Protection area management planning method for aquatic product resource survey data based on edge calculation

Abstract

The invention relates to the technical field of aquatic product resource protection, and discloses a protection area management planning method of aquatic product resource survey data based on edge calculation. According to the method, firstly, an edge computing node network is deployed in a target water area, and multi-dimensional monitoring data such as biological distribution density, water environment parameters and human activity intensity are collected in real time; carrying out space-time fusion processing on the data to generate a space-time fusion matrix containing ecological sensitivity and resource abundance features; performing initial protection area division according to the matrix to obtain a plurality of candidate protection area units; performing iterative optimization screening on the candidate units based on a dynamic efficiency evaluation model, and outputting a core protection area boundary and a control level; and finally, performing adaptive dynamic adjustment on the boundary of the core protection area according to the real-time monitoring data flow to generate a final management planning scheme. According to the method, the data processing timeliness can be improved, dynamic division and adjustment of the protection area are realized, and accurate protection and sustainable utilization of aquatic resources are facilitated.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of aquatic resources protection, and particularly to a management planning method for aquatic resources survey data in a protected area based on edge computing. BACKGROUND

[0002] Currently, the management planning of aquatic resources protection areas faces many challenges. Traditional aquatic resources surveys rely on manual sampling and regular monitoring, which is not only time-consuming and labor-intensive, but also difficult to achieve real-time coverage of water areas, resulting in lagging and limited data acquisition. In terms of data processing, due to the lack of efficient real-time processing mechanisms, a large amount of monitoring data needs to be transmitted to the cloud for analysis, which not only increases the bandwidth pressure of data transmission, but also may affect the timeliness of decision-making due to network delay.

[0003] At the same time, existing protected area division methods are often based on static historical data, which are difficult to adapt to the dynamic changes of water ecological environment. For example, the distribution of aquatic organisms will migrate with factors such as season and water temperature, and fixed protected area boundaries cannot respond to these changes in a timely manner, which may result in ineffective protection of resources in some areas or excessive protection affecting the rational use of surrounding areas.

[0004] Human activities have an increasingly significant impact on water areas, and the intensity and scope of activities such as shipping, fishing, and pollution are constantly changing. Traditional management planning methods are difficult to capture these dynamic information in real time, making the control measures of protected areas lack of pertinence and flexibility. These problems collectively result in low efficiency of aquatic resources protection area management, making it difficult to achieve precise protection and sustainable use of aquatic resources. SUMMARY

[0005] The present application aims to provide a management planning method for aquatic resources survey data in a protected area based on edge computing to solve the problems raised in the background.

[0006] To achieve the above-mentioned purpose, the present application provides a management planning method for aquatic resources survey data in a protected area based on edge computing, which comprises:

[0007] Deploying an edge computing node network in the target water area, real-time collecting multi-dimensional monitoring data including biological distribution density, water environment parameters, and human activity intensity;

[0008] Performing spatio-temporal fusion processing on the multi-dimensional monitoring data to generate a spatio-temporal fusion matrix containing ecological sensitivity and resource abundance characteristics;

[0009] Dividing the initial protected area according to the spatio-temporal fusion matrix to obtain a plurality of candidate protected area units;

[0010] Based on the dynamic performance evaluation model, the multiple candidate protection area units are iteratively optimized and screened, and the core protection area boundary and management level are output.

[0011] According to the real-time monitoring data stream, the core protection area boundary is adaptively and dynamically adjusted, and a final aquatic resource protection area management planning scheme is generated.

[0012] Preferably, the real-time acquisition includes multi-dimensional monitoring data of biological distribution density, water environment parameters and human activity intensity, including:

[0013] The sonar detection data sequence and water quality sensor array data are obtained through the edge computing node, and the biological distribution density features are extracted by using a sparse coding algorithm based on a sliding window;

[0014] The time series of temperature gradient, dissolved oxygen concentration and turbidity parameters are synchronously acquired, and the water environment parameter change mode is extracted by using a time convolution kernel;

[0015] The ship automatic identification system signal and nearshore monitoring video stream are integrated, and the human activity intensity is quantified by using a spatio-temporal compressed sensing model.

[0016] Preferably, the multi-dimensional monitoring data are spatio-temporally fused to generate a spatio-temporal fusion matrix containing ecological sensitivity and resource abundance features, including:

[0017] The biological distribution density features and the water environment parameter change mode are standardized and aligned;

[0018] The ecological correlation weight between the biological distribution density features and the water environment parameter change mode is calculated by using a dynamic graph attention mechanism;

[0019] The human activity intensity is interference-corrected according to the ecological correlation weight to generate an edge feature vector;

[0020] The edge feature vector is input into a double-channel spatio-temporal encoder to generate the spatio-temporal fusion matrix through feature cross-fusion.

[0021] Preferably, the initial protection area division is performed according to the spatio-temporal fusion matrix to obtain multiple candidate protection area units, including:

[0022] The spatio-temporal fusion matrix is subjected to initial grid division based on a watershed transformation;

[0023] The ecological sensitivity index and the resource abundance index of each grid unit are extracted;

[0024] The division threshold is set according to the numerical distribution of the ecological sensitivity index and the resource abundance index;

[0025] Clustering the continuous grid areas satisfying the division threshold value as candidate protected area units.

