A multi-dimensional data integration processing and early warning system for a breeding environment

Abstract

The application discloses a kind of breeding environment multidimensional data integration processing and early warning system.The system includes: the main control computing unit deployed in factory local, multiple sensor anchor point clusters distributed in breeding pool interior and data aggregation gateway;Each sensor anchor point cluster contains basic physicochemical probe component, environmental flow field component and biological observation component;The main control computing unit calculates physical transfer time delay according to three-dimensional flow velocity vector and aligns time series data using dynamic time warping algorithm, and the adjacency matrix between the nodes of internal model is dynamically calculated according to three-dimensional flow velocity vector to output environmental anomaly probability score;When the fluctuation rate of physicochemical parameter or environmental anomaly probability score meets the condition, data aggregation gateway sends hardware interrupt instruction to dormant biological observation component.This application can accurately compensate material propagation data time delay, accurately identify abnormalities, and reduce energy consumption using cross-modal wake-up mechanism, to ensure system space-time causal closed loop and bandwidth communication efficiency.

Description

Technical Field

[0001] This invention relates to the fields of industrial Internet of Things and aquaculture data processing, specifically to a multi-dimensional data integration, processing, and early warning system for aquaculture environment applied to high-density recirculating aquaculture (RAS) systems or large racetrack-style freshwater aquaculture ponds. Background Technology

[0002] In high-density factory aquaculture, circulating water pumps and aeration equipment artificially create complex microscopic hydrodynamic fields. Existing aquaculture monitoring systems typically employ distributed sensor arrays, connecting monitoring devices to the water surface and using hard-coded rules at the edge or cloud for early warning monitoring. However, such conventional technologies face significant technical bottlenecks in this specific physical scenario. Traditional solutions, by ignoring the physical time delay of material diffusion caused by water circulation, are prone to delayed alarms or false alarms during localized hypoxia or ammonia nitrogen surges, leading to frequent start-ups and shutdowns of wave-generating and aeration equipment, increasing energy costs at the aquaculture base. Regarding collaborative monitoring of multimodal data, high-density aquaculture requires 24 / 7 visual monitoring, placing enormous concurrent pressure on the edge computing power and local area network bandwidth of the facility. Furthermore, the data streams collected by various modal sensors suffer from severe transmission delays and timestamp asynchrony in complex transmission links. Directly splicing data in the backend system disrupts the temporal causal relationship between spatial physical hydrological field anomalies and fish biological behavioral stress, preventing early warning models from accurately extracting the spatiotemporal characteristics of disaster evolution. Currently, there is an urgent need for a system solution that can perform time delay compensation based on physical flow fields and achieve precise wake-up of software and hardware across modes. Summary of the Invention

[0003] This invention solves the technical problems of high false alarm rate and limited bandwidth in a multi-dimensional data integration processing and early warning system for aquaculture environment, and provides a multi-dimensional data integration processing and early warning system for aquaculture environment.

[0004] This application provides a multi-dimensional data integration, processing, and early warning system for aquaculture environments, including a main control computing unit deployed locally in the factory, multiple sensor anchor clusters distributed within the aquaculture ponds, and a data aggregation gateway. Each sensor anchor cluster includes a basic physicochemical probe component for collecting physicochemical parameters, an environmental flow field component, and a biological observation component. The environmental flow field component is used to collect the three-dimensional velocity vector of the flow field. The main control computing unit calculates the physical transmission time delay of fluid between different sensor anchor clusters based on the three-dimensional velocity vector and uses a dynamic time warping algorithm to align the time-series data collected by the sensor anchor clusters. A spatiotemporal graph convolutional network early warning model runs within the main control computing unit. The adjacency matrix between nodes of this spatiotemporal graph convolutional network early warning model performs dynamic fluid dynamic correlation calculation based on the three-dimensional velocity vector and outputs an environmental anomaly probability score. The biological observation component is in a dormant mode under basic operating conditions. When the fluctuation rate of the physicochemical parameters collected by the basic physicochemical probe component meets preset conditions, or when the environmental anomaly probability score exceeds a set threshold, the data aggregation gateway sends a hardware interrupt command to the biological observation component, triggering the biological observation component to enter working mode and collect images of fish behavior in the aquaculture pond.

[0005] In one specific implementation, the main control computing unit is used to select an upstream first sensor anchor cluster and a downstream second sensor anchor cluster from multiple sensor anchor clusters, extract the spatial distance vector between the first sensor anchor cluster and the second sensor anchor cluster, and calculate the average water flow velocity vector based on the three-dimensional flow velocity vector collected by the environmental flow field components of the first sensor anchor cluster and the second sensor anchor cluster respectively. The modulus of the spatial distance vector is divided by the projected velocity value of the average water flow velocity vector on the spatial distance vector to obtain the physical transmission time delay from the first sensor anchor cluster to the second sensor anchor cluster.

