An event-triggered distributed adaptive sampling method for ocean observatory network

Abstract

The application discloses an event-triggered adaptive sampling method for a distributed seabed observation network, and relates to the technical field of deep sea exploration.The method comprises the following steps: S01, collecting multi-modal sensing data of a plurality of observation nodes in the distributed seabed observation network, wherein the multi-modal sensing data comprises temperature data, pressure data and three-axis vibration data; preprocessing the multi-modal sensing data of each observation node and extracting characteristic parameters corresponding to each observation node; S02, performing abnormality detection on the characteristic parameters of each observation node, judging whether an event triggering condition is met, and if the event triggering condition is met, generating an event triggering signal and recording an event type, wherein the event triggering condition comprises one or more combinations of a vibration signal amplitude exceeding a threshold value, a temperature change rate exceeding a threshold value and a pressure change rate exceeding a threshold value. The application solves the problems of resource waste, event response lag and insufficient node cooperation caused by fixed sampling in the prior art.

Description

Technical Field

[0001] This invention relates to the field of deep-sea exploration technology, and in particular to an adaptive sampling method for a distributed seabed observation network based on event triggering. Background Technology

[0002] Long-term seabed observation systems are an important means of achieving in-situ, continuous monitoring of multiple parameters of the deep-sea environment, and are widely used in fields such as monitoring seabed geological activity, marine environmental assessment, and deep-sea resource development. Existing seabed observation systems typically integrate temperature sensors, pressure sensors, and acceleration sensors (seismic sensors), continuously acquiring seabed temperature, pressure, and triaxial vibration signals at a fixed sampling frequency, and storing the data locally or transmitting it to a shore-based data center via an acoustic communication link.

[0003] Existing systems lack the ability to respond quickly to sudden geological events on the seabed, such as earthquakes, hydrothermal eruptions, and tectonic activities. Fixed sampling modes cannot automatically increase sampling resolution to capture transient details when events occur, resulting in the loss or distortion of critical event data. Summary of the Invention

[0004] To address the aforementioned technical problems, the present invention employs an event-triggered adaptive sampling method for a distributed seabed observation network, the method comprising the following steps: S01. Collect multimodal sensing data from multiple observation nodes in a distributed seabed observation network. The multimodal sensing data includes temperature data, pressure data, and triaxial vibration data. Preprocess the multimodal sensing data of each observation node and extract the characteristic parameters corresponding to each observation node. S02. Perform anomaly detection on the characteristic parameters of each observation node to determine whether the event triggering conditions are met. If the event triggering conditions are met, generate an event triggering signal and record the event type. The event triggering conditions include at least one or more combinations of vibration signal amplitude exceeding the threshold, temperature change rate exceeding the threshold, and pressure change rate exceeding the threshold. S03. Based on the event triggering signals and event types of each observation node, a distributed collaborative decision matrix is ​​constructed. The collaborative decision matrix is ​​used to describe the triggering state of each observation node and the spatial correlation between nodes. S04. Based on the collaborative decision matrix, determine the target sampling frequency and data transmission mode of each observation node. For observation nodes that trigger events, increase their sampling frequency and enable real-time data backhaul. For observation nodes that do not trigger events but meet the spatial association conditions with the triggering node, increase their sampling frequency. For observation nodes that do not trigger events and do not meet the spatial association conditions, maintain the basic sampling frequency. S05. Each observation node performs data acquisition according to the determined target sampling frequency and transmits data according to the determined data transmission mode. S06. Collect action execution feedback data from each observation node, update feature parameters, evaluate the event state change through the event decay judgment mechanism, and restore the sampling frequency of each observation node to the basic sampling frequency when the event state is determined to have decayed.

[0005] Preferably, the event fading determination mechanism includes: A preset event decay threshold is set to determine whether an event has decayed by quantifying the difference between the current feature parameters and the baseline state before the event was triggered. in: If the degree of difference exceeds the event decay threshold, maintain the current sampling frequency; If the difference does not exceed the event fading threshold, the event is determined to have faded, and the sampling frequency is restored to the basic sampling frequency. The observation termination condition includes at least one of the following: The observation duration has reached the preset maximum observation duration; The difference between the feature parameters and the baseline state in multiple consecutive iterations remains below a preset stability threshold; Received a remote termination command from the shore-based control center.

[0006] Preferably, in step S01, the multimodal sensing data of each observation node are preprocessed to extract characteristic parameters reflecting the current state of the seabed environment, including: Temperature data, pressure data, and triaxial vibration data are filtered and denoised to eliminate environmental noise and sensor-inherent noise. Vector synthesis is performed on the filtered triaxial vibration data to calculate the root mean square value of the vibration synthesis amplitude and the vibration energy accumulation index. The step size of the sliding time window is dynamically adjusted according to the changing trend of the current pressure data. The step size is shortened during the pressure rise phase to improve the response sensitivity, and the step size is increased during the pressure stabilization phase to reduce the computational load. Calculate the rate of temperature change and the rate of pressure change based on a sliding time window; The vibration synthesis amplitude, vibration energy index, temperature change rate, and pressure change rate are used as the characteristic parameters.

[0007] As a preferred embodiment, in step S02, determining whether the event triggering conditions are met requires pre-setting a basic threshold for vibration amplitude, a basic threshold for temperature change rate, and a basic threshold for pressure change rate, and using the standard deviation of the characteristic parameters of each observation node within a preset long time window before the event triggering as a dynamic adjustment coefficient to generate an adaptive triggering threshold. When the root mean square value of the vibration synthesis amplitude of any observation node exceeds the dynamic threshold of vibration amplitude in the adaptive triggering threshold, and the vibration energy accumulation index maintains a monotonically increasing trend within a continuous time window, the event triggering condition is determined to be met, and the event type is recorded as a vibration event. When the rate of temperature change exceeds the basic threshold for both the rate of temperature change and the rate of pressure change exceeds the basic threshold for both the rate of pressure change, the event triggering condition is determined to be met, and the event type is recorded as a hydrothermal event. When the root mean square value of the composite vibration amplitude exceeds the dynamic threshold of the vibration amplitude, and at least one of the temperature change rate and pressure change rate exceeds its corresponding basic threshold, the event triggering condition is determined to be met, and the event type is recorded as a composite event.

[0008] Preferably, the construction of the distributed collaborative decision matrix in step S03 includes: S31. Construct an initial decision matrix with observation nodes as rows and event types as columns. Each matrix element represents the triggering state of the corresponding node under the current event type. S32. Based on the spatial location information of each observation node, calculate the spatial distance between nodes and establish a node adjacency table; S33. Introduce a signal delay compensation factor. Based on the seawater sound velocity profile at the location of each observation node and the spatial distance between nodes, calculate the theoretical propagation delay of the vibration / acoustic signal between different nodes, and perform delay correction on the trigger timestamp of each node. S34. For nodes that have not triggered an event, determine whether the spatial distance between the node and any node that triggered an event is less than a preset cooperative distance threshold according to the node adjacency table. If so, mark the node as a cooperative node. S35. Write the trigger status and collaborative flag of each node into the collaborative decision matrix.

[0009] Preferably, the determination of the spatial association condition in step S04 is based on a preset cooperative distance threshold. The cooperative distance threshold is dynamically determined according to the event type. Specifically, the cooperative distance threshold corresponding to a vibration event is less than the cooperative distance threshold corresponding to a hydrothermal event, and the cooperative distance threshold corresponding to a composite event is the weighted sum of the cooperative distance thresholds corresponding to the vibration event and the hydrothermal event. When the spatial distance between a node that has not triggered an event and a node that has triggered an event is less than the cooperative distance threshold, the spatial association condition is determined to be met. When the spatial distance between a node that has not triggered an event and all nodes that have triggered an event is not less than the cooperative distance threshold, it is determined that the spatial association condition is not met.

