time. At the same time, by calculating the anomaly propagation delay time and designing a window correction coefficient, a complete dynamic diagnosis mechanism of "interval division - delay quantization - coefficient correction" is formed, which effectively makes up for the shortcomings of existing technologies that lack dynamic diagnosis in the time dimension, solves the problem of ambiguous anomaly time sequence boundary positioning, and greatly improves the time accuracy and timeliness of link anomaly diagnosis.

[0050] This invention also relies on the temporal correlation characteristics of windows to locate the window when each transmission segment first becomes abnormal. The earliest abnormal transmission segment is taken as the initial segment. Multi-source data of the initial segment and the next-level transmission segment are analyzed to determine the abnormal correlation window of the next-level transmission segment. Combined with the window when the initial segment becomes abnormal, abnormal overlapping windows are obtained. The total number of abnormal windows of the initial segment and the next-level transmission segment is counted. Through the collaborative calculation of the abnormal overlap ratio and the window correction coefficient, the total number of corrected abnormal windows is obtained. An abnormal propagation forward inference model of "temporal correlation-overlap quantification-coefficient correction" is constructed. This breaks through the bottleneck of existing technologies that cannot construct abnormal propagation inference models due to the lack of temporal correlation. It can accurately distinguish between the initial abnormal segment and the propagated abnormal segment, effectively solving the problem of inter-segment confusion in multi-segment synchronous abnormal and cross-segment propagation scenarios. It avoids misjudgment of source nodes from the root and significantly improves the accuracy and reproducibility of abnormal point location in complex IoT links. Attached Figure Description

[0051] Figure 1 This is a schematic diagram of the overall method flow of the present invention;

[0052] Figure 2 This is a block diagram of the module composition of the present invention. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0054] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0055] Example:

[0056] Please see Figure 1 The present invention provides a technical solution:

[0057] A deep learning-based diagnostic analysis method for IoT data transmission links includes the following steps:

[0058] S1. Divide the data transmission link from the sensor to the gateway node in the IoT data transmission system into continuous multi-level transmission segments, and determine the link diagnosis interval formed by multiple continuous non-overlapping windows based on the current link diagnosis time.

[0059] Based on the above embodiments, the data transmission link is divided into multiple consecutive transmission segments, with the specific logic as follows:

[0060] When dividing the data transmission link, the signal transfer nodes (such as routers, switches, signal repeaters) and device interfaces (such as sensor output interfaces, cable connectors, gateway input interfaces) in the data transmission link are used as the dividing points. According to the data flow from the sensor end to the gateway end, it is divided into the first-level transmission segment, the second-level transmission segment, ..., the nth-level transmission segment from front to back.

[0061] The first-level transmission segment is the transmission path from the sensor to the first boundary point, the second-level transmission segment is the transmission path from the first boundary point to the second boundary point, and so on. The nth-level transmission segment is the transmission path from the last boundary point to the gateway. Each level of transmission segment is an independent and continuous link unit.

[0062] The data flow refers to the unidirectional and continuous data flow of sensor-collected data in the Internet of Things (IoT) data transmission system, which starts from the sensor end, passes through signal transfer nodes and device interfaces, and is gradually transmitted to the gateway end. It is the core basis for dividing the transmission segment.

[0063] Based on the above embodiments, the link diagnosis interval is determined by multiple consecutive non-overlapping windows, and the specific logic is as follows:

[0064] Starting from the current link diagnosis time, trace back a preset time period to form a link diagnosis interval, and evenly divide it into multiple continuous non-overlapping windows.

[0065] The preset duration is adaptively set according to the IoT sensor reporting cycle, transmission link rate and diagnostic real-time requirements. The typical setting range is 30 seconds to 10 minutes, and the window duration is typically set to 1 second to 60 seconds.

[0066] S2. Collect scene environment data and multi-source data of all transmission segments in each window to construct the feature vector of each transmission segment in each window, and input it into the trained deep learning model to obtain the status of each transmission segment in each window, which is normal or abnormal.

[0067] Based on the above embodiments, the scene environment data includes the temperature, humidity, and electromagnetic interference intensity of the environment in which the transmission segment is located, and the specific definitions and acquisition methods are as follows:

[0068] The ambient temperature refers to the ambient air temperature around the deployment location of the transmission segment. It is obtained by continuously sampling by deploying digital temperature sensors near key nodes of each transmission segment, using a single diagnostic window as the statistical period, extracting all sampled data within the window time range and calculating the arithmetic mean, and using this average value as the temperature of the transmission segment under the current diagnostic window.

[0069] The ambient humidity refers to the relative humidity of the air in the area where the transmission segment is deployed. An industrial-grade humidity sensor is used for synchronous and continuous sampling. The statistical period is divided into individual diagnostic windows. All sampled data within the window time range are extracted and the arithmetic mean is calculated. This average value is used as the humidity of the transmission segment under the current diagnostic window.

