Sensing node communication system and method with self-healing mechanism

Abstract

The invention discloses a sensing node communication system and method with a self-healing mechanism. The sensing node communication system comprises a node detection module, a node communication module, a preprocessing module, a mixed training module and a self-healing control module. The system realizes periodic acquisition of multi-dimensional environmental parameters by configuring a plurality of sensing nodes, shares state information based on a self-organizing network, and constructs a local network health state map. The preprocessed data is used for training a hybrid neural network, and long-term dependency features and local abnormal fluctuation features of node states are extracted from the data to assess fault risks. When the risk score exceeds a threshold value, the system automatically executes self-healing operations such as dynamic route reconstruction and standby link activation. According to the invention, the stability and the adaptive capability of the underwater sensor network are improved, and the method has wide application value.

Description

Technical Field

[0001] The present invention relates to the technical field of the Internet of Things, and in particular to a sensing node communication system and method with a self-healing mechanism. Background Art

[0002] With the rapid development of the Internet of Things, edge computing, and intelligent sensing technologies, distributed sensing systems have been widely used in fields such as environmental monitoring, industrial control, smart facilities, and disaster warning. Such systems usually consist of a large number of sensing nodes, which cooperate through a wireless communication network to collect and transmit multi-source environmental data, realizing continuous sensing and status management of the target area.

[0003] In practical applications, especially in some complex or extreme environments, such as high-humidity, high-pressure, highly corrosive atmospheric or underwater environments, the deployment and communication of sensing nodes often face severe challenges. Sensing networks in these environments not only need to have strong structural adaptability and energy consumption control capabilities, but also put forward higher requirements for the stability, communication reliability, and fault recovery capabilities of the network.

[0004] However, existing sensing communication systems mostly adopt static routing structures and centralized control architectures. When nodes fail or communication links are abnormal, it is often difficult to identify and respond dynamically in a timely manner, easily leading to network paralysis, data loss, or partial system failure. In addition, most current systems lack the ability to analyze and predict node status and environmental data in real time, cannot give effective early warnings before faults occur, and lack control mechanisms with adaptive and self-healing characteristics.

[0005] Therefore, there is an urgent need for a sensing node communication system with fault prediction capabilities and self-healing control logic, which can operate stably continuously in various complex environments, significantly improve the adaptability and overall robustness of the network, and meet the actual needs of high-reliability communication and intelligent fault tolerance. Summary of the Invention

[0006] Aiming at the deficiencies of the existing technology, the purpose of the present invention is to provide a sensing node communication system and method with a self-healing mechanism, which are used to improve the stability, adaptability, and fault response efficiency of underwater sensing networks.

[0007] To achieve the above purpose, the present invention provides the following technical solutions: A sensing node communication system with a self-healing mechanism, comprising: A node detection module, which is used to configure a plurality of sensing nodes, each of the sensing nodes includes a plurality of sensors, and each of the sensors is used to collect multi-dimensional detection data according to a preset period; A node communication module, connected to the node detection module, is used to form an ad-hoc multi-hop network to establish communication connections for each of the sensor nodes. The detection data detected by each of the sensor nodes forms a local network health status map through a status sharing protocol. The local network health status map includes a plurality of node status data, and the node status data includes node self-check data, physical environment parameters, and chemical environment parameters; A preprocessing module, connected to the node communication module, is used to perform preprocessing and normalization processing on the node self-check data, the physical environment parameters, and the chemical environment parameters in sequence to construct a training data set; A hybrid training module, connected to the preprocessing module, is used to train according to the training data set by using a hybrid neural network. The hybrid neural network includes a time series modeling layer, a local feature extraction layer, and a fully connected layer. The time series modeling layer uses a long short-term memory network to capture the long-term dependence features of each node status in the training data set. The local feature extraction layer uses a convolutional neural network layer to extract local abnormal fluctuation features within a time window in the training data set. The fully connected layer fuses the long-term dependence features and the local abnormal fluctuation features to output a fault risk score; A self-healing control module, connected to the hybrid training module and the node communication module, is used to automatically trigger self-healing response measures according to the node status data at adjacent sensor nodes when the fault risk score at the current sensor node is greater than a preset self-healing risk threshold. The self-healing response measures include dynamic routing reconstruction, standby link activation, and data load reallocation.

[0008] Further, the preprocessing module includes: A time series division unit, used to record the node self-check data, the physical environment parameters, and the chemical environment parameters according to a time series, and perform data segmentation using a sliding window technique to form continuous time series data samples; A preprocessing unit, connected to the time series division unit, is used to perform data cleaning, outlier detection, and missing value filling on the time series data samples in sequence to obtain preprocessed data samples; A normalization unit, connected to the preprocessing unit, is used to perform normalization processing on multi-dimensional data in the preprocessed data samples according to a preset normalization method to form the training data set.

[0009] Further, the node self-check data includes node voltage, node current, node energy consumption status, signal-to-noise ratio, packet loss rate, and signal delay; The physical environment parameters include temperature, ambient pressure, salinity, dissolved oxygen concentration, and ambient turbidity; The chemical environment parameters include pH value and specific ion concentration.