[0006] In one specific implementation, when the main control computing unit aligns the time-series data collected by the sensor anchor cluster using a dynamic time warping algorithm, a penalty constraint matrix is ​​configured. The initialization parameters of this penalty constraint matrix are dynamically assigned based on the physical transmission time delay. The main control computing unit constructs cost matrices for two time-series data to be aligned. When searching for the path with the minimum cumulative distance in the cost matrices, the size of the curved window limiting the search range is the product of the physical transmission time delay and a preset time tolerance parameter. A nonlinear penalty weight is applied to the coordinates of the path with the minimum cumulative distance that deviate from the diagonal, based on the magnitude of the physical transmission time delay. This nonlinear penalty weight suppresses excessive deformation of the time series during the warping operation, outputting an aligned time-series sequence set compensated for the fluid propagation delay.

[0007] In one specific implementation, when the main control computing unit performs dynamic fluid dynamic correlation calculation, it obtains the physical Euclidean distance and three-dimensional velocity vector between the sensor anchor point clusters corresponding to any two nodes, calculates the spatial water flow angle between the water flow direction and the direction of the line connecting the two nodes, and calculates the node dynamic adjacency weight based on the physical Euclidean distance, the magnitude of the three-dimensional velocity vector and the spatial water flow angle. The node dynamic adjacency weight is used to construct the adjacency matrix between nodes.

[0008] In one specific implementation, the main control computing unit is configured with computational logic to calculate the cosine value of the spatial water flow angle and determine the causal relationship of water flow between nodes based on this cosine value. When the cosine value is greater than zero, a direct water flow-driven association is determined. The main control computing unit uses the ratio of the physical Euclidean distance to the magnitude of the three-dimensional velocity vector as a characteristic time parameter, performs a nonlinear mapping on this characteristic time parameter using an exponential decay function, and multiplies the mapped value with the cosine value to generate the dynamic adjacency weight of the nodes. When the cosine value is less than or equal to zero, the main control computing unit forcibly assigns the corresponding dynamic adjacency weight of the nodes to the minimum threshold value.

[0009] In one specific implementation, the data aggregation gateway includes an edge computing module and a hardware interrupt trigger pin. The edge computing module receives dissolved oxygen concentration sequences and ammonia nitrogen concentration sequences acquired in real time by the basic physicochemical probe components, calculates the first derivative values ​​of the dissolved oxygen concentration sequences and ammonia nitrogen concentration sequences based on a preset sliding time window, and uses these first derivative values ​​as the fluctuation rate of the physicochemical parameters.

[0010] In one specific implementation, the data aggregation gateway is configured with an adaptive threshold comparison module. This module dynamically adjusts the rate of change threshold in preset conditions based on the daily average fluctuation benchmark data of water physicochemical parameters over historical periods. When the absolute value of the first derivative exceeds the rate of change threshold for multiple consecutive sampling periods, the edge computing module generates a high-level interrupt signal and sends it to the power management controller of the biological observation component via the underlying system bus through a hardware interrupt trigger pin. In response to the high-level interrupt signal, the power management controller directly closes the power supply circuit of the biological observation component, controlling it to switch from sleep mode to working mode. The acquired images of fish behavior in the aquaculture pond are packaged into burst data packets and uploaded to the main control computing unit via a short-range wireless communication link.

[0011] In one specific implementation, the basic physicochemical probe assembly includes a dissolved oxygen sensor, an ammonia nitrogen sensor, and a water temperature sensor. The environmental flow field assembly includes a high-frequency miniature Doppler current meter or an ultrasonic flow meter array. The bioobservation assembly includes an underwater high-definition camera and an auxiliary illumination source. The basic physicochemical probe assembly and the environmental flow field assembly are configured to maintain a periodic communication connection with a data aggregation gateway via a low-power wide-area network protocol.

[0012] In one specific implementation, the spatiotemporal graph convolutional network early warning model includes spatial graph convolutional layers and temporal convolutional layers. The spatial graph convolutional layer uses the adjacency matrix between nodes to perform spatial dimension aggregation on the features of each sensor anchor cluster. The temporal convolutional layer is used to perform one-dimensional convolution operations on the features after spatial aggregation to extract temporal variation features, and outputs an environmental anomaly probability score through a fully connected layer.