[0070] The electromagnetic interference intensity is the combined electromagnetic radiation and noise intensity in the space where the transmission segment is located. It is obtained by the electromagnetic intensity acquisition module through directional continuous sampling of the transmission frequency band, with a single diagnostic window as the statistical period. All sampled data within the window are extracted and the effective mean is calculated as the electromagnetic interference intensity of the transmission segment under the current diagnostic window.

[0071] Based on the above embodiments, the multi-source data includes the link transmission rate, signal attenuation amplitude, packet loss rate, and bit error rate of the transmission segment, and the specific definitions and collection methods are as follows:

[0072] The link transmission rate refers to the total amount of effective data successfully transmitted by a single transmission segment per unit time. It characterizes the data throughput and transmission efficiency of the transmission segment and is a core indicator reflecting the smoothness of the link.

[0073] Data statistics modules are deployed at the start and end nodes of the transmission segment. The statistical period is based on a single diagnostic window. The total length of valid data packets within the statistical window time range is calculated, and the ratio of the total length of valid data packets to the diagnostic window duration is used as the link transmission rate of the transmission segment under the current diagnostic window.

[0074] The signal attenuation amplitude refers to the difference between the transmitted signal strength at the beginning of the transmission segment and the received signal strength at the end, reflecting the degree of energy loss of the signal during transmission. Excessive attenuation amplitude can easily cause signal recognition failure and transmission interruption.

[0075] A signal transmission detection module is deployed at the starting node of the transmission segment to record the signal transmission strength, and a signal reception detection module is deployed at the ending node to record the signal reception strength. According to the division of a single diagnostic window as the statistical period, all sampled data within the window are extracted and the average difference is calculated. The average difference is used as the signal attenuation amplitude of the transmission segment under the current diagnostic window.

[0076] The packet loss rate refers to the proportion of data packets lost in a single diagnostic window to the total number of packets sent, and is a key indicator for measuring the reliability of link transmission.

[0077] The total number of packets sent is counted at the starting node of the transmission segment, and the number of packets successfully received is counted at the ending node. The statistical period is based on a single diagnostic window. The ratio of the number of successfully received packets to the total number of packets sent is used as the packet loss rate of the transmission segment under the current diagnostic window.

[0078] The bit error rate refers to the proportion of erroneous bits in a transmission segment to the total number of transmitted bits within a single diagnostic window, reflecting the degree of signal distortion caused by interference and link loss.

[0079] A data verification module is deployed at the receiving node of the transmission segment. Cyclic redundancy check and parity check are used to perform bit-level verification on the received data. The statistical period is based on a single diagnostic window. The ratio of the number of erroneous bits in the window to the total number of transmitted bits is used as the bit error rate of the transmission segment under the current diagnostic window.

[0080] Based on the above, it should be noted that:

[0081] This invention collects temperature, humidity, electromagnetic interference intensity, link transmission rate, signal attenuation amplitude, packet loss rate, and bit error rate, providing reliable input for deep learning link diagnosis.

[0082] Among them, scenario environment data quantifies the impact of external interference on the link and distinguishes the transmission anomalies induced by the environment; multi-source data can directly reflect the real-time operating status and fault characteristics of the link. By integrating the two types of data to form a feature vector, the operating status of the transmission segment can be comprehensively characterized, reducing misjudgments and omissions caused by single-dimensional data, and effectively improving the model's classification accuracy and diagnostic reliability for normal and abnormal link states.

[0083] Based on the above embodiments, feature vectors for each transmission segment under each window are constructed as follows:

[0084]

[0085] in, For the first Level transmission segment in the first Feature vectors of each window The first Level transmission segment in the first The temperature, humidity, electromagnetic interference intensity, link transmission rate, signal attenuation, packet loss rate, and bit error rate of each window are monitored. This is an index of the transmission segment within the data transmission link. , This refers to the number of transmission segments within the data transmission link. This is the index of the window within the link diagnosis interval. , This represents the number of windows within the link diagnostic interval.

[0086] Based on the above embodiments, the deep learning model is constructed using a deep learning network based on a multilayer perceptron. The deep neural network of the multilayer perceptron includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer, and an output layer. The first hidden layer, the second hidden layer, and the third hidden layer each have at least two neurons and all use ReLU (linear rectified unit) as the activation function.

[0087] Multiple sets of historical samples of IoT data transmission links are collected in advance. Each set of samples corresponds to a transmission segment and a feature vector constructed from scene environment data and multi-source data within a window, as well as a labeled status label. The label is divided into normal and abnormal. A dataset is constructed based on the historical samples and divided into training set, validation set and test set in a ratio of 7:2:1. The training set is used to learn the model parameters; the validation set is used to adjust the hyperparameters during training to prevent overfitting; and the test set is used to evaluate the generalization ability of the model after training.