[0010] Furthermore, it further includes: A fault detection module, connected to the node communication module, for performing fault detection on the status data of each node and generating corresponding fault detection results, and processing the fault detection results to obtain actual fault scores; An incremental training module, connected to the fault detection module and the hybrid training module, for processing the actual fault score and the fault risk score to obtain a predicted deflection value, and dynamically adjusting and then training the weight parameters of the hybrid neural network based on the predicted deviation value to perform incremental training and update on the hybrid neural network.

[0011] Furthermore, the functional expression of the hybrid neural network includes: ; ; ; ; ; ; wherein, represents the fault risk score, represents the time window length, represents the time variable, represents the Gamma weight function constructed by the current voltage and temperature of the node, represents the current voltage of the node, represents the standard deviation of the temperature, represents the mean value of the temperature, represents the current perturbation modulation term, represents the composite excitation factor formed by the product of the current and the packet loss rate, represents the sensitivity adjustment factor of the environmental pressure change, represents the historical current mean value, represents the change range of the environmental turbidity, represents the change amount of the signal-to-noise ratio, represents the communication and activation modeling function, represents the first kind of first-order Bessel function to measure the amplitude of environmental perturbation, represents the second kind of zero-order Bessel function to represent the strength of node redundant connectivity, represents the salinity perturbation frequency factor, represents the complexity parameter of the current network communication topology, represents the rate of change of the node energy consumption, Represents the time window width of the described pH value, Represents the delay attenuation and ion anomaly suppression function, Represents the maximum delay of the current signal, Represents the historical average signal delay, Represents the data channel congestion factor, Represents the time-dependent adjustment coefficient, Represents the mean value of the specific ion concentration, Represents the standard deviation of the specific ion concentration, Represents the self-healing trigger suppression function, Represents the self-healing response time activation coefficient, Represents the state consistency measurement factor between neighboring nodes, Represents the natural logarithm function, Represents a quantitative index of the complexity of ion species in the current environment.

[0012] Furthermore, the node state data further includes underwater acoustic interference data and underwater optical interference data. The underwater acoustic interference data includes power spectral density, and the underwater optical interference data includes optical image blur, color deviation drift, and image noise index.

[0013] Furthermore, it further includes a function optimization module, connected to the node communication module and the hybrid training module, for optimizing the communication and activation modeling function according to the power spectral density, the optical image blur, and the image noise index, and optimizing the self-healing trigger suppression function according to the image blur and the color deviation drift, and then updating the fault risk score according to the optimized communication and activation modeling function and the optimized self-healing trigger suppression function.

[0014] Furthermore, the optimized communication and activation modeling function is configured as: ; where Represents the optimized communication and activation modeling function, Represents the normalized power spectral density value, Represents the image noise index, Represents the blur index adjustment factor, Represents the Bessel perturbation frequency term, used to simulate the periodic excitation of PSD in underwater acoustic propagation; The optimized self-healing trigger suppression function is configured as: ; Where Represents the optimized self-healing trigger suppression function, Represents the image blur index, represents the fuzzy exponent amplification coefficient, which is used to control the interference intensity of activation, represents the color deviation drift difference, respectively represent the color difference fluctuation coefficients of the G and R channels; Then the function expression of the updated fault risk score is configured as: ; wherein, represents the updated fault risk score.

[0015] A communication method for a sensing node with a self-healing mechanism, which is applied to the above-mentioned communication system for a sensing node with a self-healing mechanism, includes: Step S1, the node detection module configures multiple sensing nodes, each of the sensing nodes includes multiple sensors, and each of the sensors is used to collect multi-dimensional detection data according to a preset period; Step S2, the node communication module forms a self-organizing multi-hop network to establish a communication connection for each of the sensing nodes, and the detection data detected by each of the sensing nodes forms a local network health status map through a status sharing protocol. The local network health status map includes multiple node status data, and the node status data includes node self-check data, physical environment parameters, and chemical environment parameters; Step S3, the preprocessing module preprocesses and normalizes the node self-check data, the physical environment parameters, and the chemical environment parameters in sequence to construct a training data set; Step S4, the hybrid training module trains according to the training data set by using a hybrid neural network. The hybrid neural network includes a time series modeling layer, a local feature extraction layer, and a fully connected layer. The time series modeling layer uses a long short-term memory network to capture the long-term dependence features of each node status in the training data set. The local feature extraction layer uses a convolutional neural network layer to extract the local abnormal fluctuation features within the time window in the training data set. The fully connected layer fuses the long-term dependence features and the local abnormal fluctuation features to output a fault risk score; Step S5, when the fault risk score at the current sensing node is greater than a preset self-healing risk threshold, the self-healing control module automatically triggers self-healing response measures according to the node status data at the adjacent sensing nodes. The self-healing response measures include dynamic routing reconstruction, standby link activation, and data load redistribution.