[0013] In one specific implementation, the main control computing unit is also equipped with a network-wide clock synchronization module. This network-wide clock synchronization module is configured to send a unified timestamp to all sensor anchor point clusters using a precise time protocol, and perform hardware-level basic time synchronization and alignment operations on the timing data based on the unified timestamp.

[0014] This application combines the physical laws of fluid mechanics to accurately compensate for data delays caused by material propagation, enabling downstream early warning models to accurately distinguish between regional persistent hypoxia disasters and invalid anomalies caused by short-term fluctuations in the ordinary water cycle, thereby improving the anti-interference capability and accuracy of the early warning system. By using the fluctuation rate of basic physicochemical parameters as a cross-modal wake-up switch at the hardware level, the contradiction between local area network bandwidth bottlenecks and all-time monitoring is resolved, ensuring the spatiotemporal causal closed loop of the underlying physical environment and biological state. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the application environment of the multi-dimensional data integration processing and early warning system for aquaculture environment provided in this embodiment of the invention.

[0016] Figure 2 This is a schematic diagram of the system structure provided in an embodiment of the present invention.

[0017] Figure 3 This is a schematic diagram of the data processing flow provided in an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. The specific embodiments described herein are merely illustrative and are not intended to limit the scope of the invention.

[0019] Example 1

[0020] Figure 1 and Figure 2As shown in the summary, this embodiment provides a multi-dimensional data integration processing and early warning system for aquaculture environments. In terms of physical topology, it includes a main control computing unit 101 deployed on a water surface platform, sensor anchor clusters 102 distributed at multiple depths within the aquaculture pond, and a data aggregation gateway 103 interconnected by communication. In the actual three-dimensional space of aquaculture, the sensor anchor clusters 102 are arranged in layers according to the spatial geometry of the aquaculture pond, specifically set along the central axis of the pond and on the upstream and downstream surfaces, forming a three-dimensional detection array capable of capturing both advection and vertical turbulence of water masses from all directions.

[0021] On the water platform side, the main control computing unit 101 is equipped with a core processing server cluster and is externally connected to a microwave communication antenna 104. This microwave communication terminal or Wi-Fi 6 wireless base station is equipped with a high anti-interference radio frequency circuit to maintain a broadband link with the plant's central control center in complex electromagnetic environments with high humidity and high-power electromechanical equipment operation, ensuring that the fused high-value early warning data can be stably transmitted back without being affected by the complex electromagnetic interference of the plant.

[0022] On the underwater detection side, each sensor anchor cluster 102 includes a basic physicochemical probe assembly 201 for collecting the physicochemical characteristics of the water, an environmental flow field assembly 202 for mapping the three-dimensional flow field characteristics, and a bio-observation assembly 203 for tracking fish community characteristics. The basic physicochemical probe assembly 201 is modularly packaged, specifically containing an integrated detection unit for detecting temperature, salinity, depth, and polarographic dissolved oxygen. The polarographic dissolved oxygen probe contacts the pool water through a specially designed breathable and water-permeable membrane, maintaining high sensitivity even in environments with high-density organic suspended matter and bioflocs. The environmental flow field assembly 202 is equipped with a high-frequency miniature Doppler current meter (operating at frequencies above 2MHz). High-frequency sound waves are used because they have a very small blind zone and can effectively reduce acoustic multipath interference caused by reflections from the pool walls within a limited water area, extracting the motion Doppler frequency shift of local micro-particles within the pool with high fidelity. The bio-observation assembly 203 includes an underwater low-light high-definition camera and a high-frequency imaging sonar, and is equipped with an adaptive auxiliary lighting source.

[0023] In terms of communication connectivity and power supply infrastructure, each sensor anchor cluster 102, suspended or fixed at a specific depth, is connected to the data aggregation gateway 103 located within the main control computing unit 101 via an underwater armored umbilical cable 105. To withstand repeated impacts and shearing damage to the cable from the high-power circulating water pump and jet jet flow, the underwater armored umbilical cable 105 is not only wrapped with tensile-resistant Kevlar fiber and stainless steel braided mesh, but also internally incorporates pure copper conductor cores for power transmission and fiber optic / twisted-pair cores for high-speed digital signal transmission. At the industrial shielded twisted-pair interface between the underwater anchor cluster and the main control computing unit, a power supply circuit module with a surge protection device is specially designed and connected. This module uses multi-stage LC filtering and a transient voltage suppression diode array to filter out surge voltages caused by parasitic inductance and equipment start-up and shutdown during long-distance underwater power transmission, providing a clean and stable DC power supply for the high-precision detection unit.