[0026] Preferably, the dynamic performance evaluation model is used to iteratively optimize and screen the plurality of candidate protected area units, outputting core protected area boundaries and management levels, including:

[0027] Calculate the connectivity index and ecological integrity coefficient of each candidate protected area unit;

[0028] Input the connectivity index and ecological integrity coefficient into a random forest classifier to obtain a preliminary protection priority;

[0029] According to the real-time human activity intensity data, the preliminary protection priority is dynamically corrected to generate a unit protection performance value;

[0030] According to the gradient distribution of the unit protection performance value, the core protected area boundaries and the corresponding management levels are determined.

[0031] Preferably, the core protected area boundaries are adaptively and dynamically adjusted according to real-time monitoring data streams, including:

[0032] Establish a protected area state transition matrix containing biological migration paths and environmental factor fluctuation parameters;

[0033] Use a hidden Markov model to predict the evolution trend of the core protected area boundaries;

[0034] When the predicted evolution trend exceeds a preset tolerance interval, a boundary re-planning instruction is triggered.

[0035] Preferably, the method further comprises a protected area performance verification mechanism:

[0036] According to the boundary re-planning instruction, incremental data of current ecological sensitivity and resource abundance characteristics are obtained;

[0037] The incremental data is updated using an inverse distance weighted interpolation algorithm to update the spatio-temporal fusion matrix;

[0038] The unit protection performance value is recalculated using the updated spatio-temporal fusion matrix;

[0039] The recalculated unit protection performance value is compared with the historical planning scheme to output an optimized protected area management planning scheme.

[0040] Preferably, the method further comprises a protected area performance verification mechanism:

[0041] After the implementation of the optimized protected area management planning scheme, actual ecological restoration indicators are continuously collected;

[0042] The actual ecological restoration index is subjected to time lag correlation analysis with the planning expected index.

[0043] The parameter weight of the dynamic efficiency evaluation model is adjusted according to the time lag correlation analysis result.

[0044] Preferably, the parameter weight of the dynamic efficiency evaluation model is adjusted according to the time lag correlation analysis result, including:

[0045] A verification feature set containing ecological restoration rate and resource regeneration rate is constructed.

[0046] An error distribution matrix of the verification feature set and historical planning scheme is calculated.

[0047] The node split rule of the random forest classifier is updated through a gradient back propagation algorithm.

[0048] The updated node split rule is applied to the subsequent unit protection efficiency value calculation process.

[0049] Preferably, the method further comprises closed-loop feedback optimization:

[0050] A model correction coefficient is generated according to the cumulative change trend of the verification feature set.

[0051] The model correction coefficient is input into the spatio-temporal fusion matrix generation process.

[0052] A new round of protected area planning iteration is started based on the corrected spatio-temporal fusion matrix.

[0053] Compared with the prior art, the method has the following beneficial effects:

[0054] By deploying an edge computing node network in the target water area, multi-dimensional monitoring data such as biological distribution density, water environment parameters and human activity intensity can be collected in real time. The localization processing capability of the edge computing node reduces the delay and bandwidth consumption in the data transmission process, ensuring the timeliness and effectiveness of the monitoring data, and providing a fresh data basis for subsequent protected area management planning.

[0055] The spatio-temporal fusion processing of multi-dimensional monitoring data generates a spatio-temporal fusion matrix containing ecological sensitivity and resource abundance features, which can organically integrate data of different dimensions, different times and different spaces, and clearly present the overall situation and dynamic change trend of the water ecosystem. This fusion processing method breaks through the limitations of traditional scattered and isolated data, enabling managers to comprehensively and systematically understand the distribution of aquatic resources and the correlation of ecological environment, and providing a scientific and comprehensive basis for the division of protected areas.

[0056] According to the spatio-temporal fusion matrix, a plurality of candidate protection area units are obtained by initial protection area division, and then the dynamic performance evaluation model is used for iterative optimization and screening to output the core protection area boundary and management level, so that the division of the protection area is no longer dependent on static historical data, but combined with real-time ecological characteristics and resource conditions. The iterative optimization process of the dynamic performance evaluation model can continuously adjust and improve the protection area unit, so as to ensure that the core protection area can accurately cover the ecologically sensitive and resource-rich areas, and at the same time, according to the characteristics of different areas, the corresponding management level is set to realize the refinement and differentiation of protection area management.

[0057] According to the real-time monitoring data stream, the boundary of the core protection area is adaptively and dynamically adjusted to generate a final aquatic resource protection area management planning scheme, so that the boundary of the protection area can be flexibly adjusted according to the changes of the ecological environment and human activities. This dynamic adjustment mechanism can timely respond to the migration of aquatic organisms, changes in water environment and changes in human activity intensity, avoiding the problems of insufficient protection or over-protection caused by fixed boundaries, enhancing the adaptability and flexibility of protection area management, and helping to realize the dynamic protection and sustainable development of aquatic resources. BRIEF DESCRIPTION OF DRAWINGS

[0058] Figure 1 A working principle diagram of the protection area management planning method for aquatic resource survey data based on edge computing is provided.

[0059] Figure 2 A flowchart for real-time collection of multi-dimensional monitoring data is provided.

[0060] Figure 3 A flowchart for spatio-temporal fusion processing is provided.

[0061] Figure 4 A flowchart for dynamic performance evaluation and optimization is provided.