[0010] Preferably, the step of determining the target sampling frequency and data transmission mode for each observation node in step S04 includes: For the observation node that triggered the event, its sampling frequency is increased to the first sampling frequency, and the data transmission mode is set to real-time feedback mode; For observation nodes that do not trigger events but meet spatial correlation conditions, their sampling frequency is increased to the second sampling frequency, and the data transmission mode is set to priority backhaul mode, wherein the second sampling frequency is lower than the first sampling frequency; For observation nodes that have not triggered events and do not meet the spatial correlation conditions, maintain the basic sampling frequency and set the data transmission mode to local storage mode; The data transmission mode is linked to the data compression strategy: in real-time return mode, a first compression ratio is used for lightweight compression transmission; in priority return mode, a second compression ratio is used for balanced compression transmission; and in local storage mode, a third compression ratio is used for high-fidelity lossless storage, wherein the first compression ratio < the second compression ratio < the third compression ratio.

[0011] Preferably, the numerical relationship between the first sampling frequency, the second sampling frequency, and the basic sampling frequency is: first sampling frequency > second sampling frequency > basic sampling frequency, and the first sampling frequency is N times the basic sampling frequency, and the second sampling frequency is M times the basic sampling frequency, where N > M > 1, and N and M are both positive integers. It also includes dynamically adjusting the values ​​of N and M based on the event trigger intensity. The event trigger intensity is calculated by weighting the over-amplitude of the vibration synthesis amplitude exceeding the dynamic threshold, the overscalar of the temperature change rate, and the pressure change rate. The higher the trigger intensity, the larger the values ​​of N and M.

[0012] Preferably, the step S06, which involves evaluating the event state change through an event fading determination mechanism, includes the following when the event state is determined to have faded: S61. The exponentially weighted moving average of the feature parameters within a preset time window before the event is triggered is used as the reference state, and the reference state is dynamically updated with the duration of stable operation to eliminate the long-term drift effect of the sensor. S62. Calculate the weighted Euclidean distance between the current feature parameter and the baseline state as the degree of difference, wherein the weight of each feature parameter is dynamically allocated according to the event type; S63. Set the fading determination holding time. When the difference level is lower than the event fading threshold for the first time, start a timer. Only after the cumulative duration of the difference level being lower than the event fading threshold exceeds the holding time will the event be finally determined to have faded, so as to prevent frequent false recovery caused by noise fluctuations.

[0013] Preferably, the method further includes node failure handling, wherein if any observation node fails to return valid action execution feedback data in a consecutive preset number of iterations, the node is marked as a failed node; By using the time difference of arrival of signals between multiple effective nodes adjacent to the failed node, spatial interpolation and reconstruction of vibration and temperature / pressure signals at the location of the failed node are performed to generate virtual node observation data, and the virtual node is temporarily incorporated into the collaborative decision matrix. When a failed node resumes normal communication, the virtual node is removed and the original node's collaborative decision matrix configuration is restored. If the failed node continues to fail to recover, it is permanently removed from the collaborative decision matrix, and the collaborative decision matrix of the remaining nodes is reconstructed.

[0014] The present invention has at least the following beneficial effects: 1. This invention detects anomalies in multimodal sensing data by setting event triggering conditions. The sampling frequency is increased and the data transmission mode is switched only when an anomaly in the seabed environment is detected. During periods of environmental stability, each observation node maintains the basic sampling frequency and stores the data locally. This effectively avoids the generation of a large amount of redundant data caused by fixed high-frequency sampling, significantly reduces the data storage pressure and the transmission burden of the underwater acoustic communication link, and significantly reduces the overall energy consumption of the system. This provides a reliable guarantee for the long-term, continuous, and in-situ deployment of the seabed observation network.

[0015] 2. This invention establishes a collaborative triggering and adaptive scheduling mechanism based on spatial correlation among multiple observation nodes by constructing a distributed collaborative decision-making matrix. When a node detects an event trigger, not only does that node increase its sampling frequency, but its spatially adjacent nodes also synchronously increase their sampling frequencies, achieving multi-node intensive observation within the event's influence area. This collaborative mechanism can effectively capture the spatiotemporal evolution of sudden seabed events, providing high-resolution multi-node synchronous observation data for the three-dimensional reconstruction and evolution analysis of geological events, significantly improving the completeness and accuracy of event observation.

[0016] 3. This invention achieves automatic recovery of the sampling strategy through an event decay determination mechanism. When the difference between the characteristic parameters and the baseline state falls below the event decay threshold, the system automatically restores the sampling frequency of each node to the baseline level, avoiding the need for manual intervention. Simultaneously, this invention also establishes a node failure handling mechanism and an event type-driven sampling strategy differentiation adjustment mechanism, further enhancing the system's robustness and adaptability in complex deep-sea environments. It can flexibly respond to different types of seabed geological events, effectively supporting long-term seabed observation missions. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the implementation structure provided in Embodiment 1 of the present invention; Figure 2 A flowchart provided for Embodiment 1 of the present invention; Figure 3 A flowchart of S03 provided in Embodiment 1 of the present invention; Figure 4 The flowchart for S06 provided in Embodiment 1 of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0021] Example 1

[0022] This embodiment provides an event-triggered adaptive sampling method for a distributed seabed observation network. The method includes the following steps: Figure 1 and Figure 2 As shown: S01. Collect multimodal sensing data from multiple observation nodes in the distributed seabed observation network (by deploying buoyancy chambers in the target sea area). The multimodal sensing data includes temperature data (temperature sensor), pressure data (pressure sensor), and triaxial vibration data (hydrophone). Preprocess the multimodal sensing data of each observation node and extract the characteristic parameters corresponding to each observation node. Furthermore, in the above embodiments, the multimodal sensing data of each observation node are preprocessed to extract characteristic parameters reflecting the current state of the seabed environment, including: Temperature data, pressure data, and triaxial vibration data are filtered and denoised to eliminate environmental noise and sensor-inherent noise. Vector synthesis is performed on the filtered triaxial vibration data to calculate the root mean square value of the vibration synthesis amplitude and the vibration energy accumulation index. The step size of the sliding time window is dynamically adjusted according to the changing trend of the current pressure data. The step size is shortened during the pressure rise phase to improve the response sensitivity, and the step size is increased during the pressure stabilization phase to reduce the computational load. Calculate the rate of temperature change and the rate of pressure change based on a sliding time window; The vibration synthesis amplitude, vibration energy index, temperature change rate, and pressure change rate are used as characteristic parameters.

[0023] Specifically, each observation node in the distributed seabed observation network acquires multimodal sensor data in real time through integrated temperature sensors, pressure sensors, and triaxial vibration data (hydrophones). Due to background noise in the deep-sea environment (such as ocean current disturbances and instrument self-noise), directly extracting features from raw data would lead to a higher false alarm rate; therefore, preprocessing is necessary. Specifically, the temperature, pressure, and triaxial vibration data from each observation node are denoised using either bandpass or low-pass filtering: For triaxial vibration data, considering that seabed microseismic signals are typically concentrated in the 0.1Hz-50Hz frequency band, a bandpass filter (such as a Butterworth bandpass filter, order 4-8) is used to filter out low-frequency ocean current interference and high-frequency circuit noise; for temperature and pressure data, whose changes are relatively gradual, a low-pass filter is used to eliminate inherent sensor noise and spike pulse interference during data acquisition, ensuring the smoothness of the output signal.