[0024] The data aggregation gateway 103 serves as the neural hub for underlying sensing. It integrates a neural processing unit with an ARM architecture and runs a trimmed and optimized embedded real-time operating system. This unit is responsible for local caching of sensor data streams, low-level protocol parsing, timestamp binding, and cross-device wake-up scheduling control. To address the asynchronous issue caused by temperature drift of different crystal oscillators among distributed nodes in multimodal data processing, a network-wide clock synchronization module 206 is installed within the main control computing unit 101. All sensors are connected to a unified timing server for reference clock synchronization before leaving the factory and before deployment in water. During operation, the network-wide clock synchronization module 206 is configured to periodically send a unified timestamp tsync to all sensor anchor clusters 102 via an industrial shielded twisted-pair network using a precise time protocol. Based on this unified timestamp tsync, the system performs hardware-level basic time synchronization alignment operations on the data frames of the underlying registers of each sensor hardware module, ensuring that the first frame sampling points of physicochemical, flow field, and biological data are absolutely aligned on the nanosecond-level physical time axis. Those skilled in the art can also achieve the underlying time alignment purpose of this application by using the NTP protocol or the Beidou satellite timing module with equivalent accuracy.

[0025] Example 2

[0026] Figure 3 As shown, during system operation, data acquisition is performed through data acquisition S301. The environmental flow field component 202 collects the three-dimensional flow velocity vector v of the flow field, and the basic physicochemical probe component 201 continuously collects basic time-series data such as dissolved oxygen and ammonia nitrogen at a sampling frequency of 1Hz. Large racetrack-style freshwater aquaculture ponds can span tens or even hundreds of meters. Sudden changes in water quality at a certain water layer are not instantaneously reflected on all sensors, but depend on the physical advection transport of water flow.

[0027] In the spatiotemporal delay compensation calculation S302, the main control calculation unit 101 performs the spatiotemporal delay compensation calculation. The main control calculation unit 101 calculates the physical transmission time delay Tdelay of the fluid between different sensor anchor clusters 102 based on the three-dimensional velocity vector v output by the high-frequency micro Doppler velocimeter or ultrasonic flowmeter array. The calculation logic is as follows: the main control calculation unit 101 selects a first sensor anchor cluster located upstream and a second sensor anchor cluster located downstream from the multiple sensor anchor clusters 102, and extracts the spatial distance vector L between the first and second sensor anchor clusters based on the three-dimensional spatial coordinates. The system reads the three-dimensional velocity vector v collected by the environmental flow field components 202 corresponding to the two nodes, calculates their geometric average, and calculates the average water flow velocity vector vˉ representing the uniform state of the flow field between the two points. Divide the magnitude of the spatial distance vector L by the projected velocity value of the average water flow velocity vector vˉ along the axis of the spatial distance vector L to obtain the physical transmission time delay Tdelay required for the material to propagate from the first sensor anchor cluster to the second sensor anchor cluster.

[0028] In the dynamic time warping alignment S303, which includes physical delay alignment, the data aggregation gateway 103 utilizes a dynamic time warping algorithm that includes physical delay alignment to perform deep alignment on multidimensional time series containing physical delays. The system performs dynamic time warping operations that include physical delay alignment:

[0029] D(i,j)=min{D(i−1,j),D(i,j−1),D(i−1,j−1)}+dval(xi,yj−Tdelay)

[0030] Here, variable D(i,j) represents the total warping cost distance accumulated up to sequence indices i and j during the alignment process; variable min is a minimum-value function used to find the optimal deformation path that minimizes the accumulated error; variable dval(xi,yj−Tdelay) represents the property distance of the corresponding time-time physicochemical parameters between the sampled value xi of the first time series at index i and the sampled value yj−Tdelay of the second time series at index (j−Tdelay) after compensating for the physical transmission time delay Tdelay. By nesting and fusing the mathematically defined dynamic time warping algorithm distance metric mechanism with the fluid dynamics-based time delay parameter Tdelay, the system can directly offset the comparison reference time axis when searching for the minimum cost path, enabling the accurate pairing of data mutation peaks with physical causal correlations.