[0062] Figure 5 A flowchart for generating a final management planning scheme is provided. DETAILED DESCRIPTION

[0063] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0064] Please refer to Figure 1 The present application provides a protection area management planning method for aquatic resource survey data based on edge computing, which comprises:

[0065] The edge computing node network is deployed at the preset sampling points of the target water area, and the node spacing is set to 300-500 meters according to the terrain of the water area. Each node is configured with a multi-beam sonar sensor, a CTD water quality monitor, an AIS receiving module, and a high-definition camera. The system synchronously collects three types of data every 5 minutes: biological echo signal, water quality parameter time series, and ship positioning and video data. All raw data are preliminarily denoised at the edge node and transmitted to the central processing platform through the LoRaWAN protocol. The platform performs spatio-temporal alignment on the multi-dimensional monitoring data: the biological density heat map parsed by the sonar, the water quality parameter change surface, and the human activity intensity distribution map are projected onto the UTM coordinate system. Through three parallel computing pipelines, a spatio-temporal fusion matrix is generated: the first layer eliminates the timestamp error between sensors, the second layer establishes a 500m x 500m geographic grid index, and the third layer fills in the ecological sensitivity score (0-1 scale) and resource abundance index (0-100 scale) of each grid cell. The initial protected area division uses an adaptive watershed algorithm: the ecological sensitivity score is used as the terrain surface to divide the watershed, and the boundary line with a gradient change of more than 0.3 is selected to form the protected area contour. After the candidate protected area unit is generated, the iterative optimization engine starts: the ecological connectivity topological matrix and species integrity parameters of each unit are recalculated every 30 minutes, and the pre-trained machine learning model is input to output the protection value score. The core protected area boundary is automatically divided into three control areas according to the score value, and the boundary coordinates are transmitted to the supervision platform in real time. The dynamic adjustment module continuously monitors the changes in environmental parameters: when the biological migration path prediction model detects that the species distribution center of gravity shifts more than 500 meters, it triggers the re-planning process and generates a new protected area solution vector map.

[0066] Example 1: refer to Figure 2 , the edge computing node adopts an embedded hardware platform architecture, integrating multi-channel signal acquisition and processing units. The node device shell meets the IP68 waterproof standard, and is internally configured with a dual-core processor and a dedicated digital signal processing module. The biological distribution density feature extraction process starts with acoustic detection, and the sonar sensor array emits a detection pulse beam at a specific frequency, with a pulse duration precisely set in the millisecond range. The receiving end captures the acoustic echo signal of the water body, which is amplified and preliminarily filtered by the analog front end, and converted into a digital signal sequence. The system sets a time sliding window mechanism to process the digital signal sequence, and the sliding window covers a fixed length of continuous sampling points. The window signal data is processed by multiple filtering to eliminate environmental noise interference, and a specific feature extraction algorithm is applied to identify the target biological reflection spectrum feature. The spectrum feature is quantized and converted, outputting the biological distribution density value in floating point format, with an accuracy of a certain number of decimal places.

[0067] The water quality monitoring module periodically collects multi-dimensional water environment parameters, and the sampling period is fixed at several seconds. The temperature sensor probe measures the temperature gradient data of different water layers. The dissolved oxygen sensor uses an electrochemical method to measure the oxygen content in the water. The turbidity sensor detects the transparency of the water body through an optical method. The three parameters are synchronously collected according to a unified timestamp to generate a parameter sequence data stream with a time marker. The time series input is processed by a real-time analysis model, which includes multiple layers of computing structures to capture the parameter change rules at different time scales. The bottom processing unit focuses on the subtle fluctuation characteristics at the minute level and extracts short-term change pattern indicators. The upper network structure analyzes longer-term trends and identifies daily or tidal periodicity rules. The water environment parameter analysis results are output as a vector structure containing multi-dimensional feature values.

[0068] The human activity monitoring function is realized through multi-source heterogeneous data fusion. The ship positioning signal receiving module continuously listens to water surface radio signals, analyzes the ship identification code and real-time position coordinate information, and performs spatial coordinate system cumulative statistical analysis on the position data. The video monitoring system is equipped with a visible light and infrared dual-mode camera, which collects water surface activity image data at a fixed frame rate. The image stream is input into an object recognition model based on deep learning for automatic analysis and processing. The model structure includes multiple feature extraction layers and spatial pyramid pooling layers, and outputs ship type recognition results and position coordinate tracking information on the water surface. The positioning signal and video recognition results are processed by spatio-temporal correlation mapping, and the repeated targets are fused and removed. The space is divided into regular grid units, and the ship activity frequency and type distribution of each grid in a unit of time are counted. The human activity intensity index takes into account factors such as activity frequency density, ship tonnage factor, and speed characteristics to generate standardized intensity level quantitative results.

[0069] In the sonar data processing flow, the original echo signal is first subjected to analog-to-digital conversion processing, with a sampling frequency set to a specific kHz range. The converted digital signal sequence is divided into several processing unit blocks, each corresponding to a fixed time length. Within each unit block, a band-pass filter set is used for signal separation, and the biological feature frequency band signal is retained and enhanced. In the feature extraction stage, a preset frequency domain analysis technique is used to convert the time domain signal, and the power spectral density distribution feature is calculated. The identification of target biological features relies on a pre-established acoustic reflection feature library, which matches the spectral peak feature distribution pattern in a specific frequency range. The final generated biological distribution density mapping table includes real-time density evaluation values for each monitoring location point.