[0024] Based on filtering and denoising, vector synthesis processing is performed on the triaxial vibration data. Specifically, assuming the triaxial acceleration data collected at a certain observation node at time t are ax(t), ay(t), and az(t), the composite vibration amplitude A(t) is calculated using the formula A(t) = sqrt(ax(t)² + ay(t)² + az(t)²). Furthermore, to more accurately characterize the intensity and energy release features of the vibration event, a sliding time window is used to calculate the root mean square (RMS) value of the composite vibration amplitude. This RMS value effectively reflects the average energy level of the vibration signal within a short time window, avoiding false triggering caused by single-point noise. Simultaneously, by integrating or summing the squares of the composite vibration amplitude within the same time window, a cumulative vibration energy index is calculated.

[0025] Furthermore, seabed pressure data can indirectly reflect rapid changes in water depth or formation stress. When the pressure is in a rapid rising phase (corresponding to sudden events such as fault displacement caused by submarine earthquakes, disturbance of overlying water caused by hydrothermal eruptions, or sediment slippage), it indicates that the environmental conditions are undergoing drastic changes. At this time, higher time resolution is needed to capture the transient details of vibration and temperature / pressure signals. Therefore, the system automatically shortens the step size of the sliding window (e.g., from the default step size to 1 / 2 or 1 / 3) to improve response sensitivity and ensure that effective feature parameters can be quickly extracted at the beginning of the event, reducing trigger delay. Conversely, when the pressure data is in a stable phase (corresponding to the normal observation period without events), the seabed environment tends to be stable. At this time, increasing the step size of the sliding window (e.g., expanding it to 1.5-2 times the default step size) can effectively reduce the computational load and power consumption of the embedded processor (such as ARM Cortex-M series MCUs or DSPs) and extend the underwater working time of the observation node while ensuring that low-frequency slow-changing trends can be monitored. For example, during the pressure stabilization period, the sliding window step size is set to 2 seconds, and the window contains 200 sampling points (sampling rate of 100Hz). The system calculates the vibration RMS value and energy accumulation index every 2 seconds. When the pressure rise rate exceeds the preset change rate threshold (e.g., 0.01MPa / s) within 3 consecutive sampling cycles, the system automatically shortens the step size to 0.5 seconds, and the window contains only 50 sampling points, achieving the output of characteristic parameters every 0.5 seconds, which greatly improves the response speed in the early stage of the event.

[0026] Simultaneously, the temperature change rate and pressure change rate were calculated using the same sliding time window. The temperature change rate was calculated by linearly fitting or differentially analyzing the temperature data within the current sliding window to obtain the temperature change per unit time, ΔT / Δt. This parameter effectively indicates geological events such as hydrothermal activity and submarine volcanic eruptions accompanied by significant heat release. The pressure change rate was obtained by differentially analyzing the pressure data, yielding ΔP / Δt, which indicates rapid changes in geostress caused by water disturbance, rapid sediment deposition, or fault activity. Finally, the root mean square value (ARMS(t)) of the synthesized vibration amplitude, the cumulative vibration energy index (E(t)), the temperature change rate ΔT / Δt, and the pressure change rate ΔP / Δt obtained from the above processing were used as characteristic parameters of the observation node, providing a quantitative basis for subsequent anomaly detection and event trigger determination.

[0027] S02. Perform anomaly detection on the characteristic parameters of each observation node to determine whether the event triggering conditions are met. If the event triggering conditions are met, generate an event triggering signal and record the event type. The event triggering conditions include at least one or more combinations of vibration signal amplitude exceeding the threshold, temperature change rate exceeding the threshold, and pressure change rate exceeding the threshold.

[0028] Furthermore, in determining whether the event triggering conditions are met, it is necessary to pre-set the basic thresholds for vibration amplitude, temperature change rate, and pressure change rate, and use the standard deviation of the characteristic parameters of each observation node within a long time window before the event is triggered as a dynamic adjustment coefficient to generate an adaptive triggering threshold. When the root mean square value of the vibration synthesis amplitude of any observation node exceeds the dynamic threshold of vibration amplitude in the adaptive trigger threshold, and the vibration energy accumulation index maintains a monotonically increasing trend within a continuous time window, the event triggering condition is determined to be met, and the event type is recorded as a vibration event. When the rate of temperature change exceeds the basic threshold for both the rate of temperature change and the rate of pressure change exceeds the basic threshold for the rate of pressure change, the event triggering condition is determined to be met, and the event type is recorded as a hydrothermal event. When the root mean square value of the composite vibration amplitude exceeds the dynamic threshold of the vibration amplitude, and at least one of the temperature change rate and pressure change rate exceeds its corresponding basic threshold, the event triggering condition is determined to be met, and the event type is recorded as a composite event.

[0029] Specifically, the digital signal processor (DSP) or main control unit (MCU) of each observation node continuously receives the preprocessed characteristic parameter stream. The system first allocates an independent circular buffer in memory to cache historical vibration, temperature, and pressure characteristic data for each observation node within a preset long time window (e.g., the first 30 minutes) before an event triggers. Based on this, the system continuously calculates the standard deviation of vibration amplitude, temperature change rate, and pressure change rate within this time window. These real-time updated standard deviations are quantized into dynamic adjustment coefficients and weighted and summed with pre-programmed baseline thresholds for vibration amplitude, temperature change rate, and pressure change rate in the firmware, respectively, to generate an "adaptive trigger threshold" that adaptively drifts with environmental background noise. When the seabed environment is in a period of high background noise (such as frequent biological activity or turbulent ocean currents), the standard deviation of the characteristic parameters increases, and the dynamic threshold automatically rises, effectively avoiding false triggers caused by random environmental fluctuations. Conversely, when the environment enters an extremely stable period, the standard deviation approaches zero, and the dynamic threshold falls back to near the baseline threshold, thus maintaining high sensitivity to weak precursory anomalies.

[0030] Regarding the specific discrimination logic and physical mechanism of event types, this step constructs differentiated judgment criteria for different types of seafloor geodynamic processes. For the discrimination of seismic events, the system not only requires the root mean square (RMS) value of the composite vibration amplitude to exceed the aforementioned dynamic threshold, but also adds a stringent temporal morphological constraint that "the accumulated vibration energy index maintains a monotonically increasing trend within a continuous time window." The physical significance of this constraint is to exclude sudden, isolated transient impact noise (such as seafloor rockfalls or biological impacts), because the vibration signals caused by real tectonic earthquakes or seafloor landslides are usually accompanied by a continuous accumulation and release of energy, and their short-term energy envelope exhibits a monotonically increasing pattern. For the discrimination of hydrothermal events, the system adopts a synergistic dual over-threshold criterion of temperature and pressure. This is because deep-sea hydrothermal eruptions are often accompanied by the upwelling of high-temperature fluids, leading to a sudden rise in local temperature, while the thermal plume disturbs the surrounding water mass, causing pressure fluctuations. The synchronous anomaly of these two factors constitutes the unique fingerprint characteristics of hydrothermal activity. For the identification of complex events, the system is designed as a logical "OR" combination of at least one of the vibration triggering condition and the temperature and pressure triggering condition. This mechanism is specifically designed to capture chain disaster processes such as submarine earthquakes triggering landslides, which in turn cause turbidity currents or hydrothermal activity to overflow.

[0031] S03. Based on the event triggering signals and event types of each observation node, construct a distributed collaborative decision matrix. The collaborative decision matrix is ​​used to describe the triggering status of each observation node and the spatial correlation between nodes.