[0031] To prevent excessive timeline distortion in traditional dynamic time warping algorithms when searching for the optimal path, the main control computing unit 101 is equipped with a penalty constraint matrix M. The initialization parameters of the penalty constraint matrix M are dynamically assigned based on the calculated physical propagation time delay Tdelay. After constructing the cost matrix C of the time series data to be aligned, the main control computing unit 101, when searching for the minimum cumulative distance path W in the cost matrix C, forces the Sakoe-Chiba bending window size, which limits the algorithm's search range, to the product of the physical propagation time delay Tdelay and a preset time tolerance parameter α. Based on the magnitude of the physical propagation time delay Tdelay, a nonlinear penalty weight wp is applied to the coordinates of the minimum cumulative distance path W that deviates from the center diagonal. Through this nonlinear penalty weight wp, the system effectively suppresses excessive distortion of the time series during warping operations, ultimately outputting an aligned time series set that strictly adheres to the causal transmission laws of matter after compensating for fluid propagation time delay. By employing a dynamic time warping mechanism with physical time delay constraints, the reasonable boundaries of data matching can be defined using hydrodynamic laws, avoiding spurious correlation data alignment caused by water cycle turbulence.

[0032] Example 3

[0033] After completing the spatiotemporal alignment of the underlying data, the system enters the early warning and judgment phase S304. A pre-trained spatiotemporal graph convolutional network early warning model runs in the memory space of the main control computing unit 101. This network model includes spatial graph convolutional layers and temporal convolutional layers. Underwater dissolved oxygen and ammonia nitrogen are not independently distributed point data, but rather a topological surface that continuously evolves with the hydrodynamic field. The spatial graph convolutional layer is configured to use the adjacency matrix between nodes to perform spatial dimension aggregation operations on the multivariate features collected by each sensor anchor cluster 102, extracting the impact of the surrounding water area. The temporal convolutional layer uses a one-dimensional dilated convolution kernel to perform temporal dimension scanning operations on the feature map after spatial aggregation to extract temporal evolution features. The fully connected layer and the Softmax activation function output an environmental anomaly probability score P, representing the current health status of the aquaculture pond environment.

[0034] In traditional graph models, the adjacency matrix is ​​often the reciprocal of a fixed physical distance. This may be effective in still water environments, but in dynamic water environments, the mutual influence between two closely spaced nodes located upstream or downstream is far greater than that between two nodes of equal distance but connected by a line perpendicular to the flow direction. When performing dynamic fluid dynamics correlation calculations, the main control computing unit 101 obtains the physical Euclidean distance *d* between any two nodes' corresponding sensor anchor clusters 102 and the magnitude of the three-dimensional flow velocity vector *v*. The main control computing unit 101 is equipped with spatial geometric projection calculation logic to calculate the spatial flow angle θ between the flow direction vector and the direction vector of the line connecting the two nodes, and to calculate the cosine value *cosθ* of this spatial flow angle θ. The system determines the causal relationship of the flow between nodes based on this cosine value *cosθ*.

[0035] When the calculated cosine value cosθ is greater than zero, the system determines that there is a direct water flow-driven correlation between upstream and downstream nodes, indicating the physical feasibility of material transport along the flow. The main control calculation unit 101 uses the ratio of the physical Euclidean distance d to the magnitude of the three-dimensional velocity vector v as the characteristic time parameter tf. The system performs a dynamic adjacency weight construction calculation based on fluid mechanics:

[0036] Aij=exp(−σ2d−v×cosθ×Δt)

[0037] In this formula, Aij represents the calculated dynamic adjacency weight of a node, characterizing the degree of dynamic impact of environmental changes on another node; exp is a power function with natural base, used here as an exponential decay function for nonlinear mapping; d represents the physical Euclidean distance between two nodes; v is the magnitude of the three-dimensional velocity vector; θ is the angle between the aforementioned flow vector and the direction of the line connecting the two nodes; Δt is the time step; and σ² is the smoothing variance constant used to control the width of the Gaussian kernel distribution. This formula maps distance, velocity, and flow angle into a nonlinear exponential decay space and couples them with cosine values ​​to accurately simulate how sudden changes in dissolved oxygen or ammonia nitrogen exceedances at the upstream inlet or aeration end are transmitted to the downstream outlet via the circulating hydrodynamic field and exhibit Gaussian diffusion and decay processes.

[0038] When the cosine value cosθ is less than or equal to zero, it indicates that the substance cannot diffuse backwards to that node. The main control computing unit 101 forcibly assigns the corresponding node's dynamic adjacency weight Aij to a preset extremely low value, i.e., the minimum threshold, actively disconnecting the feature information transmission link between these two nodes in the graph network at this moment. By introducing a construction algorithm that judges the direction of the spatial water flow angle and executes the blocking assignment, the topological dynamic change characteristics of the directed graph can be utilized to eliminate interference from irrelevant environmental noise disturbances to nodes in non-downstream areas.

[0039] Example 4

[0040] In high-density, factory-style aquaculture monitoring scenarios, keeping video and sonar equipment constantly on will deplete battery reserves or increase the thermal load on power cables. The massive amounts of multimedia data generated can also cause momentary malfunctions in limited local area network bandwidth. This system establishes cross-modal event-driven wake-up logic using the S305-based cross-modal event-driven wake-up mechanism.