[0070] The time series processing of the water environment analysis model includes multiple parallel computing branches. The temperature parameter time series is processed by a specific length convolution kernel to extract the trend characteristics at different time scales. The dissolved oxygen concentration data analysis includes an outlier detection step, which applies a statistical outlier detection method to correct measurement errors. The short-term fluctuation characteristics of the turbidity parameter are suppressed by a moving average process to suppress random noise interference. The analysis results of each parameter are integrated in the multi-scale fusion layer, and the final comprehensive environmental index vector containing time dynamic characteristics is generated.

[0071] In the water surface moving target tracking process, the video analysis model performs motion target detection on the continuous frame sequence, and the background difference method combined with the optical flow method identifies the moving object area. The identified area is processed by a convolutional neural network for ship classification, and the classification result is associated with the estimated physical size of the ship. The positioning signal receiver continuously scans a specific frequency range and receives location information in a specific data packet format. The timestamps of the two data sources are aligned under a unified time reference, and the spatial coordinates use the same geodetic coordinate reference system. The target matching process calculates the spatio-temporal distance similarity between the video recognition target and the positioning signal, and the cross-modal data correlation fusion is realized through threshold judgment.

[0072] In the intensity quantification calculation stage, the grid elements of spatial region division are fixed-length square grids with a side length of meters. Each grid is associated with multiple quantization factors: the number of ship passes per unit time, the average tonnage order of magnitude value of the ship, the maximum observed speed of the ship, and the distribution weight of the ship activity type. Each factor is weighted and fused by a specific calculation method combined with standardized parameters to generate a human activity intensity index matrix for the region grid. The index value range is normalized to a floating point number between 0 and 1.

[0073] At the device network configuration level, edge computing nodes are distributed and deployed according to topological rules, and the node spacing is dynamically adjusted based on water characteristics. Data transmission uses a specific Internet of Things protocol for encapsulation, and the data packet format includes time markers, device identification codes, and payload structures. The node device has a built-in power management unit, which uses solar photovoltaic panels combined with lithium-ion batteries to provide continuous energy supply. The system internal clock achieves millisecond-level synchronization accuracy through a specific time synchronization protocol. Raw data is preliminarily compressed locally in the node, and specific algorithms are used to reduce data transmission volume without losing core feature information.

[0074] Example 2: see Figure 3In the data standardization processing stage, the feature data of different sources is subjected to independent data transformation operations. The biological distribution density value is processed by using a global scale adjustment method. Through the statistical value distribution characteristics of all monitoring points within a specific time window, the data center trend index and the data dispersion index are calculated, which are used to map all newly collected data to a unified numerical range interval. The water environment parameter data is subjected to an interval compression processing method. First, the numerical fluctuation upper and lower boundaries of each parameter within the historical time range are obtained. According to the boundary value, the numerical conversion rule is set to linearly project all observation values into the standard numerical interval. After the transformation of the two types of data, the data has the comparability feature, while the original data distribution rule is maintained.

[0075] The spatial topological relationship construction adopts a dynamic graph structure expression method. All monitoring points are regarded as vertex elements in the graph, and the vertex position information is derived from the geographic coordinate system. The relationship between vertices is established by a distance weight function, and the relationship strength index is generated according to the spatial proximity. There is a strong connection relationship between monitoring points with closer distance. The weight value of the edge in the graph presents a continuous decay trend with the increase of distance, and the weight tends to zero at a certain distance threshold.

[0076] The ecological correlation weight calculation is completed through a multi-stage attention mechanism. First, the statistical correlation between biological features and environmental parameters is analyzed within the local neighborhood range to generate a preliminary correlation strength parameter. Then, the feature interaction relationship is analyzed in a larger spatial range, and the correlation feature distribution is obtained through a parallel computing structure from a global perspective. Finally, the analysis results of different scales are summarized, and the aggregation function is used to output the comprehensive ecological correlation weight coefficient, which reflects the interaction strength of ecological system elements at a specific location.

[0077] The human activity interference correction process takes the ecological correlation weight as a control factor. The original human activity intensity index is multiplied by the weight coefficient, and the regional background activity level value is added in the superposition calculation. The corrected activity intensity value retains the original observation characteristics and reflects the influence of the sensitivity difference of the ecological system. The corrected value and the biological density and environmental parameters processed in the previous stage form a three-dimensional feature vector array.

[0078] The spatial feature channel adopts a graph structure neural network for feature extraction. The network input is the complete spatial topological relationship graph and the feature vector matrix, and the network level is set to a specific layer depth. The feature propagation of the graph nodes in the neighborhood range follows a specific aggregation rule, and the new state of each node is determined by the features and relationship weights of its adjacent nodes. Nonlinear activation transformation is introduced in the state updating process of each layer of the network to enhance the expression ability of the model. The spatial dimension depth feature mapping is generated through multi-layer feature propagation.

[0079] The time feature channel adopts a sequence processing model to capture the timing regularity. The environmental data is arranged in an equal interval time sequence, and a special memory gate structure is used to realize the long and short term feature preservation mechanism. The forward processing process records the historical feature evolution path, and the backward processing process obtains the future trend prediction feature. The bidirectional processing results are spliced and integrated in the fusion layer to generate a comprehensive time sequence feature array with time dependence.