[0016] Advantages of the present invention: The present invention collects key indicators such as node voltage, current, energy consumption status, signal-to-noise ratio, packet loss rate, and signal delay, realizes node-level self-diagnosis and health assessment, and significantly enhances the timeliness and accuracy of fault identification; By integrating physical and chemical parameters such as temperature, ambient pressure, salinity, dissolved oxygen, pH value, turbidity, and specific ion concentration, the present invention can comprehensively monitor the changes in the underwater environment; The present invention also adopts a hybrid neural network architecture that combines LSTM and CNN, which can not only capture the time dependence in the sensing data, but also identify short-term local fluctuation anomalies, significantly improving the depth and robustness of fault prediction; After identifying high-risk nodes, the present invention can automatically trigger multi-level response strategies including dynamic routing reconstruction, standby link activation, and data load redistribution based on the states of neighboring nodes, effectively preventing network paralysis or data interruption; Facing the complex and changeable physical and chemical environment underwater, the present invention has the ability of real-time adaptation and reconstruction, significantly improving the stability and fault tolerance of the sensing network in underwater operations. Brief Description of the Drawings

[0017] Figure 1 It is a schematic structural diagram of the sensing node communication system in the present invention.

[0018] Reference Signs: 1, node detection module; 2, node communication module; 3, preprocessing module; 31, time series division unit; 32, preprocessing unit; 33, normalization unit; 4, hybrid training module; 5, self-healing control module; 6, fault detection module; 7, incremental training module; 8, function optimization module. Detailed Description of the Embodiments

[0019] The present invention will be further described in detail below with reference to the drawings and embodiments. The same components are denoted by the same reference signs. It should be noted that the terms "front", "rear", "left", "right", "upper" and "lower" used in the following description refer to the directions in the drawings, and the terms "bottom surface" and "top surface", "inner" and "outer" refer to the directions towards or away from the geometric center of a specific component respectively.

[0020] Example 1, referring to Figure 1 , which is the first embodiment of the present invention. This embodiment provides a sensing node communication system with a self-healing mechanism, which can improve the stability, adaptive ability, and fault response efficiency of the underwater sensing network, including: Node detection module 1, which is used to configure multiple sensing nodes. Each sensing node includes multiple sensors, and each sensor is used to collect multi-dimensional detection data at a preset period; The node communication module 2 is connected to the node detection module 1 and is used to form a self-organizing multi-hop network to establish communication connections for each sensing node. The detection data detected by each sensing node forms a local network health status map through a status sharing protocol. The local network health status map includes multiple node status data, and the node status data includes node self-check data, physical environment parameters, and chemical environment parameters; The preprocessing module 3 is connected to the node communication module 2 and is used to perform preprocessing and normalization processing on the node self-check data, physical environment parameters, and chemical environment parameters in sequence to construct a training data set; The hybrid training module 4 is connected to the preprocessing module 3 and is used to train according to the training data set using a hybrid neural network. The hybrid neural network includes a time series modeling layer, a local feature extraction layer, and a fully connected layer. The time series modeling layer uses a long short-term memory network to capture the long-term dependence features of each node status in the training data set. The local feature extraction layer uses a convolutional neural network layer to extract the local abnormal fluctuation features within the time window in the training data set. The fully connected layer fuses the long-term dependence features and the local abnormal fluctuation features to output a fault risk score; The self-healing control module 5 is connected to the hybrid training module 4 and the node communication module 2 and is used to automatically trigger self-healing response measures according to the node status data at adjacent sensing nodes when the fault risk score at the current sensing node is greater than the preset self-healing risk threshold. The self-healing response measures include dynamic routing reconstruction, standby link activation, and data load redistribution.

[0021] Specifically, in this embodiment, the node self-check data includes node voltage, node current, node energy consumption status, signal-to-noise ratio, packet loss rate, and signal delay; The physical environment parameters include temperature, environmental pressure, salinity, dissolved oxygen concentration, and environmental turbidity; The chemical environment parameters include pH value and specific ion concentration.

[0022] Working principle of Embodiment 1: In a protection project of a sunken ship site in the coastal area, a sensing node communication system based on Embodiment 1 is deployed to construct an underwater perception network covering key areas around the site. The system consists of multiple sensing nodes, and each node integrates multiple sensors and an intelligent communication module.

[0023] Node data acquisition: Each sensing node is configured with a sensing device to detect the following detection data: Node self-check data: including self-check indicators such as node voltage, current, energy consumption status, signal-to-noise ratio, packet loss rate, and signal delay, which can reflect the running health status of the node device itself in real time; Physical environment parameters: including information such as temperature, environmental pressure, salinity, dissolved oxygen concentration, and environmental turbidity, which are used to evaluate environmental changes; Chemical environmental parameters: including the pH value of the water body and the concentration of specific ions (such as chloride ions, sodium ions, etc.), which are used to judge the environmental corrosiveness and potential impact on cultural relics.

[0024] The sensing nodes synchronously collect the above multi-dimensional detection data at preset time intervals, and construct a self-organizing multi-hop network through underwater acoustic communication to achieve data sharing and status coordination.

[0025] Data processing and model training: All data are first normalized and outliers are removed through the preprocessing module 3 to form a structured training data set. The system uses a hybrid neural network model: Composition of the input data of the hybrid neural network model: multiple sensing data time series of a sensing node within a certain time window (such as 30 minutes); Data dimension: Node self-check data (such as voltage, current, packet loss rate, signal delay, signal-to-noise ratio) - a total of 5 dimensions; Physical parameters (such as temperature, ambient pressure, salinity, dissolved oxygen, turbidity) - a total of 5 dimensions; Chemical parameters (such as pH value, ion concentration) - a total of 2 dimensions; Total: 12-dimensional data channels × time series to form an input tensor.