[0032] Furthermore, in the above embodiments, constructing a distributed collaborative decision-making matrix includes ( Figure 3 (as shown) S31. Construct an initial decision matrix with observation nodes as rows and event types as columns. Each matrix element represents the triggering state of the corresponding node under the current event type. S32. Based on the spatial location information of each observation node, calculate the spatial distance between nodes and establish a node adjacency table; S33. Introduce a signal delay compensation factor. Based on the seawater sound velocity profile at the location of each observation node and the spatial distance between nodes, calculate the theoretical propagation delay of the vibration / acoustic signal between different nodes, and perform delay correction on the trigger timestamp of each node. S34. For nodes that have not triggered an event, determine whether the spatial distance between the node and any node that has triggered an event is less than the preset collaborative distance threshold according to the node adjacency table. If so, mark the node as a collaborative node. S35. Write the trigger status and collaborative flag of each node into the collaborative decision matrix.

[0033] Specifically, in the deep-sea seabed observation network, an initial two-dimensional decision matrix is ​​constructed, with all nodes in the network as row vectors and event types such as seismic events, hydrothermal events, and combined events as column vectors. Each element in the matrix is ​​initialized with the local triggering state of the corresponding node under the current event type. For example, if node A detects that the seismic amplitude exceeds the threshold, the element is assigned the value "triggered," otherwise it is "not triggered." Based on this, the system calculates the Euclidean spatial distance between nodes using latitude and longitude or relative coordinates based on the pre-stored spatial location information of each node, and uses this to establish a node adjacency table. This table records the distance values ​​between each node and its surrounding nodes, providing a data foundation for subsequent collaborative distance determination.

[0034] Because the propagation speed of seabed vibrations or acoustic signals in water is limited (approximately 1500 m / s), and the distances between different observation nodes and the event source vary, the arrival time of signals generated by the same event at different nodes differs significantly. For example, if an earthquake event occurs near node A, node A detects the signal trigger at t=0, while node B, located 500 meters from the event source, detects the signal only after t≈0.33 seconds, and node C, located 1000 meters away, triggers the signal only after t≈0.67 seconds. If the system directly uses the original trigger times of each node for collaborative judgment, it will mistakenly assume that the trigger times of nodes B and C lag behind those of node A, thus incorrectly determining that the triggering events of the three are not synchronized, making collaborative observation impossible. To solve this problem, the system first acquires the seawater sound velocity profile data (sound velocity variation curves with depth, temperature, and salinity) at the location of each observation node. Combined with the previously calculated spatial distances between nodes, it uses ray tracing or equivalent average sound velocity methods to calculate the theoretical propagation delay of vibration / acoustic signals from the potential event source to each node. Subsequently, using the event occurrence time of the first triggered node as the time reference, the original trigger timestamps of each subsequent node are subtracted from the corresponding theoretical propagation delay. This allows the time coordinates of each node to be uniformly corrected to the "actual occurrence time of the signal," thereby effectively eliminating the interference of signal propagation delay on the trigger synchronization judgment and enabling the system to accurately identify which nodes' trigger signals originate from the same seabed geological event.

[0035] During operation, for nodes that have not triggered an event, the system queries the previously established node adjacency table to determine whether the spatial distance between the node and any node that has triggered an event is less than a preset collaborative distance threshold. It is important to emphasize that this collaborative distance threshold is not a fixed value, but rather dynamically determined based on the current event type: For seismic events, the elastic waves generated have a limited propagation distance and attenuate rapidly in the seabed strata, so the corresponding collaborative distance threshold is set to a smaller value (e.g., 300 meters), activating only nodes within a limited range near the seismic source for collaborative, intensive observation; for hydrothermal events, the accompanying temperature and pressure anomalies spread widely with ocean currents, so the corresponding collaborative distance threshold is set to a larger value (e.g., 800 meters) to cover a broader area affected by hydrothermal plumes; and for composite events, which possess both seismic and temperature / pressure anomaly characteristics, the collaborative distance threshold is a weighted sum of the two, ensuring that the collaborative range matches the actual physical impact area. If the distance between a non-triggered node and a triggered node is less than the dynamic cooperative distance threshold, the node is marked as a "cooperative node." This means that although it has not detected the event trigger itself, it is still within the event's influence range in space, and its sampling frequency needs to be increased to help capture the event's spatiotemporal evolution information. Finally, the system writes the original triggering state of each node, the synchronization triggering judgment result after time delay correction, and the cooperative marker into the decision matrix to form a complete distributed cooperative decision matrix.

[0036] S04. Based on the collaborative decision matrix, determine the target sampling frequency and data transmission mode for each observation node. For observation nodes that trigger events, increase their sampling frequency and enable real-time data backhaul. For observation nodes that do not trigger events but meet the spatial association conditions with the triggering node, increase their sampling frequency. For observation nodes that do not trigger events and do not meet the spatial association conditions, maintain the basic sampling frequency.

[0037] Furthermore, the spatial association condition in the above embodiments is determined based on a preset cooperative distance threshold. The cooperative distance threshold is dynamically determined according to the event type. Specifically, the cooperative distance threshold corresponding to a vibration event is less than the cooperative distance threshold corresponding to a hydrothermal event, and the cooperative distance threshold corresponding to a composite event is the weighted sum of the cooperative distance thresholds corresponding to the vibration event and the hydrothermal event. When the spatial distance between a node that has not triggered an event and a node that has triggered an event is less than the collaborative distance threshold, the spatial association condition is determined to be met. When the spatial distance between a node that has not triggered an event and all nodes that have triggered an event is not less than the collaborative distance threshold, the spatial association condition is determined not to be met.

[0038] Specifically, during the system initialization phase, based on the geological environment characteristics, sediment types, and historical event statistics of the area where the seabed observation network is located, the baseline cooperative distance thresholds Dv and Dt corresponding to seismic events are preset, respectively, with Dv < Dt. For example, in a typical seabed observation scenario, Dv can be set to 500 meters (effective radius of influence of seismic waves) and Dt to 3000 meters (effective radius of influence of hydrothermal diffusion). When the event type determined in step S02 is a pure seismic event, the system directly takes Dv as the current cooperative distance threshold; when the event type is a pure hydrothermal event, the system takes D_t as the current cooperative distance threshold. For composite events (seismic and hydrothermal events occur simultaneously, such as volcanic earthquakes accompanied by hydrothermal eruptions), the spatial influence range of the physical field is between the two. The system uses a weighted sum method to calculate the composite cooperative distance threshold Dc = α·Dv + β·Dt, where α and β are preset weighting coefficients, and α + β = 1. The values ​​of α and β can be dynamically adjusted based on the relative intensity of the seismic and hydrothermal components in the composite event: when the seismic component dominates (e.g., the magnitude of the seismic composite amplitude exceeding the standard is significantly higher than that of the temperature and pressure exceeding the standard), α is set to a larger value (e.g., 0.7) and β to a smaller value (e.g., 0.3), causing the composite threshold to bias towards the seismic threshold; conversely, when the hydrothermal component dominates, α is set to a smaller value and β to a larger value. Through this dynamic value selection strategy, the system can adaptively adjust the coordination range according to the actual physical characteristics of the event, achieving optimal allocation of observation resources.