[0080] The spatial features and time features are cooperatively fused in the cross-processing layer. A feature gate mechanism is designed to control the interaction ratio of the two types of information, generating a composite feature tensor structure. The tensor maintains a geographical grid arrangement in the spatial dimension, with each grid cell containing multi-dimensional feature descriptions. The final output data array is organized in a two-dimensional matrix form, with the cell positions corresponding to geographical coordinates and the data content including ecological evaluation indicators and resource statistical indicators.

[0081] An image processing method is used in the initial division stage of the protected area. The resource abundance distribution map is used as the input layer, and the edge detection operator scans the continuous feature change gradient to identify the spatial discontinuity position as the potential boundary reference line. The core position of the region is determined in combination with the ecological sensitivity distribution characteristics, and the core position is expanded outward following the principle of minimum gradient until it reaches the area with sharp feature changes. This region growing method divides the water area into multiple independent spatial units.

[0082] Unit index quantization statistics are performed in the divided grid area. The ecological sensitivity value is the average of all monitoring points in the unit, and the resource abundance value is the minimum value in the unit. The statistical process excludes the sampling points near the boundary line to avoid partition error interference.

[0083] The partition determination standard is set according to the overall statistical distribution characteristics. The ecological sensitivity mean threshold is set to a specific quantile value, and the resource abundance minimum value requirement is set to a fixed standard line. These two indicators together form the necessary and sufficient conditions for regional screening.

[0084] The spatial clustering process identifies continuous grid groups that meet the conditions. Grid cells that are adjacent in position and meet the index requirements are connected, and the neighborhood range is determined according to a specific distance parameter. The connected cell set is checked for spatial connectivity, and small independent areas are merged into adjacent large areas. The final geographical region unit is retained as a polygon boundary coordinate set and stored as an analyzable geographic information data structure.

[0085] Example 3: see Figure 4In the dynamic performance evaluation model implementation phase, the connectivity index of the candidate protected area unit is calculated first. This index is realized by using the electrical network simulation method, which converts the spatial structure of water into an equivalent resistance network structure. The center point of each grid unit is set as the node of the circuit, and the resistor device is connected between adjacent nodes, and the resistance value is proportional to the actual geographical distance. A constant potential difference is applied to the selected core unit node, and the current distribution of the boundary node is measured. By calculating the proportion of the outflow current to the total input current, the spatial connectivity performance is quantified, and the distance factor and path complexity are also included in the evaluation system. The connectivity value range is set between 0 and 1, and the higher the value, the stronger the core of the unit in the ecological corridor.

[0086] Ecological integrity assessment includes two dimensions of morphological characteristics and biodiversity. Morphological characteristics are obtained by calculating the geometric properties of the unit polygon, considering the mathematical relationship between perimeter and area, and identifying the degree of deviation from normal shape regularity. Biodiversity analysis is based on acoustic and optical monitoring data, and the spatial distribution density difference of different species is identified. The species abundance information is processed by logarithmic transformation, and the diversity comprehensive evaluation value is generated by combining the distribution uniformity index. The final integrity coefficient is the weighted combination result of the shape factor and the diversity index, and the coefficient weight is preset according to the fixed proportion of the water ecological characteristics.

[0087] Before the random forest model processes the unit evaluation parameters, the model training configuration phase needs to be completed. The training data set comes from historical water protected area cases, containing thousands of sample unit feature label pairs. The model is composed of multiple decision trees, and each tree generates a branch structure using specific rules. When a new unit parameter vector is input, each decision tree independently outputs a classification decision result, and the voting results of all trees are statistically merged to generate the initial protection priority value of the unit, which reflects the basic evaluation of the ecological protection value of the unit.

[0088] The real-time human activity intensity monitoring system continuously updates the data stream, and collects the latest ship distribution information every fixed period. The data analysis engine compares the deviation amplitude of the current intensity value from the historical baseline level, and constructs a time decay function model. This function describes the credibility decay law of the historical priority evaluation results over time. The protection priority correction mechanism converts the real-time activity intensity deviation into an adjustment factor, which is applied to the basic priority value by subtraction, and sets a reasonable interval range limit for the output value.

[0089] The unit protection performance value is calculated by integrating the connectivity, integrity and adjusted priority parameters. The comprehensive calculation formula is introduced as follows:

[0090] Among them: P represents the unit protection performance value, P represents the corrected protection priority, is the connectivity index, is the ecological integrity coefficient, is the preset connectivity weight factor. The calculation result is divided into three management level categories according to a certain numerical interval standard, and each level corresponds to differentiated management and control strategy requirements. The core protection zone boundary is composed of continuous spatial units that meet the highest level standard.

[0091] The boundary dynamic prediction mechanism is based on a state transition model. The biological migration path data comes from the results of acoustic marker tracking experiments, recording the spatial migration trajectory rules of species under different environmental conditions. The environmental factor fluctuation parameters include dynamic indicators such as diurnal water temperature variation amplitude and dissolved oxygen concentration fluctuation frequency. The spatial state transition matrix has a dimension equal to the total number of protection zone units, and the matrix elements reflect the biological transfer probability between units under specific environmental conditions.

[0092] The Hidden Markov Model configures three types of hidden state categories, corresponding to the stability, expansion and contraction characteristics of the protection zone spatial pattern. The observation layer includes the processing results of environmental monitoring parameters, such as water quality parameter change trend slope and biological density spatial variation coefficient. The model parameter estimation uses an iterative optimization algorithm to adjust the distribution parameters of state transition probability and observation probability at each iteration. The prediction process is based on the current observation sequence to calculate the probability distribution of each hidden state, and continuously monitors the frequency of the dominant state.