[0026] The time series modeling layer uses an LSTM network to capture the long-term change trends of time series signals such as node voltage, current, and signal-to-noise ratio, and form long-term dependence features. The network structure of the time series modeling layer includes 1 to 2 stacked LSTM units, and each layer contains 64 or 128 LSTM units; The local feature extraction layer uses a CNN to extract features from the fluctuations of parameters such as pH value, salinity, and packet loss rate within a short time window; The fully connected layer includes a feature fusion layer. The feature fusion layer concatenates the long-term dependence features extracted by the time series modeling layer and the local features extracted by the local feature extraction layer, and then uses a Flatten layer to convert it into a vector; and then inputs it into the fully connected structure for fusion modeling to output the fault risk score of each node.

[0027] Self-healing response mechanism: When the risk score of a certain node exceeds the self-healing risk threshold due to drastic environmental changes or equipment aging, the system immediately triggers the self-healing mechanism according to the status data of adjacent nodes in the graph: If the signal-to-noise ratio of this node continues to decline and the signal delay increases, the system judges that its communication is unstable and automatically performs dynamic routing reconstruction; If the node voltage drops significantly and the energy consumption status is abnormal, the standby communication link is activated; If the turbidity and pH value of the node monitoring area increase abnormally, affecting the quality of data collection, the data load reallocation mechanism is started, and the task is temporarily switched to the adjacent healthy node.

[0028] The whole process does not require manual intervention, realizing continuous, stable and safe collection of key underwater data, and ensuring the efficient development of the site environment assessment and protection work.

[0029] Preferably, the preprocessing module 3 includes: The timing division unit 31 is used to record the node self-check data, physical environment parameters, and chemical environment parameters in a time series, and use the sliding window technology to segment the data to form continuous time series data samples; The preprocessing unit 32 is connected to the timing division unit 31 and is used to perform data cleaning, outlier detection, and missing value filling on the time series data samples in sequence to obtain preprocessed data samples; The normalization unit 33 is connected to the preprocessing unit 32 and is used to perform normalization processing on the multi-dimensional data in the preprocessed data samples according to the preset normalization method to form a training data set.

[0030] Specifically, in the embodiment, the timing division unit 31 records data every 10 seconds, and adopts a sliding window mechanism, and makes the following settings: Window length: 60 (i.e., 600 seconds = 10 minutes); Sliding step: 30 (semi-overlapping window, 30 samples as one "step"). A sample sequence is constructed for each window, and each sample sequence contains 60 consecutive sequences of 12 types of data such as node voltage, current, delay, SNR, temperature, ambient pressure, turbidity, salinity, dissolved oxygen, pH value, specific ion concentration, and data packet loss rate.

[0031] The preprocessing unit 32 sequentially executes the following preprocessing processes on each window sample: Data cleaning: Remove obviously incorrect data (such as invalid outliers like voltage < 2V, pH < 1, etc.); Mark the points exceeding the set boundary (such as z-score > 3).

[0032] Outlier detection: Use an algorithm based on the median absolute deviation (MAD) to detect mutation points; Instead of deleting a single jump value immediately, replace it with the median within the window.

[0033] Missing value filling: Use the two-way interpolation method (two-way linear + short-term mean) to fill in the missing data; Ensure the continuity of the time series and provide input integrity for subsequent LSTM modeling.

[0034] Example: If the pH value is missing at the 38th second, it is filled with the linearly extrapolated mean value of the 37th and 39th seconds; if the continuous missing exceeds 3, it is replaced with the average pH value in the window.

[0035] Normalization unit 33: After preprocessing, the system performs a normalization operation: all variables uniformly use the Z-score normalization method, and the normalized data is saved in the training dataset.

[0036] This preprocessing module 3 can reduce more than 85% of the abnormal data inputs; the normalization strategy eliminates the scale deviation between the sensing data, making the model learning more stable; the sliding window mechanism increases the sample size, improves the training sufficiency and the model robustness; this embodiment ensures that the model can effectively operate in the real and complex marine environment and accurately predict the fault trend of underwater nodes.

[0037] Embodiment 2 is the second embodiment of the present invention. Different from the previous embodiment, this embodiment provides a fault detection module 6 and an incremental training module 7, which can realize the adaptive correction and retraining of the model prediction and optimize the overall risk prediction performance of the system. It includes: Fault detection module 6, connected to the node communication module 2, is used to detect faults according to the status data of each node and generate corresponding fault detection results, and process the fault actual score based on the fault detection results; Incremental training module 7, connected to the fault detection module 6 and the hybrid training module 4, is used to process the prediction deflection value according to the fault actual score and the fault risk score, and perform dynamic adjustment and retraining on the weight parameters of the hybrid neural network based on the prediction deviation value to perform incremental training and update on the hybrid neural network.

[0038] Working principle of Embodiment 2: The fault detection module 6 is connected to the node communication module 2, and its main task is to convert the operation status data of each node into a "practical fault degree" score value that can be judged by the system. Specifically, it includes: Data collection and integration: Receive the node self-check data (such as voltage, current, delay, signal-to-noise ratio, packet loss rate), environmental parameters (temperature, pH, salinity, image quality, etc.) and the communication status feedback of other adjacent nodes; Rule recognition and fault determination: Based on a set of detection rules and models (for example: if the voltage is lower than the limit and the delay is high, then there is a fault; if the image appears strongly blurred and color offset, it is determined as abnormal), perform a logical judgment on the current node status; Score generation: Map the determination result to the fault actual score , indicating the severity of the fault of the node in the current cycle. Among them: Normal; : Slight to moderate abnormality; : Severe fault, self-healing response required.