[0088] The structure of a deep learning network with a multilayer perceptron is as follows:

[0089] Input layer: Used to receive feature vectors constructed from scene environment data and multi-source data;

[0090] The first hidden layer has 128 neurons and uses ReLU as the activation function.

[0091] The second hidden layer has 64 neurons and also uses the ReLU activation function;

[0092] The third hidden layer has 32 neurons and uses the ReLU activation function;

[0093] Output layer: It has 1 neuron, which is used to output the state of each transmission segment in each window.

[0094] The process of training a deep learning model is as follows:

[0095] Supervised learning is employed to train the model. Feature vectors from multiple historical samples within a single window corresponding to a transmission segment are used as model input, and the state of each sample is used as the classification label output, thus constructing a supervised training task. A binary cross-entropy loss function is used to calculate the loss between the model's predictions and the true labels. The model parameters are iteratively updated using a backpropagation algorithm. When the training loss value is within a certain range... If the loss does not decrease significantly within the specified interval for 10 consecutive training rounds, the model is considered to have converged and training is stopped.

[0096] S3. Locate the window when the state of each transmission segment first becomes abnormal, take the earliest abnormal transmission segment as the initial segment and locate the earliest abnormal window, analyze the multi-source data of the initial segment and the next level transmission segment to determine the abnormal association window of the next level transmission segment, combine the window when the state of the initial segment becomes abnormal to obtain the abnormal overlapping window, calculate the abnormal propagation delay time and count the total number of abnormal windows of the initial segment and the next level transmission segment.

[0097] Based on the above embodiments, the window at which each transmission segment first becomes abnormal is located. The transmission segment that first becomes abnormal is taken as the initial segment, and the earliest abnormal window is located. The specific logic is as follows:

[0098] The window at which the first abnormal state occurs in each transmission segment is taken as its starting abnormal window. The time sequence of the starting abnormal windows of each transmission segment is compared, and the transmission segment with the earliest time is selected as the initial segment. The starting abnormal window of the initial segment is defined as the earliest abnormal window.

[0099] If two or more transmission segments have the same earliest start abnormal window time, the transmission segment closest to the sensor end will be determined as the initial segment.

[0100] Based on the above, it should be noted that:

[0101] Multi-level transmission segments are divided sequentially according to the unidirectional transmission path of data flow from the sensor end to the gateway end. Abnormal states of IoT data transmission links usually follow the propagation law of propagation from upstream to downstream and along the direction of data flow. Upstream transmission segment faults are more likely to propagate and spread downstream, thus causing multiple transmission segments to experience abnormalities simultaneously. Downstream transmission segment abnormalities usually do not have a reverse effect on upstream transmission segments.

[0102] When the initial anomaly window times of multiple transmission segments are tied for the earliest, and the initial segment cannot be uniquely determined by the time dimension, taking the upstream transmission segment closest to the sensor as the initial segment can follow the physical laws of anomaly propagation, ensure the uniqueness and determinism of the initial segment selection rule, avoid ambiguity in algorithm judgment, and at the same time fit the diagnostic logic of troubleshooting IoT transmission link faults from the source step by step, improving the accuracy and reliability of subsequent anomaly propagation analysis and root cause location.

[0103] Based on the above embodiments, the abnormal association window of the next-level transmission segment is determined, and combined with the window where the initial segment has an abnormal state, an abnormal overlap window is obtained. The specific logic is as follows:

[0104] The earliest anomaly window is used as the baseline window, and multi-source data within the baseline window of the initial segment are extracted to construct the baseline vector.

[0105] Extract multi-source data from each window of the next-level transmission segment to generate comparison vectors one by one. Calculate the cosine similarity between each comparison vector and the reference vector. Select windows with a cosine similarity greater than or equal to a preset similarity threshold that are not located before the reference window as abnormal association windows of the next-level transmission segment.

[0106] The abnormal windows in the initial segment refer to all windows that are determined to be abnormal by the deep learning model. The abnormal associated windows in the next-level transmission segment refer to windows that are highly matched with the baseline abnormal features of the initial segment after being filtered by cosine similarity. The intersection operation of the two sets of windows is performed to extract the common windows that exist in both sets of windows. These common windows are the abnormal overlapping windows. The total number of abnormal overlapping windows is counted to obtain the number of windows that are both abnormal.

[0107] Cosine similarity is a commonly used vector similarity metric in this field, used to quantify the correlation between two sets of vectors. It assesses similarity by calculating the cosine of the angle between two vectors in the vector space. The calculation rule is as follows: calculate the dot product and magnitude of each vector, and use the ratio of the dot product to the product of their magnitudes as the cosine similarity result. The value range of this index is... The closer the value is to 1, the higher the similarity of the vectors formed by the two sets of multi-source data.