[0093] The boundary stability judgment sets numerical threshold standards. When the model prediction result indicates that the protection zone contraction state probability exceeds the predetermined warning level for multiple consecutive periods, it is determined to be a potential ecological risk state. At this time, the system generates an electronic command signal to trigger the subsequent boundary re-planning process. The judgment mechanism sets a safety buffer parameter to avoid excessive sensitivity response due to temporary environmental fluctuations. The re-planning instruction contains spatial position encoding and time marker information, providing a starting instruction basis for the data update process. All prediction logs are stored in a special database for subsequent model optimization reference.

[0094] Example 4: refer to Figure 5 After the boundary re-planning instruction is triggered, the system automatically records the instruction timestamp T0 as the starting point of incremental data collection. Taking the Yangtze River Estuary A water protection zone as an example, when the Hidden Markov Model outputs the contraction probability exceeding the 0.8 threshold for three consecutive times, the system immediately starts a 24-hour rolling monitoring window. The sensor array deployed in the G07 grid region enters high-frequency collection mode, and the biological sonar sampling interval is compressed from 5 minutes to 1 minute, and the water quality sensor starts real-time back transmission at the second level. The newly added mobile monitoring platform cruises along the protection zone boundary to supplement the data collection of 12 temporary points.

[0095] Incremental data processing adopts a hierarchical update strategy. Taking the dissolved oxygen parameter as an example, the original time series is [8.2, 8.1, 8.0, 7.9] mg / L (time points T-3 to T0), and the new data points are [8.3, 8.4] (T+1 to T+2). The data smoothing process retains the last value of the original sequence and fuses the new observation value according to the preset proportion to form the updated sequence [8.2, 8.1, 8.0, 8.1, 8.2]. The human activity intensity update adopts ship trajectory density recalculation, and the original grid cell division remains unchanged, but the statistical period is switched to a rolling 24-hour window. Table 1 below shows the incremental update parameters of some grids in the G07 area.

[0096] Table 1: Incremental update parameter table of water area grid characteristics.

[0097] Grid encoding Biological density increment (ind / m3) Dissolved oxygen change rate (%) Turbidity fluctuation value (NTU) Ship activity frequency Intensity correction coefficient A-23 +0.12 -3.2 +0.15 7→9 0.82→0.79 B-15 -0.05 +1.8 -0.08 3→2 0.91→0.93 C-09 +0.31 -5.1 +0.22 5→12 0.75→0.68

[0098] In the protected area feature matrix reconstruction stage, the spatio-temporal encoder loads the pre-trained model weights. The biological density change is filled in the blank area using the spatial interpolation method: taking the monitoring buoy M08 in the G07 area as the center point (coordinates 121.35°E, 31.41°N), its new density value 0.15 ind / m³ affects the surrounding three blank grids according to the distance decreasing principle. The distance parameter is inversely proportional to the weight. After generating the updated matrix, the protected area recalculation process is started, and the unit connectivity evaluation adds the migration corridor constraint condition: when detecting the dolphin activity trajectory crossing the G15-G18 area, the ecological connectivity index of the grid units on this path is additionally increased by 0.15.

[0099] The effectiveness comparison adopts the grid overlay analysis method. The original scheme plans the core protected area with an area of 46.7 km², and the adjusted new scheme is 42.3 km². The spatial difference engine identifies two main change areas: 1) the protected area boundary in the East Beach area shrinks by 1.2 km to the land side, corresponding to the addition of 13 ship activity frequencies in the B-09 to B-14 grid units; 2) a new protected zone is added in the North Port waterway, covering the newly appeared Chinese herring spawning ground C-03 unit. The change difference rate calculation result shows that the shrinkage area accounts for 18.6% of the original area, and the expansion area accounts for 7.2% of the new scheme. After the scheme change takes effect, the supervision system automatically pushes the boundary change coordinates to the navigation terminals of the 6 patrol ships.

[0100] The effectiveness verification mechanism is started 30 days after implementation. The ecological restoration monitoring group sets up 10 sampling points at Wusongkou and uses standard plankton nets (net diameter 0.5 mm) to sample three times a week. The actual collected data shows that: the first week, the average phytoplankton abundance is 45×10 4 cells / L, while the planned expected value is 50×10 4 cells / L; by the fourth week, the actual value increases to 55×10 4 cells / L, close to the expected value of 58×104 The time-lag analysis was calculated by a sliding window: a 14-day window was set, and the expected change curve was compared with the measured data day by day. The highest correlation of the expansion rate of the spawning ground was found at a 17-day delay (Pearson coefficient 0.89).

[0101] The model parameter adjustment process was based on the time-lag analysis results. In the dimension of fish resource assessment, the growth coefficient in the original prediction model was initially set to 0.12, and the analysis found that there was a daily deviation of 0.8% from the actual recovery rate. By adjusting the equation parameters, the slope of the simulation curve was reduced by 10%, and a temperature compensation term was added. The updated random forest model increased the seasonal weight in the water depth feature, automatically increasing the ecological sensitivity evaluation weight of the shallow area (<5m) during the spring spawning period by 0.15. The parameter update log showed that the dissolved oxygen influence factor was corrected from 0.32 to 0.28, and the decay rate parameter of the ship interference coefficient was adjusted from 0.7 to 0.65. After the revised model was redeployed, the newly collected monitoring data of Hengsha Island showed that the matching degree of the revised buffer zone boundary with the actual biological enrichment area was improved by 8 percentage points.