[0039] Furthermore, the system has established a node adjacency table when constructing the collaborative decision matrix, which stores the spatial distance matrix between each observation node. When a node (denoted as node A) is determined to be a trigger event node, the system traverses all its adjacent nodes (denoted as node B) in the adjacency table, extracts the Euclidean spatial distance dAB between node A and node B one by one, and compares it with the collaborative distance threshold D determined in step S41 above. If dAB < D, then node B is determined to meet the spatial association condition, and node B is marked as a collaborative node, so the operation of "increasing the sampling frequency and enabling priority backhaul" is performed; if dAB ≥ D, then node B is determined not to meet the spatial association condition, and the basic sampling frequency is maintained. In the case of multiple trigger event nodes (e.g., multiple nodes simultaneously detect different stages of the same geological event), the system takes the union of the collaborative ranges of each trigger node, that is, any non-triggered node is marked as a collaborative node as long as the spatial distance between it and at least one trigger node is less than the corresponding collaborative distance threshold.

[0040] S05. Each observation node performs data acquisition according to the determined target sampling frequency and transmits data according to the determined data transmission mode.

[0041] Specifically, this embodiment configures differentiated data flow and transmission strategies based on different data transmission modes (real-time backhaul mode, priority backhaul mode, and local storage mode). In real-time backhaul mode, data from the triggering event node is assigned the highest transmission priority. After packaging, the data packet is immediately sent to the transmission queue of the underwater acoustic communication device and transmitted to the surface acoustic communication buoy via the underwater acoustic transducer. The buoy, through a communication relay device powered by a solar panel, forwards the data in real-time to the shore-based data control center via a satellite communication link. In this mode, to ensure the real-time performance of critical data, the system can be configured as a "pass-through mode," where data is not written to the local storage card but is directly sent to the communication transmission queue from the acquisition buffer, eliminating the additional latency introduced by storage writes. In priority backhaul mode, cooperating nodes that meet the spatial association conditions mark the acquired data as "high priority" and send it to the transmission queue, but its priority is lower than that of data packets in real-time backhaul mode. When the bandwidth of the underwater acoustic communication link is limited, the data in this mode can be appropriately delayed, but it takes priority over regular data. In local storage mode, nodes that have not triggered events and do not meet the coordination conditions will only write the collected data to the local large-capacity storage card, without occupying the underwater acoustic communication bandwidth. The data can be read uniformly when the equipment is retrieved after the observation task is completed, or it can be retransmitted at the lowest priority during the system's idle period.

[0042] S06. Collect action execution feedback data from each observation node, update feature parameters, evaluate the event state change through the event decay judgment mechanism, and restore the sampling frequency of each observation node to the basic sampling frequency when the event state is determined to have decayed.

[0043] Furthermore, in the above embodiments, the event state change is evaluated through an event decay determination mechanism. When the event state is determined to have decayed, it includes ( Figure 4 (as shown) S61. The exponentially weighted moving average of the characteristic parameters within a preset time window before the event is triggered is used as the baseline state, and the baseline state is dynamically updated with the duration of stable operation to eliminate the long-term drift effect of the sensor. S62. Calculate the weighted Euclidean distance between the current feature parameters and the baseline state as the degree of difference, where the weights of each feature parameter are dynamically allocated according to the event type. S63. Set the duration for fading judgment. When the degree of difference is lower than the event fading threshold for the first time, start the timer. Only after the cumulative duration for which the degree of difference is continuously lower than the event fading threshold exceeds the duration will the event be finally judged to have faded, in order to prevent frequent false recovery caused by noise fluctuations.

[0044] Specifically, before an event is triggered, the system continuously records a sequence of characteristic parameters within a preset time window (e.g., 60 seconds before triggering). This sequence is assigned an exponentially decaying weight, with data closer to the trigger time having a higher weight. A weighted average is then calculated as the initial baseline state. The advantage of EWMA (Extended Wake-up Mapping) is its ability to more sensitively reflect the latest environmental state before triggering, avoiding the dragging effect of outdated data on the baseline value. More importantly, after the system determines that the event has subsided and returns to the basic sampling frequency, it does not freeze the baseline state. Instead, as the subsequent stable operating time increases, new characteristic parameters are slowly integrated into the baseline state with a very small smoothing coefficient, achieving a rolling update of the baseline state. This dynamic update mechanism directly addresses the zero-point drift and sensitivity drift problems caused by the long-term immersion of deep-sea sensors in seawater.

[0045] Secondly, the system uses weighted Euclidean distance to quantify the deviation between the current state and the reference state. Let the feature parameter vector at the current moment be X=(x1,x2,x3,x4), corresponding to the vibration synthesis amplitude, vibration energy accumulation index, temperature change rate, and pressure change rate, respectively; the reference state vector is X0=(x0...). 1 ,x0 2 ,x0 3 ,x0 4 The system calculates the weighted Euclidean distance D = sqrt[λ1·(x1-x0)]. 1 )²+λ²·(x²-x0) 2 )²+λ3·(x3-x0 3 )²+λ4·(x4-x0 4)²], where λ1 to λ4 are the weight coefficients corresponding to each characteristic parameter. The weight allocation is not fixed, but dynamically adjusted according to the event type determined in step S02. When the event type is a vibration event, the vibration-related characteristic parameters (synthetic amplitude, energy index) are the dominant judgment criteria during the decay process. Therefore, λ1 and λ2 are assigned higher weights (e.g., 0.35 each), while the weights of temperature change rate and pressure change rate are relatively lower (0.15 each). When the event type is a hydrothermal event, temperature change rate and pressure change rate become dominant, and λ3 and λ4 are correspondingly increased (e.g., 0.35 each). For composite events, each weight is allocated proportionally according to the relative contribution of the vibration component and the hydrothermal component when the event is triggered, ensuring that the comprehensive judgment result can simultaneously reflect the decay process of both types of physical fields. The advantage of weighted Euclidean distance over simple arithmetic distance is that it can effectively avoid the judgment bias caused by the differences in the dimensions and numerical ranges of different characteristic parameters, and at the same time, it can accurately focus on the key parameters that are most representative of the current event type through weight allocation.

[0046] Finally, this embodiment introduces a debouncing mechanism to prevent frequent false recovery caused by environmental noise or transient disturbances. The specific implementation of this mechanism is as follows: the system presets a fading determination holding time T. h (e.g., T) h =10 seconds). After calculating the current difference level D, the system compares it with the preset event decay threshold Dth. When D first falls below Dth, the system does not immediately determine that the event has decayed, but instead starts a timer. In subsequent sampling iterations, the system continuously monitors the relationship between the difference level D and the event decay threshold Dth: if D rises back to Dth or above at any point during the timing process, it indicates that the event has not truly decayed or a secondary energy release has occurred, and the timer is immediately reset and monitoring resumes; only when the cumulative duration of D continuously falling below Dth strictly exceeds T... h Only after the required duration (i.e., after which the system makes a formal determination that the event has subsided) does it finally execute the operation to restore the sampling frequency of each observation node to the basic sampling frequency. The core principle of this mechanism is similar to a hardware de-jitter circuit in electronic engineering. Background noise in the seabed environment, transient interference from biological activity, and even the arrival of weak far-field seismic waves from nearby sea areas can all cause a temporary drop in characteristic parameters at a certain moment. If the sampling frequency is restored immediately without differentiation, the system will frequently "tug-of-war" at the edge of event subsidence, increasing computational overhead and potentially causing the loss of critical data due to re-triggering shortly after restoration. Duration T h The value of T can be flexibly configured according to the application scenario. For event types such as vibration events, where energy release is rapid and decay is steep, T... h It can be set to a shorter value (e.g., 5 seconds); for event types such as hydrothermal events where the temperature and pressure fields change slowly and decay gradually, Th It should be set to a relatively long value (such as 30 seconds) to ensure that the temperature and pressure fields are fully balanced in spatial distribution before recovery, so as to avoid premature exit from the encrypted observation state due to the instantaneous drop in local temperature and pressure.