[0039] Result output: Transmit the actual fault score and the fault risk score of the current node to the incremental training module together.

[0040] The incremental training module 7 is connected to the fault detection module 6 and the hybrid training module 4, aiming to achieve online adjustment and adaptive learning of the model and maintain prediction accuracy. Specifically, it includes: Prediction deviation calculation: Receive the prediction score currently output by the hybrid neural network, calculate the difference with the actual score output by the fault detection module, and calculate the prediction deviation : .

[0041] Error judgment and activation mechanism: The system sets a deviation tolerance threshold. When is greater than the deviation tolerance threshold: and continuously exceeds the set period (such as 3 rounds), the system determines that the current model has a "prediction deviation" and weight update is required.

[0042] Training sample construction: Temporarily store such high-deviation samples (including input sequences and error annotations) in the "incremental training sample pool".

[0043] Weight fine-tuning and incremental training: Perform mini-batch backpropagation based on high-deviation samples; use a low learning rate (such as 0.0005) to fine-tune the weights of the LSTM layer, CNN layer or fully connected layer in the network; output the updated parameter weights after training.

[0044] Model update and iteration: Automatically load the new weights into the hybrid neural network model; start using the updated model to continue scoring and predicting in the next cycle; form a continuous closed-loop self-evolving model update mechanism.

[0045] This embodiment ensures that the model not only has high prediction ability, but also can continuously correct its own reasoning deviation according to the actual operation feedback, enabling the model to have the intelligent characteristics of "self-learning", "self-correction" and "self-evolution", and maintaining high-precision prediction performance and adaptability during long-term operation.

[0046] Preferably, the functional expression of the hybrid neural network includes: ; ; ; ; ; ; Among them, represents the fault risk score, represents the time window length, represents the time variable, represents the Gamma weight function constructed by the current voltage and temperature of the node, represents the current voltage of the node, represents the standard deviation of the temperature, represents the mean value of the temperature, represents the current perturbation modulation term, represents the composite excitation factor formed by the product of the current and the packet loss rate, represents the sensitivity adjustment factor of the environmental pressure change, represents the historical current mean value, represents the change range of the environmental turbidity, represents the change amount of the signal-to-noise ratio, represents the communication and activation modeling function, represents the first kind of first-order Bessel function to measure the amplitude of environmental perturbation, represents the second kind of zero-order Bessel function representing the node redundancy connectivity strength, represents the salinity perturbation frequency factor, represents the complexity parameter of the current network communication topology, represents the rate of change of the node energy consumption, represents the time window width of the pH value, represents the delay attenuation and ion anomaly suppression function, represents the maximum delay of the current signal, represents the historical average signal delay, represents the data channel congestion factor, represents the time-dependent adjustment coefficient, represents the mean value of the specific ion concentration, represents the standard deviation of the specific ion concentration, represents the self-healing trigger suppression function, represents the self-healing response time activation coefficient, represents the state consistency measurement factor between adjacent nodes, represents the natural logarithm function, represents the complexity quantitative index of the ion species in the current environment.

[0047] Specifically, in this embodiment, in a Song Dynasty shipwreck site area in a certain sea area of the South China Sea, the protection unit deployed a set of Internet of Things underwater self-healing control systems based on the present invention, which altogether included 20 distributed intelligent underwater nodes. The node numbered A12 was located at a water depth of 11.8 meters and was specifically used to monitor local water chemical changes. This node showed high-frequency fluctuations for three consecutive days and was identified as a potential abnormal node by the system model, triggering a risk score assessment.

[0048] Data collected by the node (average value, unit standardized): Voltage , Current , Average temperature , Standard deviation , Environmental pressure factor , Historical average current , Turbidity range , Signal-to-noise ratio fluctuation , Salinity excitation frequency , Communication topology complexity , Node energy consumption rate , PH window , Maximum signal delay , Average delay , Congestion coefficient , Current average specific ion concentration , Standard deviation , Self-healing trigger coefficient , Node state consistency , Ion complexity index , Time integration window , The system calculates the integral at intervals of one second.

[0049] Fit a Gamma-Gaussian mixture curve through voltage and temperature to reflect the influence of temperature on node power supply fluctuations; Construct a perturbation enhancement model through the logarithmic relationship between current and environmental pressure and the changes in turbidity and signal-to-noise ratio; Simulate the non-linear interference and PH-related excitation in the communication process to make the score sensitive to the link condition; Couple and model the specific ion concentration anomaly and signal delay problem as key risk factors; Suppress rapid response actions through hyperbolic secant to prevent the system from being frequently mis-triggered; Accumulate the full dynamic risk curve within 10 minutes in integral form to obtain the failure risk score of the node.