[0108] The preset similarity threshold is determined through historical data statistics. First, based on historical samples of link anomalies, the cosine similarity benchmark value between the benchmark vector and the vector to be compared under different anomaly types is statistically obtained. Then, based on this benchmark value, it is floated upward by 20% as the preset similarity threshold.

[0109] The anomaly types are classified according to the fault manifestations and performance degradation forms of IoT transmission links. Combined with the multi-source data and scenario environment data collected in this solution, they specifically include signal attenuation anomalies, transmission rate fluctuation anomalies, data packet loss anomalies, data bit error anomalies, and composite transmission anomalies induced by environmental factors such as temperature, humidity, and electromagnetic interference.

[0110] Based on the above embodiments, the abnormal propagation delay time is calculated and the total number of abnormal windows between the initial segment and the next level transmission segment is counted. The specific logic is as follows:

[0111] Extract the start timestamps of all anomaly-associated windows in the next-level transmission segment, filter out the anomaly-associated window with the earliest time, and subtract the start timestamp of the earliest anomaly window from the start timestamp of the anomaly-associated window. The resulting time difference is the anomaly propagation delay time.

[0112] Perform a union operation on the abnormal windows of the initial segment and the abnormal associated windows of the next-level transmission segment, merge the two sets of windows and remove duplicate window numbers to obtain a total window set covering all common abnormal periods of the two transmission segments; count the number of windows in this total window set, and the result is the total number of abnormal windows of the initial segment and the next-level transmission segment.

[0113] Based on the above, it should be noted that:

[0114] The above steps can accurately determine the initial segment of the anomaly and the earliest anomaly window. Relying on the cosine similarity matching of feature vectors from multiple sources, as well as set operations of intersection and union, the analysis and statistics of anomaly association windows and anomaly overlapping windows are completed. At the same time, by calculating the anomaly propagation delay time and counting the total number of anomaly windows in the initial segment and the next level of transmission segment, the quantitative assessment of the anomaly propagation lag and the scope of anomaly impact can be achieved. This can effectively distinguish between anomalies of the same source propagation type and independent sporadic anomalies, providing data support for subsequent anomaly propagation relationship determination and fault root cause tracing.

[0115] S4. Determine the corresponding window correction coefficient based on the abnormal propagation delay time. Calculate the abnormal overlap ratio based on the number of abnormal overlapping windows and the number of windows when the initial segment is in an abnormal state. Using the total number of abnormal windows as a baseline, combine the abnormal overlap ratio and the window correction coefficient to obtain the corrected total number of abnormal windows, thereby locking in the time range of the abnormality and completing the location of the abnormal point in the transmission link.

[0116] Based on the above embodiments, the corresponding window correction coefficient is determined according to the abnormal propagation delay time, and the specific logic is as follows:

[0117] The delay scenario category is determined based on the abnormal propagation delay time, and a corresponding window correction coefficient is assigned to each delay scenario category, according to the following rules:

[0118] If the abnormal propagation delay time is less than or equal to the zero-delay threshold, the delay scenario category is determined to be zero-delay, and the assigned window correction coefficient is... ;

[0119] If the abnormal propagation delay time is less than the short-long delay boundary threshold, and the abnormal propagation delay time is greater than the no-delay threshold, the delay scenario is classified as short delay, and the assigned window correction coefficient is... ;

[0120] If the abnormal propagation delay time is greater than or equal to the short-long delay boundary threshold, the delay scenario is classified as long delay, and the assigned window correction coefficient is... ;

[0121] Among them, the no-delay threshold is less than the short-long delay boundary threshold, and .

[0122] Based on the above rules, the abnormal propagation delay time is matched and calculated to determine the delay scenario category to which it belongs, and then the corresponding window correction coefficient is obtained.

[0123] Based on the above, the no-delay threshold and the short-long delay boundary threshold are determined through statistical analysis of historical abnormal samples of IoT data transmission links:

[0124] A large amount of abnormal propagation delay time data in historical abnormal samples were sorted and processed, and the lower quartile and upper quartile of the delay time data were calculated respectively. Based on the above statistical results, the calculated lower quartile was determined as the no-delay threshold, and the calculated upper quartile was determined as the short-long delay boundary threshold.

[0125] Based on the above, it should be noted that:

[0126] In a zero-latency scenario, the anomaly propagation delay is less than or equal to the zero-latency threshold. The time it takes for an anomaly to propagate from the initial segment to the next transmission segment is extremely short. It is assumed that anomalies in both transmission segments occur almost synchronously. The anomaly propagation process does not introduce a significant lag bias into the statistical results of the anomaly window, thus eliminating the need to amplify or shrink the total number of anomaly windows. Therefore, the window correction coefficient corresponding to the zero-latency scenario is... It is set to 1.0, which serves as a reference standard for assigning window correction coefficients and quantity corrections in short-latency and long-latency scenarios.