[0102] Example 5: Verification feature data collection was systematically carried out within a quarterly cycle. The ecological restoration index monitoring system included multiple dimensions of observation content: plankton biomass used underwater imaging system to obtain sample volume concentration; benthic biodiversity was counted by sea bottom trawl sampling combined with image recognition; economic fish resources used underwater acoustic scanning to generate three-dimensional density distribution map; spawning ground space range used multispectral satellite remote sensing to identify substrate feature change area. Each item of data was stored in a time series database monthly, each record containing geographic coordinates, collection timestamp, parameter type and quantitative value fields. The collection process strictly followed the standard operating procedures, and the shipborne equipment performed sensor calibration program before each operation, and the unmanned aerial remote sensing platform preset fixed flight path covered the entire territory of the protected area.

[0103] Error analysis was based on historical planning target comparison. The system automatically retrieved the prediction data curve at the beginning of the planning, and aligned the actual collection sequence according to the same time coordinate axis. For each verification index, a difference analysis model was constructed to calculate the dispersion degree of the measured value and the predicted value at the same time node. The difference calculation process considered environmental background fluctuation factors, and introduced the historical mean of meteorology and hydrology for standardized correction. The dispersion error was aggregated according to the geographical unit, and the error distribution heat map of the east, west, south and north four partitions of the protected area was generated. The difference feature analysis module identified the error space aggregation area, and marked the key observation area with error amplitude exceeding the set threshold for three consecutive quarters.

[0104] The model parameter update performs a gradient-driven iterative strategy. The random forest classifier dynamically adjusts the decision tree node judgment conditions while retaining the original structural framework. The loss function defines the comprehensive error space distribution and time persistence dual factors, and calculates the cumulative deviation of the current model output from the validation features. The back propagation process retraces the decision tree path to locate the split nodes that need to be optimized, and implements a slight shift adjustment to the classification judgment threshold. The node update operation follows the conservative principle, and the adjustment amplitude of a single iteration is controlled within the preset safe range. Major parameter changes need to be verified by historical data before execution. The updated model introduces a sample weight rebalancing mechanism, and the observation data in the area with significant recent errors are given higher weight coefficients.

[0105] The closed-loop feedback establishes a quarterly cycle polling mechanism. Every 90 days, the system automatically starts the validation evaluation process, calculates the change trend slope of each ecological indicator in the statistical period. The change trend analysis uses the moving window method to detect the mutation point, and excludes seasonal fluctuation interference. Based on the change trend of the validation features, a dynamic correction coefficient array is generated, which contains positive and negative floating point values, and the absolute value reflects the adjustment intensity. The coefficient generation process integrates expert experience rule library, when the recovery rate of a specific biological population shows abnormal changes, the corresponding coefficient is added with an additional adjustment factor.

[0106] The correction coefficient acts on the protected area planning basic data flow. The spatio-temporal fusion matrix generation module receives the correction coefficient input and applies the coefficient intervention in the feature encoding stage. The data standardization processing link introduces a coefficient weighting item to make the ecological sensitivity evaluation biased towards the key correction area. The feature propagation process of the graph neural network is affected by the coefficient to adjust the connection weight between nodes and change the calculation method of spatial correlation. The matrix output layer sets up a coefficient response channel to fine-tune the final ecological sensitivity score and resource abundance index according to the partition coefficient value.

[0107] The self-triggered restart mechanism is activated under certain conditions. When the average of the continuous correction coefficients exceeds the critical threshold, the system determines that the planning basic data needs to be updated comprehensively. At this time, the candidate protected area cache database is automatically cleaned up, and the seed point selection logic of the watershed segmentation algorithm is reinitialized. The new round of planning process inherits the historical validation experience data, and the updated spatio-temporal fusion matrix starts the space grid reconstruction. The closed-loop system retains the complete version of the iteration record, and each iteration version is associated with the current environmental feature snapshot data to build a continuously evolving protected area planning knowledge base. This periodic self-updating mechanism enables the planning system to adapt to the long-term dynamic changes of the water ecosystem, while ensuring that the key technical parameters always reflect the latest ecological development needs.

[0108] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting; it is not intended to exclude myriad other embodiments of the present application that other inventors can develop based on the same general inventive concepts embodied by the described embodiments. That is, although the present application is described in terms of particular embodiments and illustrative figures, it should be apparent that the scope of the present application is not limited to these specific embodiments.

[0109] While the embodiments of the application have been shown and described herein, it will be understood by those skilled in the art that many changes, modifications, substitutions and alterations to these embodiments can be made without departing from the principles and spirits of the application, and it is intended that the scope of the application be limited solely by the scope of the appended claims and the equivalents thereof.

Claims

1. A protected area management planning method based on aquatic resource survey data using edge computing, characterized in that, include: Deploy an edge computing node network in the target water area to collect multi-dimensional monitoring data in real time, including biological distribution density, water environment parameters, and intensity of human activities. The multidimensional monitoring data is subjected to spatiotemporal fusion processing to generate a spatiotemporal fusion matrix that includes ecological sensitivity and resource abundance characteristics; The initial protection zone is delineated based on the spatiotemporal fusion matrix, resulting in multiple candidate protection zone units; Based on the dynamic performance evaluation model, the multiple candidate protected area units are iteratively optimized and screened to output the core protected area boundary and control level. The boundaries of the core protected area are adaptively and dynamically adjusted based on real-time monitoring data streams to generate the final management plan for the aquatic resource protection area.