[0050] After the model integral operation, the final score calculation result of node A12 is: ; The system scoring judgment logic is as follows:

[0051] Since , the system determines that the node is in the "high - risk abnormal warning range (0.7 to 1.0)", and automatically executes the following self - healing strategies: Dynamic routing reconstruction: Switch the original data path forwarding task of node A12 to adjacent nodes A11 and A13; Standby link activation: Activate the low - frequency acoustic channel link to take over part of its data communication; Task load redistribution: Redistribute the water chemistry monitoring tasks of this node to adjacent nodes in the ratios of 20%, 30%, and 50%; Incremental storage of training data: Add the data features of this time to the "risk abnormal sample library" of the model for subsequent training model optimization.

[0052] Embodiment 3, referring to the figure, is the second embodiment of the present invention. Different from the previous embodiment, this embodiment provides a function optimization module 8, which can accurately respond to latent faults caused by image degradation, unstable acoustic channels, etc., and achieve higher - accuracy risk prediction and system adaptability. Among them, the node state data also includes underwater acoustic interference data and underwater optical interference data. The underwater acoustic interference data includes power spectral density, and the underwater optical interference data includes optical image blur, color deviation drift, and image noise index.

[0053] It also includes a function optimization module 8, which is connected to the node communication module 2 and the hybrid training module 4, and is used to optimize the communication and activation modeling functions according to the power spectral density, optical image blur, and image noise index, and optimize the self - healing trigger suppression function according to the image blur and color deviation drift, and then update the fault risk score according to the optimized communication and activation modeling function and the optimized self - healing trigger suppression function.

[0054] The working principle of Embodiment 3: In a complex underwater environment, due to non - structural changes such as fluctuations in underwater image quality and acoustic channel interference, traditional feature models often cannot capture them in time, resulting in a large deviation between the risk prediction result and the actual node state.

[0055] For this reason, the function optimization module 8 introduces acoustic and optical interference factors and embeds them in the fault risk scoring function in the form of complex functions, enabling the system to have the ability to accurately respond to non - linear and multi - dimensional perturbation signals.

[0056] The core task of the function optimization module 8 is to embed the above interference factors into the scoring function (communication and activation modeling function and self-healing trigger suppression function) in a composite mathematical expression, so as to dynamically adjust the input of complex function structures such as activation function, Bessel function, and error function during scoring calculation, and then adjust the output score.

[0057] When the interference term shows rapid fluctuations (for example: the image noise index doubles in a short time), the function optimization module 8 will directly amplify the local fluctuation response term in the scoring function, causing the risk score of the model to rise rapidly in the short term, realizing early warning of latent anomalies.

[0058] At the same time, by dynamically adjusting the amplification coefficient and normalization scale in the function structure, the system can automatically adapt to different degrees of interference effects according to different water areas and node deployment situations, maintaining the balance between the stability and sensitivity of the score.

[0059] By introducing the function optimization module 8 and structurally embedding the underwater interference factors into the risk scoring function, the system can effectively improve the recognition accuracy of latent faults such as image degradation and unstable sound channels; at the same time, it can avoid misjudging optical or acoustic quality fluctuations as ordinary physical anomalies; furthermore, it can achieve a higher-confidence fault score output, improving the prediction accuracy and adaptive ability of the overall system.

[0060] Preferably, the optimized communication and activation modeling function is configured as: ; where represents the optimized communication and activation modeling function, represents the normalized power spectral density value, represents the image noise index, represents the fuzzy index adjustment factor, represents the Bessel perturbation frequency term, which is used to simulate the periodic excitation of PSD in underwater acoustic propagation; [[ID=2,7]]The optimized self-healing trigger suppression function is configured as: ; where represents the optimized self-healing trigger suppression function, represents the image blur index, represents the fuzzy index amplification coefficient, which is used to control the interference intensity of the blur on activation, represents the color deviation drift difference, respectively represent the color difference fluctuation coefficients of the G and R channels. The above optimization makes the self-healing trigger suppression function more sensitive and adjustable to blur and color deviation, and avoids frequent self-healing caused by misjudgment; Then the function expression of the updated fault risk score is configured as: ; Among them, represents the updated failure risk score.

[0061] Specifically, in this embodiment, in a protection project of a Qing Dynasty sunken ship site in the East China Sea, the team deployed the underwater self-healing control system based on the Internet of Things of the present invention, and the system is composed of multiple intelligent sensing nodes. The node numbered A09 is installed near the starboard area of the main hull and undertakes the tasks of underwater image acquisition, environmental monitoring, and acoustic link maintenance in the key cultural relics area.

[0062] This node has frequently occurred features such as blurred images, unstable links, and serious color deviation in the past two days, and the system automatically calls the optimized risk scoring function to comprehensively evaluate its failure risk by combining image and acoustic data.

[0063] Key input parameter values (after standardization) collected by the node: Signal-to-noise ratio fluctuation , Power spectral density , Signal delay , Topological complexity , Image blur index , Image noise index , Color deviation drift , Color difference fluctuation coefficient , Image channel consistency index , Voltage , Current , PH value window , Temperature mean , Temperature standard deviation , Energy consumption rate , Specific ion concentration , Specific ion concentration , Blur index adjustment coefficient , Image blur amplification factor , Salinity frequency term , Bessel perturbation frequency term , PSD interference amplitude term ; Substitute the above parameters into the function expression of the updated failure risk score, and we can get .