[0127] The longer the anomaly propagation delay, the more significant the lag in the propagation of the anomaly from the initial segment to the next level of transmission. The greater the offset of the anomaly association window in the next level of transmission relative to the earliest anomaly window in the initial segment, the more significant the lag. The total number of anomaly windows obtained solely through the original union operation cannot fully cover the anomaly time period extended by the propagation lag, leading to a smaller anomaly time range and incomplete anomaly location. To compensate for the statistical bias caused by the propagation delay and ensure that the corrected total number of anomaly windows fully covers the actual anomaly wave and time period, the original statistical results need to be positively amplified and corrected. Furthermore, the longer the delay, the more pronounced the statistical gaps caused by the lag offset, and the greater the required compensation. Therefore, the window correction coefficient for short-delay scenarios... Window correction coefficient for long latency scenarios All are set to values ​​greater than 1.0, and satisfy the following conditions: .

[0128] Based on the correlation between anomaly propagation delay and missing window statistics, in short-latency scenarios of IoT data transmission links, the time it takes for anomalies to propagate from the initial segment to the next level of transmission is relatively short, resulting in only a small number of missed anomaly windows. Verification based on measured data and historical fault statistics of IoT data transmission links shows that the missing anomaly window statistics in such scenarios are approximately 10%-30%. To compensate for this statistical gap and fully cover the actual anomaly occurrence period, the total number of original anomaly windows needs to be increased by 1.1-1.3 times. Therefore, the window correction coefficient corresponding to short-latency scenarios can be... The value is 1.2.

[0129] In IoT data transmission links, the time taken for anomaly propagation increases significantly in long-latency scenarios, with a noticeable lag effect. This further widens the statistical gap in anomaly window numbers. Verification based on measured data and historical fault statistics of IoT data transmission links shows that the statistical gap in such scenarios can reach 40%-60%. To fully compensate for the statistical bias caused by the lag and ensure that the corrected results fully match the actual anomaly impact range, the total number of original anomaly windows needs to be increased by 1.4-1.6 times. Therefore, the window correction coefficient corresponding to long-latency scenarios can be... The value is 1.5.

[0130] Based on the above embodiments, the abnormal overlap ratio is calculated based on the number of abnormal overlapping windows and the number of windows with an abnormal initial segment state, using the following formula:

[0131]

[0132] in, The abnormal overlap ratio is used to quantitatively evaluate the co-origin correlation and synchronous transmission degree of the anomalies between the initial segment and the next-level transmission segment by combining two index parameters: the number of abnormal overlap windows and the number of windows in the initial segment with an abnormal state. The larger the abnormal overlap ratio, the higher the proportion of the abnormal window of the initial segment and the next-level transmission segment are synchronously overlapped, and the stronger the correlation between the anomalies of the two levels of transmission segments originating from the same initial fault.

[0133] In the formula, This represents the number of abnormally overlapping windows. This represents the number of windows in which the initial segment is in an abnormal state within the link diagnostic interval;

[0134] Based on the above, it should be noted that:

[0135] If the number of abnormal overlapping windows increases, while the number of windows with abnormal status in the initial segment remains unchanged, the proportion of synchronous overlap between the abnormality in the initial segment and the abnormality in the next-level transmission segment increases accordingly. The correlation between abnormalities in the two-level transmission segments and the degree of synchronous transmission are enhanced, thus leading to an increase in the abnormal overlap ratio. Therefore, the number of abnormal overlapping windows and the abnormal overlap ratio are positively correlated. Thus, the number of abnormal overlapping windows is used as the numerator of the formula, so that its positive change can directly drive the abnormal overlap ratio to change in the same direction.

[0136] If the number of windows with an abnormal state in the initial segment increases, the proportion of synchronous overlap between the abnormality in the initial segment and the abnormality in the next-level transmission segment decreases accordingly, provided that the number of abnormal overlapping windows remains unchanged. The correlation between abnormalities in the two-level transmission segments and the degree of synchronous transmission weaken, thus reducing the abnormal overlap ratio. Therefore, the number of windows with an abnormal state in the initial segment is negatively correlated with the abnormal overlap ratio. Thus, the number of windows with an abnormal state in the initial segment is used as the denominator of the formula, so that its change can adjust the abnormal overlap ratio in the opposite direction.

[0137] Therefore, the above ratio is used to express the functional relationship between the abnormal overlap ratio, the number of abnormal overlap windows, and the number of windows with an abnormal initial segment state.

[0138] Based on the above embodiments, the total number of abnormal windows after correction is obtained by combining the abnormal overlap ratio and the window correction coefficient. The specific logic is as follows:

[0139]

[0140] in, This represents the total number of abnormal windows after correction. This represents the total number of abnormal windows. This is the window correction factor for the current delay scenario category.