[0041] Under normal operating conditions, the power module of the bio-observation component 203 within the sensor anchor cluster 102 is cut off, placing it in a deep sleep mode to release total communication bandwidth and reduce overall system power consumption. The basic physicochemical probe component 201 and the environmental flow field component 202 are configured to maintain a lightweight, periodic heartbeat communication connection with the data aggregation gateway 103 via a packaged low-power wide-area network protocol such as LoRa or NB-IoT.

[0042] The edge computing module 204 within the data aggregation gateway 103 acts as a forward sentinel for hard real-time response. The edge computing module 204 continuously receives dissolved oxygen concentration sequences (O2) and ammonia nitrogen concentration sequences (N) acquired in real-time at a frequency of 1Hz from the basic physicochemical probe component 201. Based on a preset sliding time window containing 60 seconds of data, the system calculates the first derivative value R of the dissolved oxygen concentration sequence O2 and the ammonia nitrogen concentration sequence N at the current time step, and uses the absolute value of this first derivative value R as the current rate of change of the physicochemical parameters.

[0043] The data aggregation gateway 103 integrates an adaptive threshold comparison module. Due to differences in baseline water quality across seasons or water cycle periods, this module dynamically adjusts the rate of change threshold Rth based on the daily average fluctuations of water physicochemical parameters over a historical week. In the specific stress detection scenario of this embodiment, a hard trigger threshold for the rate of decrease in dissolved oxygen concentration is set to 0.1 mg / (L⋅s), or an extreme threshold for the rate of increase in ammonia nitrogen concentration is set. This indicates that the system has detected extreme conditions such as a sharp drop in dissolved oxygen or an abnormal surge in ammonia nitrogen.

[0044] When the edge computing module 204 determines that the absolute value of the first derivative R exceeds the aforementioned rate of change threshold for multiple consecutive sampling periods, or when the environmental anomaly probability score P output by the aforementioned spatiotemporal graph convolutional network early warning model is higher than the set threshold of 0.75, the preset critical condition of the dual channels is met. The register of the edge computing module 204 immediately generates an interrupt request, generates a high-level interrupt signal, and sends the high-level interrupt signal directly to the power management controller inside the biological observation component 203 via the dedicated hardware interrupt trigger pin 205 inside the data aggregation gateway, using the underlying I2C communication system bus.

[0045] In response to the high-level interrupt signal that cannot be shielded by software, the power management controller of the bioobservation component 203 instantaneously turns on its internal field-effect transistor switch, directly closing the power supply circuit of the bioobservation component. The underwater low-light camera and imaging sonar complete system startup within 500 milliseconds, switching from an extremely low-power sleep mode to a high-load operating mode within seconds using the underlying bus, and immediately setting the camera and sonar frame rate to 30fps and the acoustic frequency band.

[0046] The system captures images of high-density schools of fish exhibiting behaviors such as surfacing or frantic swimming in response to stress, enabling accurate evidence collection without consuming excessive local area network bandwidth. Upon confirmation of this high-risk situation, the system immediately generates the highest-level high-risk warning signal.

[0047] To rapidly transmit massive amounts of disaster site evidence in confined channels, the data aggregation gateway 103 is responsible for packaging and uploading multimodal fusion data. This packaging process combines the multidimensional feature matrix of the flow field with a temporal compression algorithm for multidimensional image data, extracting keyframes representing the fish school outline, and merging them with the currently triggered 3D flow velocity vector and dissolved oxygen drop curve into a structured burst stress data packet. The data aggregation gateway 103 uploads this burst data packet in its entirety to the main control computing unit 101 via an internal RF antenna using a short-range wireless communication link or directly through a high-bandwidth industrial Ethernet channel, and then transfers it to the cloud. By employing a hardware-level interrupt triggering mechanism based on first-derivative mutation of the underlying bus combined with a high- and low-power mode hard switching mechanism, the high-frequency small data packets of physical sensors can act as precise valves for opening high-power vision devices, achieving an optimal balance between energy consumption and bandwidth management.