[0047] Example 2

[0048] Based on the above embodiment one, the event fading determination mechanism indicated in this embodiment includes: A preset event decay threshold is set to determine whether an event has decayed by quantifying the difference between the current feature parameters and the baseline state before the event was triggered. in: If the difference exceeds the event decay threshold, maintain the current sampling frequency; If the difference does not exceed the event decay threshold, the event is determined to have decayed, and the sampling frequency is restored to the basic sampling frequency. The observation termination conditions include at least one of the following: The observation duration has reached the preset maximum observation duration; The difference between the feature parameters and the baseline state in multiple consecutive iterations remains below a preset stability threshold; Received a remote termination command from the shore-based control center.

[0049] Specifically, when an event is triggered, the system uses the average or exponentially weighted moving average of the characteristic parameters within a preset time window before the event as the baseline state. This baseline state represents the background environmental level before the event. In each subsequent iteration of sampling, the system calculates in real time the weighted Euclidean distance between the current temperature change rate, pressure change rate, and composite vibration amplitude, and the baseline state, quantifying this distance as a difference index. When the difference exceeds a preset event fading threshold, it means that the current environmental state still significantly deviates from the background level, and the system determines that the event is still ongoing, thus maintaining the current high-frequency sampling frequency. Conversely, when the difference gradually decreases and eventually does not exceed the event fading threshold, it indicates that the environmental state has returned to a normal level, and the system determines that the event has faded, gradually restoring the sampling frequency of each observation node to the base sampling frequency. To prevent false recovery caused by sudden noise spikes, the system also sets a duration for determining fading. Even if the difference level is lower than the threshold for the first time, the system will not immediately determine fading. Instead, it will start a timer and continuously monitor. Only when the cumulative duration of the difference level being lower than the threshold exceeds the duration of ...

[0050] Meanwhile, to address special scenarios such as prolonged event duration, sensor drift leading to baseline distortion, or the need for manual intervention, this embodiment also includes a multi-level observation termination condition configuration, complementing the event decay determination mechanism. The first type of termination condition is "the observation duration reaches the preset maximum observation duration." This condition serves as a safety mechanism to prevent the system from maintaining high-frequency sampling indefinitely under abnormal conditions. For example, during system initialization, the maximum observation duration can be set to 72 hours based on battery capacity and storage space. Once this time limit is reached, regardless of whether the event has completely subsided, the system will forcibly exit the event observation mode, ensuring the overall survivability of the platform. The second type of termination condition is "the difference between the feature parameters and the baseline state in multiple consecutive iterations remains below a preset stability threshold." This condition works in conjunction with the event decay determination mechanism: when the system repeatedly detects that the difference remains at an extremely low level (below a stability threshold that is more stringent than the decay threshold) within a relatively long time window, it indicates that the environment has not only recovered but is also in a highly stable state. At this point, the system terminates the observation cycle early, further saving energy. The third type of termination condition is "receiving a remote termination command from the shore-based control center." This condition grants shore-based operators the ability to intervene remotely. When a false alarm is detected, key data acquisition has been completed, or equipment malfunctions, shore-based personnel can send a termination command to underwater nodes via the satellite-underwater acoustic communication link. Upon receiving the command, each node immediately exits its current observation state and resumes its default operating mode. These three types of termination conditions cover the time, state, and manual intervention dimensions, respectively. Their priority order is as follows: remote termination command has the highest priority (immediate execution), followed by maximum observation duration (hard termination), and stability threshold condition has the lowest priority (soft early termination). Through the organic combination of the event decay judgment mechanism and multiple termination conditions, the system can promptly release system resources after an event ends, ensure its own safety during periods of anomaly, and respond quickly to manual intervention, thus achieving full-scenario coverage and high reliability assurance for the sampling strategy exit mechanism.

[0051] Example 3

[0052] Based on Embodiment 1, this embodiment determines the target sampling frequency and data transmission mode for each observation node, including: For the observation node that triggered the event, its sampling frequency is increased to the first sampling frequency, and the data transmission mode is set to real-time feedback mode; For observation nodes that do not trigger events but meet the spatial correlation conditions, their sampling frequency is increased to the second sampling frequency, and the data transmission mode is set to priority backhaul mode, where the second sampling frequency is lower than the first sampling frequency. For observation nodes that have not triggered events and do not meet the spatial correlation conditions, maintain the basic sampling frequency and set the data transmission mode to local storage mode; The data transmission mode and data compression strategy are linked: in real-time return mode, a first compression ratio is used for lightweight compression transmission; in priority return mode, a second compression ratio is used for balanced compression transmission; and in local storage mode, a third compression ratio is used for high-fidelity lossless storage, with the first compression ratio < the second compression ratio < the third compression ratio.

[0053] Furthermore, the numerical relationship between the first sampling frequency, the second sampling frequency, and the basic sampling frequency is as follows: first sampling frequency > second sampling frequency > basic sampling frequency, and the first sampling frequency is N times the basic sampling frequency, the second sampling frequency is M times the basic sampling frequency, where N > M > 1, and N and M are both positive integers. It also includes dynamically adjusting the values ​​of N and M based on the event trigger intensity. The event trigger intensity is calculated by weighting the over-amplitude of the vibration synthesis amplitude exceeding the dynamic threshold, the overscalar of the temperature change rate, and the pressure change rate. The higher the trigger intensity, the larger the values ​​of N and M.

[0054] Specifically, the system divides all nodes in the distributed observation network into three levels based on the judgment results of steps S02 and S03. The first level consists of triggering event nodes, which are nodes whose own characteristic parameters meet the event triggering conditions. These nodes are located in the core area where the event occurs and bear the highest value data acquisition tasks. Therefore, the system increases their sampling frequency to the highest first sampling frequency (denoted as f1) and enables real-time backhaul mode. That is, the data is sent directly to the surface buoy and forwarded to the shore via the underwater acoustic communication link without going through local storage buffer, ensuring that key data arrives at the analysis end with minimal delay. The second level consists of cooperating nodes, which are adjacent nodes that have not triggered the event themselves but meet the spatial association conditions with at least one triggering node. Although these nodes are not directly in the core of the event, their location may capture the attenuated signal propagating outward from the event, which has important supplementary value for analyzing the spatiotemporal evolution of the event. The system increases their sampling frequency to a medium second sampling frequency (denoted as f2) and sets it to priority backhaul mode. That is, the data is sent in a queue according to priority after a short local buffer. When communication resources are scarce, it takes precedence over regular data but is less than real-time backhaul data. The third level consists of ordinary nodes, which are the remaining nodes that neither trigger events nor meet the spatial correlation conditions. These nodes are far from the event's influence area, and the data they collect contributes little to the current event analysis. The system maintains its basic sampling frequency (denoted as f0) and sets it to local storage mode, meaning that the data is only stored in a local high-capacity storage card and does not actively initiate underwater acoustic communication transmission. It is only retrieved during routine system maintenance or data recycling, thereby maximizing the saving of valuable communication energy.