[0064] The system scoring judgment logic is as follows:

[0065] Since , the A09 node is in the high-risk interval (0.91), and the system immediately executes the following self-healing control: Image acquisition migration: Pause the video acquisition task of A09 and transfer it to the adjacent node A08 to perform supplementary shooting from the same perspective; Data path reconstruction: Reset the data streams of images and acoustic signals, and use the backup communication link (audio 2) to bypass A09; Node status announcement broadcast: Through the status sharing protocol, convey the fault status of A09 to all nodes within the network; Model feedback learning: Package the input and result of the scoring function this time and store them in the model incremental training dataset; Administrator alert push: The system pops up a high-risk prompt at the control terminal and generates a scoring curve chart for operation and maintenance analysis.

[0066] This embodiment shows that after introducing the modeling of acoustic and optical interference factors, the scoring function can accurately respond to latent faults caused by image degradation, unstable sound channels, etc., and achieve higher-accuracy risk prediction and system self-adaptive capabilities.

[0067] A communication method for sensing nodes with a self-healing mechanism, applied to the above-mentioned communication system for sensing nodes with a self-healing mechanism, includes: Step S1, the node detection module 1 configures multiple sensing nodes, each sensing node includes multiple sensors, and each sensor is used to collect multi-dimensional detection data according to a preset period; Step S2, the node communication module 2 forms a self-organizing multi-hop network to establish communication connections for each sensing node. The detection data detected by each sensing node forms a local network health status map through the status sharing protocol. The local network health status map includes multiple node status data, and the node status data includes node self-check data, physical environment parameters, and chemical environment parameters; Step S3, the preprocessing module 3 performs preprocessing and normalization processing on the node self-check data, physical environment parameters, and chemical environment parameters in sequence to construct a training dataset; Step S4, the hybrid training module 4 trains according to the training dataset using a hybrid neural network. The hybrid neural network includes a time series modeling layer, a local feature extraction layer, and a fully connected layer. The time series modeling layer uses a long short-term memory network to capture the long-term dependence features of each node status in the training dataset. The local feature extraction layer uses a convolutional neural network layer to extract the local abnormal fluctuation features within the time window in the training dataset. The fully connected layer fuses the long-term dependence features and the local abnormal fluctuation features to output a fault risk score; Step S5, when the fault risk score at the current sensing node is greater than the preset self-healing risk threshold, the self-healing control module 5 automatically triggers self-healing response measures according to the node status data at the adjacent sensing nodes. The self-healing response measures include dynamic routing reconstruction, standby link activation, and data load redistribution.

[0068] The above are only the preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions falling within the concept of the present invention belong to the protection scope of the present invention. It should be noted that for those of ordinary skill in the art, several improvements and refinements made without departing from the principle of the present invention should also be regarded as within the protection scope of the present invention.

Claims

1. A sensing node communication system with a self-healing mechanism, characterized in that, Including: A node detection module (1) for configuring a plurality of sensing nodes, each of the sensing nodes including a plurality of sensors, and each of the sensors being used to collect multi-dimensional detection data according to a preset period; A node communication module (2) connected to the node detection module (1) for forming a self-organizing multi-hop network to establish a communication connection for each of the sensing nodes. The detection data detected by each of the sensing nodes forms a local network health status map through a state sharing protocol. The local network health status map includes a plurality of node status data, and the node status data includes node self-check data, physical environment parameters, and chemical environment parameters; A preprocessing module (3) connected to the node communication module (2) for preprocessing and normalizing the node self-check data, the physical environment parameters, and the chemical environment parameters in sequence to construct a training data set; A hybrid training module (4) connected to the preprocessing module (3) for training according to the training data set by using a hybrid neural network. The hybrid neural network includes a time series modeling layer, a local feature extraction layer, and a fully connected layer. The time series modeling layer uses a long short-term memory network to capture the long-term dependence features of each node status in the training data set. The local feature extraction layer uses a convolutional neural network layer to extract the local abnormal fluctuation features within a time window in the training data set. The fully connected layer fuses the long-term dependence features and the local abnormal fluctuation features to output a fault risk score; A self-healing control module (5) connected to the hybrid training module (4) and the node communication module (2) for automatically triggering a self-healing response measure according to the node status data at a neighboring sensing node when the fault risk score at the current sensing node is greater than a preset self-healing risk threshold. The self-healing response measures include dynamic routing reconstruction, standby link activation, and data load reallocation.

2. The communication system of the sensing node with a self-healing mechanism according to claim 1, characterized in that, The preprocessing module (3) includes: A time series partitioning unit (31) for recording the node self-check data, the physical environment parameters, and the chemical environment parameters in a time series and performing data segmentation by using a sliding window technique to form continuous time series data samples; A preprocessing unit (32) connected to the time series partitioning unit (31) for sequentially performing data cleaning, outlier detection, and missing value filling on the time series data samples to obtain preprocessed data samples; A normalization unit (33) connected to the preprocessing unit (32) for normalizing the multi-dimensional data in the preprocessed data samples according to a preset normalization method to form the training data set.

3. The communication system of the sensing node with a self-healing mechanism according to claim 1, wherein: The node self-check data includes node voltage, node current, node energy consumption status, signal-to-noise ratio, packet loss rate, and signal delay; The physical environment parameters include temperature, environmental pressure, salinity, dissolved oxygen concentration, and environmental turbidity; The chemical environment parameters include pH value and specific ion concentration.