[0048] In summary, the multi-dimensional data integration processing and early warning system for aquaculture environments provided in this embodiment abandons the traditional approach of relying solely on a fixed absolute value threshold for dissolved oxygen concentration for alarm purposes. A high-frequency micro Doppler velocimeter or ultrasonic flowmeter array constructs a three-dimensional microscopic flow field in the aquaculture pond area in real time. The system uses the acquired three-dimensional flow velocity vector *v* as the core constraint of the physics engine. On one hand, it is injected into the dynamic time warping calculation, proactively compensating for the time lag in the arrival of sensor data at different nodes caused by water flow transport. On the other hand, it is fed into the adjacency matrix generation formula of the spatiotemporal graph convolutional network, ensuring that the graph network fully conforms to the partial differential physics causality law of fluid dynamics for information aggregation. When data that has eliminated short-term water flow disturbances and is correctly aligned is calculated to have an abnormally high environmental score, or when the dissolved oxygen index drops sharply with an abnormal slope, the edge side immediately activates the photoelectric sonar device using the hard real-time characteristics of the I2C bus low-level interrupt. After the visual algorithm confirms the abnormal stress characteristics of the fish population, the system finally packages and reports the data in a multi-modal manner for confirmation.

[0049] By deeply integrating environmental flow field parameters with physicochemical early warning algorithms, the problem of false alarms and excessive warnings caused by complex circulating hydrodynamic fields is fundamentally solved, avoiding asset losses caused by frequent equipment malfunctions. An innovative cross-domain collaborative mechanism driven by the rate of change of physical parameters to activate biological multimedia sensors balances the stringent battery power limitations of edge detection devices with the rigid requirement for full-band audio-visual monitoring and evidence collection during sudden disasters, thus improving the digital disaster relief and operational efficiency of factory-style smart fishery facilities. The specific sensor models and algorithm constants set for the aquaculture pond layout described above are merely preferred options based on optimal application scenarios. Simple substitutions or parameter scaling made by those skilled in the art based on the specific hydrological characteristics of the aquaculture area, without departing from the principles of overall flow field delay compensation and cross-triggered architecture, should be included within the scope of protection.

Claims

1. A multi-dimensional data integration processing and early warning system for aquaculture environments, characterized in that, The system includes a main control computing unit deployed locally at the factory, multiple sensor anchor clusters distributed within the aquaculture ponds, and a data aggregation gateway. Each sensor anchor cluster contains a basic physicochemical probe assembly for collecting physicochemical parameters, an environmental flow field assembly, and a biological observation assembly. The environmental flow field assembly is used to collect the three-dimensional velocity vector v of the flow field. The main control computing unit calculates the physical transmission time delay of the fluid between different sensor anchor clusters based on the three-dimensional velocity vector v, and uses a dynamic time warping algorithm to align the time-series data collected by the sensor anchor clusters. The main control computing unit runs a spatiotemporal graph convolutional network early warning model. The adjacency matrix between nodes of the spatiotemporal graph convolutional network early warning model is used to calculate the dynamic fluid dynamic correlation based on the three-dimensional flow velocity vector v, and outputs an environmental anomaly probability score P. The biological observation component is in a dormant mode under basic operating conditions. When the fluctuation rate of the physicochemical parameters collected by the basic physicochemical probe component meets a preset condition, or when the environmental anomaly probability score P is higher than a set threshold, the data aggregation gateway sends a hardware interrupt command to the biological observation component, triggering the biological observation component to enter the working mode to collect images of fish behavior in the aquaculture pond.

2. The multi-dimensional data integration processing and early warning system for aquaculture environment as described in claim 1, characterized in that, The main control computing unit is used to select the first upstream sensor anchor cluster and the second downstream sensor anchor cluster from the plurality of sensor anchor clusters, extract the spatial distance vector L between the first sensor anchor cluster and the second sensor anchor cluster, and calculate the average water flow velocity vector vˉ based on the three-dimensional flow velocity vector v collected by the environmental flow field components of the first sensor anchor cluster and the second sensor anchor cluster respectively. The unit divides the magnitude of the spatial distance vector L by the projected velocity value of the average water flow velocity vector vˉ on the spatial distance vector L to obtain the physical transmission time delay Tdelay from the first sensor anchor cluster to the second sensor anchor cluster.

3. The multi-dimensional data integration processing and early warning system for aquaculture environment as described in claim 2, characterized in that, When the main control computing unit aligns the time-series data collected by the sensor anchor cluster using the dynamic time warping algorithm, it is configured with a penalty constraint matrix M. The initialization parameters of the penalty constraint matrix M are dynamically assigned according to the physical transmission time delay Tdelay. The main control computing unit constructs two cost matrices C of time-series data to be aligned. When searching for the minimum cumulative distance path W in the cost matrix C, the size of the curved window that limits the search range is the product of the physical transmission time delay Tdelay and the preset time tolerance parameter α. Based on the value of the physical transmission time delay Tdelay, a nonlinear penalty weight wp is applied to the coordinates of the minimum cumulative distance path W that deviate from the diagonal, and the aligned time-series sequence set after compensating for the fluid propagation delay is output.