[0055] The three-level sampling frequencies described above satisfy a clear numerical progression relationship: f1 > f2 > f0, and f1 = f0 × N, f2 = f0 × M, where N and M are both positive integers greater than 1 and N > M. This numerical relationship ensures that a distinct sampling gradient is formed between the three levels of nodes, rather than a simple binary switching, enabling the system to smoothly adapt to the information attenuation gradient from the core area of ​​the event to the peripheral area. More importantly, N and M are not fixed constants, but are adaptively adjusted according to the dynamic calculation results of the event trigger intensity. The calculation logic of the event trigger intensity is as follows: the system obtains the overamplitude ΔA (denoted as δA after normalization) of the vibration synthesis amplitude exceeding the dynamic threshold, the overscalar ΔT (denoted as δT after normalization) of the temperature change rate exceeding the temperature threshold, and the overscalar ΔP (denoted as δP after normalization) of the pressure change rate exceeding the pressure threshold, respectively. Then, the system performs a weighted summation using preset weighting coefficients to obtain the comprehensive trigger intensity index I = w A ·δ A +w T ·δ T +w P ·δ P , where w A w T w P The system uses preset weighting coefficients, with the sum of the three factors being 1. The trigger intensity I is mapped to a continuous interval between 0 and 1, with I approaching 1 indicating a stronger event. The system has multiple preset intensity level intervals: for example, 0 ≤ I < 0.3 indicates a low-intensity event (N=4, M=2); 0.3 ≤ I < 0.7 indicates a medium-intensity event (N=8, M=4); and 0.7 ≤ I ≤ 1.0 indicates a high-intensity event (N=16, M=8). Through this dynamic mapping mechanism, when a strong earthquake event is triggered, the system automatically increases the sampling frequency of the core nodes to 16 times the base frequency (e.g., from 10Hz to 160Hz), while weak disturbances are only increased to 4 times (40Hz), achieving a match between sampling resources and event intensity.

[0056] Example 4

[0057] Based on the above embodiment 1, this example also includes node failure handling. When any observation node fails to return valid action execution feedback data in a consecutive preset number of iterations, the node is marked as a failed node. By using the time difference of arrival of signals between multiple effective nodes adjacent to the failed node, spatial interpolation and reconstruction of vibration and temperature / pressure signals at the location of the failed node are performed to generate virtual node observation data, and the virtual node is temporarily incorporated into the collaborative decision matrix. When a failed node resumes normal communication, the virtual node is removed and the original node's collaborative decision matrix configuration is restored. If the failed node continues to fail to recover, it is permanently removed from the collaborative decision matrix, and the collaborative decision matrix of the remaining nodes is reconstructed.

[0058] Specifically, during normal operation, each observation node returns action execution feedback data, including data acquisition status, equipment self-test information, and timestamps, to the master node or surface buoy after each sampling iteration. In each iteration, the system monitors whether all nodes return valid feedback data within a preset timeout period. If a node fails to return valid feedback data for a preset number of consecutive iterations (e.g., 5 consecutive iterations), the system determines that the node is in a failed state, marks it as a failed node, and temporarily removes it from the current collaborative decision matrix to prevent it from being included in invalid calculations in subsequent decision-making processes. This continuous counting determination strategy effectively avoids misjudgments caused by transient communication interference, improving the accuracy and robustness of failure determination.

[0059] When a node (denoted as node F) is marked as failed, the system first extracts multiple valid nodes adjacent to node F from the node adjacency table (usually four or more are selected to ensure interpolation accuracy). Then, the system obtains the arrival times of the vibration signals and the temperature and pressure data change curves recorded by these adjacent valid nodes within the same event response period. By comparing the signal arrival time differences between adjacent nodes, and combining the precise spatial coordinates of each node with the pre-measured seawater sound velocity profile and seabed sediment sound velocity parameters of the area, the system uses the inverse distance weighting method or Kriging interpolation method to reverse-calculate the theoretical arrival times and corresponding signal amplitudes of the vibration wavefront and the temperature and pressure anomaly front at the location of the failed node F. For example, suppose that in a certain vibration event, node E, located 400 meters east of node F, records the arrival time of the seismic wave at t=10.00s with a peak acceleration of 0.12g; node W, located 400 meters west, records at t=10.15s with a peak acceleration of 0.08g; node S, located 400 meters south, records at t=10.12s with a peak acceleration of 0.10g; and node N, located 400 meters north, records at t=10.08s with a peak acceleration of 0.11g. Based on the spatial coordinates of each node and the signal propagation velocity model, the system calculates, through weighted interpolation, that the theoretical arrival time at node F is approximately t=10.09s and the peak acceleration is approximately 0.10g. Similarly, for temperature and pressure data, the system also uses the temperature and pressure change curves of adjacent nodes for spatial interpolation reconstruction to generate virtual temperature and pressure values ​​at the failed node. After the above processing, the system temporarily incorporates the virtual node and its reconstructed data into the collaborative decision matrix. In subsequent collaborative triggering judgments and sampling frequency allocations, the virtual node enjoys the same decision weight as the real node. This means that when the calculated feature parameters of the virtual node's location meet the event triggering conditions, other real nodes in its adjacent area will also initiate collaborative sampling due to the "trigger" of the virtual node, thereby maintaining the spatial continuity of the event perception area.

[0060] When a failed node resumes normal communication, the system senses the node's recovery status by receiving valid action execution feedback data again. It then automatically removes the corresponding virtual node, restores the original real node to the collaborative decision matrix, and re-establishes its adjacency relationships with neighboring nodes. If a failed node fails to resume communication within a preset long-term window (e.g., 72 consecutive hours), the system determines that the node is permanently failed, permanently removes it from the collaborative decision matrix, and reconstructs the node adjacency table and collaborative decision matrix based on the spatial distribution of the remaining valid nodes. In the case of permanent removal, the system appropriately expands the collaborative distance threshold between neighboring nodes that previously relied on the failed node for spatial interpolation to compensate for the decrease in local observation capability caused by permanent removal. For example, in an area densely populated with hexagonal honeycomb-like observation nodes, after removing the central node, the collaborative distance threshold between its six neighboring nodes expands outward by approximately 10%, allowing the event collaboration range that originally required the central node to be covered by the re-establishment of pairwise relationships between the peripheral nodes.

[0061] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An adaptive sampling method for a distributed seabed observation network based on event triggering, characterized in that, The method includes the following steps: S01. Collect multimodal sensing data from multiple observation nodes in a distributed seabed observation network. The multimodal sensing data includes temperature data, pressure data, and triaxial vibration data. Preprocess the multimodal sensing data of each observation node and extract the characteristic parameters corresponding to each observation node. S02. Perform anomaly detection on the characteristic parameters of each observation node to determine whether the event triggering conditions are met. If the event triggering conditions are met, generate an event triggering signal and record the event type. The event triggering conditions include at least one or more combinations of vibration signal amplitude exceeding the threshold, temperature change rate exceeding the threshold, and pressure change rate exceeding the threshold. S03. Based on the event triggering signals and event types of each observation node, a distributed collaborative decision matrix is ​​constructed. The collaborative decision matrix is ​​used to describe the triggering state of each observation node and the spatial correlation between nodes. S04. Based on the collaborative decision matrix, determine the target sampling frequency and data transmission mode of each observation node. For observation nodes that trigger events, increase their sampling frequency and enable real-time data backhaul. For observation nodes that do not trigger events but meet the spatial association conditions with the triggering node, increase their sampling frequency. For observation nodes that do not trigger events and do not meet the spatial association conditions, maintain the basic sampling frequency. S05. Each observation node performs data acquisition according to the determined target sampling frequency and transmits data according to the determined data transmission mode. S06. Collect action execution feedback data from each observation node, update feature parameters, evaluate the event state change through the event decay judgment mechanism, and restore the sampling frequency of each observation node to the basic sampling frequency when the event state is determined to have decayed.