4. The communication system of sensing nodes with a self-healing mechanism according to claim 1, characterized in that, It also includes: A fault detection module (6), connected to the node communication module (2), is configured to perform fault detection on the status data of each node, generate corresponding fault detection results, and process the fault detection results to obtain actual fault scores; An incremental training module (7), connected to the fault detection module (6) and the hybrid training module (4), is configured to process the actual fault scores and the fault risk scores to obtain predicted deflection values, and perform dynamic adjustment and retraining on the weight parameters of the hybrid neural network based on the predicted deviation values, so as to perform incremental training and update on the hybrid neural network.

5. The communication system of the sensing node with a self-healing mechanism according to claim 3, characterized in that: The functional expression of the hybrid neural network includes: ； ； ； ； ； ； Among them, represents the failure risk score, represents the time window length, represents the time variable, represents the Gamma weight function constructed by the current voltage and temperature of the node, represents the current voltage of the node, represents the standard deviation of the temperature, represents the mean value of the temperature, represents the current disturbance modulation term, represents the composite excitation factor formed by the product of the current and the packet loss rate, represents the sensitivity adjustment factor of the environmental pressure change, represents the historical current mean value, represents the change interval of the environmental turbidity, represents the change amount of the signal-to-noise ratio, represents the communication and activation modeling function, represents the first kind of first-order Bessel function to measure the amplitude of environmental disturbance, represents the second kind of zero-order Bessel function to represent the node redundancy connectivity strength, represents the salinity disturbance frequency factor, represents the complexity parameter of the current network communication topology, represents the rate of change of the node energy consumption, represents the time window width of the pH value, represents the delay attenuation and ion anomaly suppression function, represents the maximum current signal delay, represents the historical average signal delay, represents the data channel congestion factor, represents the time-dependent adjustment coefficient, represents the mean value of the specific ion concentration, represents the standard deviation of the specific ion concentration, represents the self-healing trigger suppression function, represents the self-healing response time activation coefficient, represents the state consistency measurement factor between adjacent nodes, represents the natural logarithm function, represents the complexity quantitative index of the ion species in the current environment.

6. The communication system of the sensing node with a self-healing mechanism according to claim 5, characterized in that: The node status data further includes underwater acoustic interference data and underwater optical interference data. The underwater acoustic interference data includes power spectral density, and the underwater optical interference data includes optical image blur, color deviation drift, and image noise index.

7. The communication system of the sensing node with a self-healing mechanism according to claim 6, characterized in that, It further includes a function optimization module (8), connected to the node communication module (2) and the hybrid training module (4), configured to optimize the communication and activation modeling function according to the power spectral density, the optical image blur, and the image noise index, and optimize the self-healing trigger suppression function according to the image blur and the color deviation drift, and then update the fault risk score according to the optimized communication and activation modeling function and the optimized self-healing trigger suppression function.

8. The communication system of the sensing node with a self-healing mechanism according to claim 7, characterized in that: The optimized communication and activation modeling function is configured as: ; wherein, represents the optimized communication and activation modeling function, represents the normalized power spectral density value, represents the image noise index, represents the fuzzy index adjustment factor, represents the Bessel perturbation frequency term, which is used to simulate the periodic perturbation of PSD in underwater acoustic propagation; The optimized self-healing trigger suppression function is configured as: ； Among them, represents the optimized self-healing trigger suppression function, represents the image blurriness index, represents the blurriness index amplification factor, which is used to control the interference strength of the activated interference, represents the color deviation drift difference, respectively represent the color difference fluctuation coefficients of the G and R channels; Then the functional expression of the updated fault risk score is configured as: ； Among them, represents the updated failure risk score.

9. A communication method for a sensing node with a self-healing mechanism, which is applied to the sensing node communication system with a self-healing mechanism described in any one of claims 1-8, characterized in that, Including: Step S1, the node detection module (1) configures multiple sensing nodes, each of the sensing nodes includes multiple sensors, and each of the sensors is configured to collect multi-dimensional detection data according to a preset period; Step S2, the node communication module (2) forms a self-organizing multi-hop network to establish communication connections for each of the sensing nodes. The detection data detected by each of the sensing nodes forms a local network health status map through a status sharing protocol. The local network health status map includes multiple node status data, and the node status data includes node self-check data, physical environment parameters, and chemical environment parameters; Step S3, the preprocessing module (3) performs preprocessing and normalization processing on the node self-check data, the physical environment parameters, and the chemical environment parameters in sequence to construct a training data set; Step S4, the hybrid training module (4) performs training using a hybrid neural network according to the training data set. The hybrid neural network includes a time series modeling layer, a local feature extraction layer, and a fully connected layer. The time series modeling layer uses a long short-term memory network to capture the long-term dependence features of each node status in the training data set. The local feature extraction layer uses a convolutional neural network layer to extract local abnormal fluctuation features within a time window in the training data set. The fully connected layer fuses the long-term dependence features and the local abnormal fluctuation features to output a fault risk score; Step S5, when the failure risk score at the current sensing node of the self-healing control module (5) is greater than a preset self-healing risk threshold, the self-healing response measures are automatically triggered according to the node state data at the sensing nodes adjacent thereto. The self-healing response measures include dynamic routing reconstruction, standby link activation, and data load redistribution.

Images