4. The multi-dimensional data integration processing and early warning system for aquaculture environment as described in claim 1, characterized in that, When performing the dynamic fluid dynamics correlation calculation, the main control computing unit obtains the physical Euclidean distance d between the sensor anchor point clusters corresponding to any two nodes and the three-dimensional flow velocity vector v, calculates the spatial flow angle θ between the water flow direction and the direction of the line connecting the two nodes, and calculates the node dynamic adjacency weight Aij based on the physical Euclidean distance d, the magnitude of the three-dimensional flow velocity vector v and the spatial flow angle θ. The node dynamic adjacency weight Aij is used to construct the adjacency matrix between nodes.

5. The multi-dimensional data integration processing and early warning system for aquaculture environment as described in claim 4, characterized in that, The main control computing unit is configured with computing logic to calculate the cosine value cosθ of the spatial water flow angle θ, and to determine the causal relationship of water flow between nodes based on the cosine value cosθ. When the cosine value cosθ is greater than zero, it is determined that there is a direct water flow driving association. The main control computing unit uses the ratio of the physical Euclidean distance d to the magnitude of the three-dimensional velocity vector v as a feature time parameter tf, performs a nonlinear mapping on the feature time parameter tf using an exponential decay function, and multiplies the mapped value with the cosine value cosθ to generate the node dynamic adjacency weight Aij. When the cosine value cosθ is less than or equal to zero, the main control computing unit forcibly assigns the corresponding node dynamic adjacency weight Aij to the minimum threshold value.

6. The multi-dimensional data integration processing and early warning system for aquaculture environment as described in claim 1, characterized in that, The data aggregation gateway includes an edge computing module and a hardware interrupt trigger pin. The edge computing module is used to receive the dissolved oxygen concentration sequence O2 and ammonia nitrogen concentration sequence N collected in real time by the basic physicochemical probe component, calculate the first derivative value R of the dissolved oxygen concentration sequence O2 and the ammonia nitrogen concentration sequence N based on a preset sliding time window, and use the first derivative value R as the fluctuation rate of the physicochemical parameters.

7. The multi-dimensional data integration processing and early warning system for aquaculture environment as described in claim 6, characterized in that, The data aggregation gateway is equipped with an adaptive threshold comparison module. This module dynamically adjusts the rate of change threshold Rth in the preset conditions based on the daily average fluctuation benchmark data of the water body's physicochemical parameters over historical periods. When the absolute value of the first derivative R exceeds the rate of change threshold Rth for multiple consecutive sampling periods, the edge computing module generates a high-level interrupt signal. This high-level interrupt signal is sent to the power management controller of the biological observation component via the underlying system bus through the hardware interrupt trigger pin. In response to the high-level interrupt signal, the power management controller directly closes the power supply circuit of the biological observation component, controlling it to switch from sleep mode to working mode. The collected images of fish behavior in the aquaculture pond are packaged into burst data packets and uploaded to the main control computing unit via a short-range wireless communication link.

8. The multi-dimensional data integration processing and early warning system for aquaculture environment as described in claim 1, characterized in that, The basic physicochemical probe assembly includes a dissolved oxygen sensor, an ammonia nitrogen sensor, and a water temperature sensor. The environmental flow field assembly includes a high-frequency miniature Doppler current meter or an ultrasonic flow meter array. The biological observation assembly includes an underwater high-definition camera and an auxiliary lighting source. The basic physicochemical probe assembly and the environmental flow field assembly are configured to maintain a periodic communication connection with the data aggregation gateway via a low-power wide area network protocol.

9. The multi-dimensional data integration processing and early warning system for aquaculture environment as described in claim 1, characterized in that, The spatiotemporal graph convolutional network early warning model includes a spatial graph convolutional layer and a temporal convolutional layer. The spatial graph convolutional layer uses the adjacency matrix between nodes to perform spatial dimension aggregation on the features of each sensor anchor cluster. The temporal convolutional layer is used to perform one-dimensional convolution operation on the features after spatial aggregation to extract temporal change features. The environmental anomaly probability score P is output through a fully connected layer.

10. The multi-dimensional data integration processing and early warning system for aquaculture environment as described in claim 1, characterized in that, The main control computing unit is also equipped with a network-wide clock synchronization module. The network-wide clock synchronization module is configured to send a unified timestamp tsync to all the sensor anchor clusters using a precise time protocol, and perform hardware-level basic time synchronization and alignment operations on the time series data based on the unified timestamp tsync.

Images