2. The adaptive sampling method for a distributed seabed observation network based on event triggering according to claim 1, characterized in that, The event decay determination mechanism includes: A preset event decay threshold is set to determine whether an event has decayed by quantifying the difference between the current feature parameters and the baseline state before the event was triggered. in: If the degree of difference exceeds the event decay threshold, maintain the current sampling frequency; If the difference does not exceed the event fading threshold, the event is determined to have faded, and the sampling frequency is restored to the basic sampling frequency. The observation termination condition includes at least one of the following: The observation duration has reached the preset maximum observation duration; The difference between the feature parameters and the baseline state in multiple consecutive iterations remains below a preset stability threshold; Received a remote termination command from the shore-based control center.

3. The adaptive sampling method for a distributed seabed observation network based on event triggering according to claim 1, characterized in that, In step S01, the multimodal sensing data from each observation node are preprocessed to extract characteristic parameters reflecting the current state of the seabed environment, including: Temperature data, pressure data, and triaxial vibration data are filtered and denoised to eliminate environmental noise and sensor-inherent noise. Vector synthesis is performed on the filtered triaxial vibration data to calculate the root mean square value of the vibration synthesis amplitude and the vibration energy accumulation index. The step size of the sliding time window is dynamically adjusted according to the changing trend of the current pressure data. The step size is shortened during the pressure rise phase to improve the response sensitivity, and the step size is increased during the pressure stabilization phase to reduce the computational load. Calculate the rate of temperature change and the rate of pressure change based on a sliding time window; The vibration synthesis amplitude, vibration energy index, temperature change rate, and pressure change rate are used as the characteristic parameters.

4. The adaptive sampling method for a distributed seabed observation network based on event triggering according to claim 1, characterized in that, In S02, determining whether the event triggering conditions are met requires pre-setting basic thresholds for vibration amplitude, temperature change rate, and pressure change rate, and using the standard deviation of characteristic parameters within a long time window preset by each observation node before the event triggering as a dynamic adjustment coefficient to generate an adaptive triggering threshold. When the root mean square value of the vibration synthesis amplitude of any observation node exceeds the dynamic threshold of vibration amplitude in the adaptive triggering threshold, and the vibration energy accumulation index maintains a monotonically increasing trend within a continuous time window, the event triggering condition is determined to be met, and the event type is recorded as a vibration event. When the rate of temperature change exceeds the basic threshold for both the rate of temperature change and the rate of pressure change exceeds the basic threshold for both the rate of pressure change, the event triggering condition is determined to be met, and the event type is recorded as a hydrothermal event. When the root mean square value of the composite vibration amplitude exceeds the dynamic threshold of the vibration amplitude, and at least one of the temperature change rate and pressure change rate exceeds its corresponding basic threshold, the event triggering condition is determined to be met, and the event type is recorded as a composite event.

5. The adaptive sampling method for a distributed seabed observation network based on event triggering according to claim 1, characterized in that, The construction of the distributed collaborative decision-making matrix in S03 includes: S31. Construct an initial decision matrix with observation nodes as rows and event types as columns. Each matrix element represents the triggering state of the corresponding node under the current event type. S32. Based on the spatial location information of each observation node, calculate the spatial distance between nodes and establish a node adjacency table; S33. Introduce a signal delay compensation factor. Based on the seawater sound velocity profile at the location of each observation node and the spatial distance between nodes, calculate the theoretical propagation delay of the vibration / acoustic signal between different nodes, and perform delay correction on the trigger timestamp of each node. S34. For nodes that have not triggered an event, determine whether the spatial distance between the node and any node that triggered an event is less than a preset cooperative distance threshold according to the node adjacency table. If so, mark the node as a cooperative node. S35. Write the trigger status and collaborative flag of each node into the collaborative decision matrix.

6. The adaptive sampling method for a distributed seabed observation network based on event triggering according to claim 1, characterized in that, The determination of the spatial association condition in step S04 is based on a preset cooperative distance threshold. The cooperative distance threshold is dynamically determined according to the event type. The cooperative distance threshold corresponding to the vibration event is less than the cooperative distance threshold corresponding to the hydrothermal event. The cooperative distance threshold corresponding to the composite event is the weighted sum of the cooperative distance thresholds corresponding to the vibration event and the hydrothermal event. When the spatial distance between a node that has not triggered an event and a node that has triggered an event is less than the cooperative distance threshold, the spatial association condition is determined to be met. When the spatial distance between a node that has not triggered an event and all nodes that have triggered an event is not less than the cooperative distance threshold, it is determined that the spatial association condition is not met.

7. The adaptive sampling method for a distributed seabed observation network based on event triggering according to claim 1, characterized in that, The step S04, which involves determining the target sampling frequency and data transmission mode for each observation node, includes: For the observation node that triggered the event, its sampling frequency is increased to the first sampling frequency, and the data transmission mode is set to real-time feedback mode; For observation nodes that do not trigger events but meet spatial correlation conditions, their sampling frequency is increased to the second sampling frequency, and the data transmission mode is set to priority backhaul mode, wherein the second sampling frequency is lower than the first sampling frequency; For observation nodes that have not triggered events and do not meet the spatial correlation conditions, maintain the basic sampling frequency and set the data transmission mode to local storage mode; The data transmission mode is linked to the data compression strategy: in real-time return mode, a first compression ratio is used for lightweight compression transmission; in priority return mode, a second compression ratio is used for balanced compression transmission; and in local storage mode, a third compression ratio is used for high-fidelity lossless storage, wherein the first compression ratio < the second compression ratio < the third compression ratio.

8. The adaptive sampling method for a distributed seabed observation network based on event triggering according to claim 7, characterized in that, The numerical relationship between the first sampling frequency, the second sampling frequency, and the basic sampling frequency is as follows: first sampling frequency > second sampling frequency > basic sampling frequency, and the first sampling frequency is N times the basic sampling frequency, and the second sampling frequency is M times the basic sampling frequency, where N > M > 1, and N and M are both positive integers. It also includes dynamically adjusting the values ​​of N and M based on the event trigger intensity. The event trigger intensity is calculated by weighting the over-amplitude of the vibration synthesis amplitude exceeding the dynamic threshold, the overscalar of the temperature change rate, and the pressure change rate. The higher the trigger intensity, the larger the values ​​of N and M.

9. The adaptive sampling method for a distributed seabed observation network based on event triggering according to claim 1, characterized in that, The event state change assessment mechanism described in S06, whereby the event state is determined to have faded, includes: S61. The exponentially weighted moving average of the feature parameters within a preset time window before the event is triggered is used as the reference state, and the reference state is dynamically updated with the duration of stable operation to eliminate the long-term drift effect of the sensor. S62. Calculate the weighted Euclidean distance between the current feature parameter and the baseline state as the degree of difference, wherein the weight of each feature parameter is dynamically allocated according to the event type; S63. Set the fading determination holding time. When the difference level is lower than the event fading threshold for the first time, start a timer. Only after the cumulative duration of the difference level being lower than the event fading threshold exceeds the holding time will the event be finally determined to have faded, so as to prevent frequent false recovery caused by noise fluctuations.

10. The adaptive sampling method for a distributed seabed observation network based on event triggering according to claim 1, characterized in that, The method also includes node failure handling, whereby a node is marked as a failed node when any observation node fails to return valid action execution feedback data in a preset number of consecutive iterations. By using the time difference of arrival of signals between multiple effective nodes adjacent to the failed node, spatial interpolation and reconstruction of vibration and temperature / pressure signals at the location of the failed node are performed to generate virtual node observation data, and the virtual node is temporarily incorporated into the collaborative decision matrix. When a failed node resumes normal communication, the virtual node is removed and the original node's collaborative decision matrix configuration is restored. If the failed node continues to fail to recover, it is permanently removed from the collaborative decision matrix, and the collaborative decision matrix of the remaining nodes is reconstructed.

Images