[0141] Based on the above, it should be noted that:

[0142] Total number of abnormal windows To correct the calculated baseline value, which reflects the total number of uncompensated initial outlier windows, all subsequent corrections are based on this baseline value;

[0143] Under this premise, the abnormal overlap ratio It is mainly used to determine the homology and conduction attribution relationship between anomalies in the initial segment and the next level transmission segment. The higher the value, the greater the degree of synchronization overlap between the two-stage transmission segment anomalies, and the more it confirms that the next-stage anomaly is formed by the propagation of the initial segment, rather than an anomaly generated independently;

[0144] Since this scheme corrects the total number of abnormal windows, its core purpose is to compensate for the statistical loss of abnormal windows caused by the lag in the transmission of abnormalities during the hierarchical transmission process. Therefore, the correction operation is directly related to the degree of correlation between abnormal transmission and the correction operation.

[0145] Specifically, such statistical omissions are only necessary to compensate for when the next-level anomaly is indeed propagated from the initial segment; and The higher the value, the stronger the correlation between anomalies. The corresponding statistical missing value compensation should be increased accordingly. The increase in compensation will directly increase the total number of corrected anomaly windows. Therefore, the anomaly overlap ratio is positively correlated with the total number of corrected anomaly windows.

[0146] Based on the aforementioned positive correlation, a higher proportion of abnormal overlap indicates a stronger transmission correlation between the two transmission stages, and the statistical gaps caused by transmission lag have greater real compensatory significance, requiring a higher correction weighting. Therefore, the proportion of abnormal overlap needs to be included in the correction calculation as a weighting coefficient.

[0147] Meanwhile, window correction factor Used to quantify the degree of lag in the transmission of anomalies and to match the corresponding statistical missing compensation magnitude;

[0148] Based on the aforementioned classification rules for delayed scenarios, the longer the delay time of anomaly propagation, the more significant the lag effect corresponding to the delayed scenario, the more anomaly windows are missed in the original statistics, and the greater the statistical missing range.

[0149] Based on this, the window correction factor The assigned values ​​are positively correlated with the degree of transmission lag and the magnitude of statistical missing values. A larger value indicates a greater degree of compensation required for the baseline value. A higher compensation level, assuming the baseline value and weighting coefficients remain unchanged, will result in a larger total number of corrected outlier windows. Therefore, the window correction coefficient... It is positively correlated with the total number of corrected abnormal windows.

[0150] Due to abnormal overlap ratio With window correction factor Both are dimensionless coefficients that positively influence the correction results, and their effects are independent of each other, with no overlap in their compensation logic. Responsible for assigning corrective weights based on the reliability of homology transmission. Responsible for determining the compensation ratio based on the degree of delay or deficiency.

[0151] Combining the two in a multiplicative manner allows for the linear superposition of dual positive effects, forming a comprehensive weighted compensation coefficient that jointly reflects the reliability of abnormal transmission and the strength of missing compensation.

[0152] Finally, by multiplying the comprehensive weighted compensation coefficient by the benchmark value, the credibility weighting and lag missing compensation can be completed simultaneously based on the total number of abnormal windows, ultimately obtaining the corrected total number of abnormal windows that fits the true range of abnormalities.

[0153] Based on the above embodiments, the time range of the anomaly is locked to locate the point of anomaly in the transmission link. The specific logic is as follows:

[0154] The link diagnosis interval consists of continuous, non-overlapping fixed-duration windows. The number of abnormal windows directly corresponds to the duration of the abnormality. The corrected total number of abnormal windows has compensated for the statistical bias of transmission lag and incorporated the reliability of abnormality propagation from the same source, thus accurately reflecting the complete period of abnormality propagation. Based on the corrected total number of abnormal windows, combined with the window's temporal position and the duration of each window, the start and end times of the abnormality can be determined, defining the complete abnormal time range. Furthermore, by combining the temporal sequence of the earliest abnormal window in the initial segment, the overlapping windows, and the associated windows, as well as the transmission segment attribution, the initial occurrence point and propagation path of the abnormality can be traced, ultimately achieving precise location of the abnormal point in the transmission link.

[0155] Based on the above, it should be noted that:

[0156] By quantifying and correcting the abnormal propagation delay, a dynamic diagnostic mechanism of "interval division - delay quantization - coefficient correction" is constructed to make up for the lack of dynamic diagnosis in the time dimension of existing technologies and improve the accuracy and timeliness of abnormal timing location. At the same time, relying on the collaborative calculation of the abnormal overlap ratio and correction coefficient, an abnormal propagation forward inference model is established to achieve accurate differentiation between the initial abnormal segment and the propagated abnormal segment, solve the problem of inter-segment confusion and source misjudgment in cross-segment propagation and multi-segment synchronous anomalies, and improve the accuracy and reproducibility of abnormal point location in complex IoT links.