2. The protected area management and planning method based on edge computing aquatic resource survey data according to claim 1, characterized in that, The real-time acquisition includes multi-dimensional monitoring data on biological distribution density, aquatic environmental parameters, and the intensity of human activities, including: Sonar detection data sequences and water quality sensor array data are acquired through edge computing nodes, and biological distribution density features are extracted using a sparse coding algorithm based on a sliding window. Time series of temperature gradient, dissolved oxygen concentration and turbidity parameters were collected simultaneously, and the change patterns of water environment parameters were extracted using time convolution kernels. By integrating signals from the Automatic Identification System (AIS) and nearshore surveillance video streams, the intensity of human activities is quantified using a spatiotemporal compression sensing model.

3. The protected area management and planning method based on edge computing aquatic resource survey data according to claim 2, characterized in that, The process of performing spatiotemporal fusion processing on the multidimensional monitoring data to generate a spatiotemporal fusion matrix containing ecological sensitivity and resource abundance characteristics includes: The biological distribution density characteristics are standardized and aligned with the water environment parameter change patterns. The ecological association weights between the biological distribution density characteristics and the water environment parameter change patterns were calculated using a dynamic graph attention mechanism. The intensity of human activities is corrected for interference based on the ecological association weights to generate edge feature vectors; The edge feature vector is input into a dual-channel spatiotemporal encoder, and the spatiotemporal fusion matrix is ​​generated through feature cross-fusion.

4. The protected area management and planning method based on edge computing aquatic resource survey data according to claim 3, characterized in that, The initial protection zone delineation based on the spatiotemporal fusion matrix yields multiple candidate protection zone units, including: The spatiotemporal fusion matrix is ​​subjected to initial grid partitioning based on watershed transformation; Extract the ecological sensitivity indicators and resource abundance indicators for each grid cell; The division threshold is set based on the numerical distribution of the ecological sensitivity index and resource abundance index; Continuous grid regions that meet the defined segmentation threshold are clustered into candidate protected area units.

5. The protected area management and planning method based on edge computing aquatic resource survey data according to claim 4, characterized in that, The dynamic performance evaluation model is used to iteratively optimize and screen the multiple candidate protected area units, outputting the core protected area boundary and control level, including: Calculate the connectivity index and ecological integrity coefficient for each candidate protected area unit; The connectivity index and ecological integrity coefficient are input into a random forest classifier to obtain a preliminary protection priority. The initial protection priority is dynamically corrected based on real-time human activity intensity data to generate unit protection effectiveness values. The boundaries of the core protection zone and the corresponding control level are determined based on the gradient distribution of the unit protection effectiveness values.

6. The protected area management planning method based on edge computing aquatic resource survey data according to claim 5, characterized in that, The adaptive dynamic adjustment of the core protection zone boundary based on real-time monitoring data stream includes: Establish a state transition matrix for the protected area, including biological migration routes and environmental factor fluctuation parameters; Hidden Markov Model (HMM) is used to predict the evolution trend of the core protected area boundary; When the predicted evolution trend exceeds the preset tolerance range, a boundary replanning instruction is triggered.

7. The protected area management planning method based on edge computing aquatic resource survey data according to claim 6, characterized in that, The process of generating the final aquatic resource protection zone management plan includes: Incremental data on current ecological sensitivity and resource abundance characteristics are obtained according to the boundary replanning instructions; The spatiotemporal fusion matrix is ​​updated using an inverse distance weighted interpolation algorithm on the incremental data; The unit protection effectiveness value is recalculated using the updated spatiotemporal fusion matrix; The recalculated unit protection effectiveness value is compared with the historical planning scheme, and the optimized protection zone management plan is output.

8. The protected area management planning method based on edge computing aquatic resource survey data according to claim 7, characterized in that, It also includes a mechanism for verifying the effectiveness of protected areas: After the implementation of the optimized protected area management plan, actual ecological restoration indicators will be continuously collected. A time-lag correlation analysis was performed between the actual ecological restoration indicators and the planned expected indicators. The parameter weights of the dynamic performance evaluation model are adjusted based on the results of the time-delay correlation analysis.

9. The protected area management planning method based on edge computing aquatic resource survey data according to claim 8, characterized in that, The adjustment of parameter weights in the dynamic performance evaluation model based on the time-delay correlation analysis results includes: Construct a validation feature set that includes ecological restoration rate and resource regeneration rate; Calculate the error distribution matrix between the verification feature set and the historical planning scheme; The node splitting rules of the random forest classifier are updated using the gradient backpropagation algorithm. The updated node splitting rules are then applied to the subsequent calculation of unit protection effectiveness values.

10. The protected area management planning method based on edge computing aquatic resource survey data according to claim 9, characterized in that, It also includes closed-loop feedback optimization: Model correction coefficients are generated based on the cumulative change trend of the verification feature set; The model correction coefficients are input into the spatiotemporal fusion matrix generation process; A new round of protected area planning iterations will be launched based on the revised spatiotemporal fusion matrix.

Images