[0157] Please see Figure 2 The present invention also provides a technical solution:

[0158] A deep learning-based IoT data transmission link diagnostic analysis system, the system being used to execute any of the deep learning-based IoT data transmission link diagnostic analysis methods described above, comprising:

[0159] The segmentation module is used to divide the data transmission link from the sensor to the gateway node in the IoT data transmission system into continuous multi-level transmission segments, and to determine the link diagnosis interval formed by multiple continuous non-overlapping windows based on the current link diagnosis time.

[0160] The state simulation module is used to collect scene environment data and multi-source data of all transmission segments in each window to construct the feature vector of each transmission segment in each window, and input it into the trained deep learning model to obtain the state of each transmission segment in each window, which is normal or abnormal.

[0161] The window statistics module is used to locate the window when the state of each transmission segment first becomes abnormal. The earliest transmission segment to become abnormal is taken as the initial segment and the earliest abnormal window is located. The multi-source data of the initial segment and the next level transmission segment are analyzed to determine the abnormal associated window of the next level transmission segment. The abnormal overlapping window is obtained by combining the window when the initial segment becomes abnormal. The abnormal propagation delay time is calculated and the total number of abnormal windows of the initial segment and the next level transmission segment is counted.

[0162] The time positioning module is used to determine the corresponding window correction coefficient based on the abnormal propagation delay time. Based on the number of abnormal overlapping windows and the number of windows when the initial segment is in an abnormal state, the abnormal overlap ratio is calculated. Using the total number of abnormal windows as the baseline value, the total number of abnormal windows is obtained after correction by combining the abnormal overlap ratio and the window correction coefficient, so as to lock the time range of the abnormality and complete the location of the abnormal point in the transmission link.

[0163] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0164] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by software, electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0165] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0166] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A deep learning based diagnosis analysis method for an Internet of Things data transmission link, characterized in that, The specific steps include: S1. Divide the data transmission link from the sensor to the gateway node in the IoT data transmission system into continuous multi-level transmission segments, and determine the link diagnosis interval formed by multiple continuous non-overlapping windows based on the current link diagnosis time. S2. Collect scene environment data and multi-source data of all transmission segments in each window to construct the feature vector of each transmission segment in each window, and input it into the trained deep learning model to obtain the status of each transmission segment in each window, which is normal or abnormal. S3. Locate the window when the state of each transmission segment first becomes abnormal, take the earliest abnormal transmission segment as the initial segment and locate the earliest abnormal window, analyze the multi-source data of the initial segment and the next level transmission segment to determine the abnormal association window of the next level transmission segment, combine the window when the state of the initial segment becomes abnormal to obtain the abnormal overlapping window, calculate the abnormal propagation delay time and count the total number of abnormal windows of the initial segment and the next level transmission segment. S4. Determine the corresponding window correction coefficient based on the abnormal propagation delay time. Calculate the abnormal overlap ratio based on the number of abnormal overlapping windows and the number of windows when the initial segment is in an abnormal state. Using the total number of abnormal windows as a baseline, combine the abnormal overlap ratio and the window correction coefficient to obtain the corrected total number of abnormal windows, thereby locking in the time range of the abnormality and completing the location of the abnormal point in the transmission link.

2. The deep learning based Internet of Things data transmission link diagnosis analysis method according to claim 1, characterized in that, The scene environment data includes the temperature, humidity, and electromagnetic interference intensity of the environment in which the transmission segment is located; the multi-source data includes the link transmission rate, signal attenuation amplitude, packet loss rate, and bit error rate of the transmission segment.

3. The IoT data transmission link diagnostic analysis method based on deep learning according to claim 1, characterized in that, The data transmission link is divided into multiple consecutive transmission segments, and the specific logic is as follows: Using the signal transfer nodes and device interfaces in the data transmission link as the dividing points, the data flow is divided into the first-level transmission segment, the second-level transmission segment, ..., the nth-level transmission segment, according to the transmission path from the sensor end to the gateway end. The first-level transmission segment is the transmission path from the sensor to the first boundary point, the second-level transmission segment is the transmission path from the first boundary point to the second boundary point, and so on, with the nth-level transmission segment being the transmission path from the last boundary point to the gateway. The link diagnostic interval is determined by multiple consecutive, non-overlapping windows, and the specific logic is as follows: Starting from the current link diagnosis time, trace back a preset time period to form a link diagnosis interval, and evenly divide it into multiple continuous non-overlapping windows.

4. The IoT data transmission link diagnostic analysis method based on deep learning according to claim 1, characterized in that, The window at which the first abnormal state occurs in each transmission segment is located. The earliest transmission segment with an abnormal state is used as the initial segment, and the earliest abnormal window is located. The specific logic is as follows: The window at which the first abnormal state occurs in each transmission segment is taken as its starting abnormal window. The time sequence of the starting abnormal windows of each transmission segment is compared, and the transmission segment with the earliest time is selected as the initial segment. The starting abnormal window of the initial segment is defined as the earliest abnormal window. If two or more transmission segments have the same earliest start abnormal window time, the transmission segment closest to the sensor end will be determined as the initial segment.

5. The IoT data transmission link diagnostic analysis method based on deep learning according to claim 2, characterized in that, The abnormal association window for the next level transmission segment is determined, and combined with the windows where the initial segment has an abnormal status, an abnormal overlap window is obtained. The specific logic is as follows: The earliest anomaly window is used as the baseline window, and multi-source data within the baseline window of the initial segment are extracted to construct the baseline vector. Extract multi-source data from each window of the next-level transmission segment to generate comparison vectors one by one. Calculate the cosine similarity between each comparison vector and the reference vector. Select windows with a cosine similarity greater than or equal to a preset similarity threshold that are not located before the reference window as abnormal association windows of the next-level transmission segment. An intersection analysis is performed on the abnormal windows of the initial segment and the abnormal associated windows of the next-level transmission segment to identify the abnormal overlapping windows and count their number.

6. The IoT data transmission link diagnostic analysis method based on deep learning according to claim 5, characterized in that, The abnormal propagation delay time is calculated, and the total number of abnormal windows between the initial segment and the next level transmission segment is counted. The specific logic is as follows: Extract the start timestamps of all anomaly-associated windows in the next-level transmission segment, filter out the anomaly-associated window with the earliest time, and subtract the start timestamp of the earliest anomaly window from the start timestamp of the anomaly-associated window. The resulting time difference is the anomaly propagation delay time. Perform a union analysis on the abnormal windows of the initial segment and the abnormal associated windows of the next-level transmission segment to determine the total number of abnormal windows.

7. The IoT data transmission link diagnostic analysis method based on deep learning according to claim 1, characterized in that, The window correction coefficient is determined based on the anomaly propagation delay time, and the specific logic is as follows: The delay scenario category is determined based on the abnormal propagation delay time, and a corresponding window correction coefficient is assigned to each delay scenario category, according to the following rules: If the abnormal propagation delay time is less than or equal to the zero-delay threshold, the delay scenario category is determined to be zero-delay, and the assigned window correction coefficient is... ; If the abnormal propagation delay time is less than the short-long delay boundary threshold, and the abnormal propagation delay time is greater than the no-delay threshold, the delay scenario is classified as short delay, and the assigned window correction coefficient is... ; If the abnormal propagation delay time is greater than or equal to the short-long delay boundary threshold, the delay scenario is classified as long delay, and the assigned window correction coefficient is... ; Among them, the no-delay threshold is less than the short-long delay boundary threshold, and ; Based on the above rules, the abnormal propagation delay time is matched and calculated to determine the delay scenario category to which it belongs, and then the corresponding window correction coefficient is obtained.

8. The IoT data transmission link diagnostic analysis method based on deep learning according to claim 7, characterized in that, The abnormal overlap ratio is calculated based on the number of abnormal overlapping windows and the number of windows with an abnormal initial segment state, using the following formula: in, This indicates an abnormal overlap ratio. This represents the number of abnormally overlapping windows. This represents the number of windows in the initial segment that are in an abnormal state. The total number of abnormal windows after correction is obtained by combining the abnormal overlap ratio and the window correction coefficient. The specific logic is as follows: in, This represents the total number of abnormal windows after correction. This represents the total number of abnormal windows. This is the window correction factor for the current delay scenario category.

9. A deep learning-based IoT data transmission link diagnostic analysis system, the system being used to execute the deep learning-based IoT data transmission link diagnostic analysis method according to any one of claims 1-8, characterized in that, include: The segmentation module is used to divide the data transmission link from the sensor to the gateway node in the IoT data transmission system into continuous multi-level transmission segments, and to determine the link diagnosis interval formed by multiple continuous non-overlapping windows based on the current link diagnosis time. The state simulation module is used to collect scene environment data and multi-source data of all transmission segments in each window to construct the feature vector of each transmission segment in each window, and input it into the trained deep learning model to obtain the state of each transmission segment in each window, which is normal or abnormal. The window statistics module is used to locate the window when the state of each transmission segment first becomes abnormal. The earliest transmission segment to become abnormal is taken as the initial segment and the earliest abnormal window is located. The multi-source data of the initial segment and the next level transmission segment are analyzed to determine the abnormal associated window of the next level transmission segment. The abnormal overlapping window is obtained by combining the window when the initial segment becomes abnormal. The abnormal propagation delay time is calculated and the total number of abnormal windows of the initial segment and the next level transmission segment is counted. The time positioning module is used to determine the corresponding window correction coefficient based on the abnormal propagation delay time. Based on the number of abnormal overlapping windows and the number of windows when the initial segment is in an abnormal state, the abnormal overlap ratio is calculated. Using the total number of abnormal windows as the baseline value, the total number of abnormal windows is obtained after correction by combining the abnormal overlap ratio and the window correction coefficient, so as to lock the time range of the abnormality and complete the location of the abnormal point in the transmission link.

Images