Natural oyster reef ecological environment restoration method and system based on symbiotic competition mechanism

Abstract

The invention discloses a natural oyster reef ecological environment restoration method and system based on a symbiotic competition mechanism, and relates to the technical field of ecological environment restoration, and the method comprises the steps: collecting time series data, building a symbiotic competition relation graph based on the collected data and a graph neural network, recognizing potential key factors, and carrying out the feature importance analysis based on a random forest model. Screening out indication factors which are high in sensitivity and easy to monitor; calculating the ratio of a target area factor to a health group mean value, generating a factor characteristic curve, judging the degree that the factor deviates from a health state through a time domain characteristic, and formulating a basic restoration measure according to an indication factor threshold value; a reinforcement learning algorithm is introduced, and measure combination is dynamically optimized through real-time interaction with an ecological system. According to the invention, the factor influence weight is quantified through the random forest algorithm, and the indication factor identification accuracy is improved; reinforcement learning and real-time monitoring are combined, a restoration scheme can be adaptively adjusted according to ecological changes, and efficiency is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ecological environment restoration, and more particularly to a natural oyster reef ecological environment restoration method and system based on a symbiotic competition mechanism. BACKGROUND

[0002] At present, oyster culture is in a high-density cultivation state due to blind expansion of scale, which seriously threatens the natural growth environment of natural spat areas. As an important oyster habitat, natural spat areas are likely to have a certain scale of natural oyster reefs, so the protection of natural spat areas is also the protection of oyster reefs.

[0003] However, as one of the most threatened coastal ecosystems, oyster reefs have lost more than 85% of their habitats worldwide. As a biological reef, oyster reefs have important ecosystem service functions such as water purification, biological habitat, coastal protection, and biological carbon sequestration, and have high economic value.

[0004] At the same time, affected by factors such as overfishing, water pollution, and coastal development, the habitat of natural spat areas is in crisis. More importantly, due to the lack of systematic investigation and data, the current situation and the degree of threat of natural spat areas are in an unknown state, which brings great difficulties to the protection, restoration, and sustainable use of oyster reef resources. In addition, the public's lack of understanding of the importance of oyster reefs, their ecological functions, and their interaction with mangroves, coral reefs, and seagrass beds also hinders their effective protection.

[0005] Therefore, how to propose a natural oyster reef ecological environment restoration method and system based on a symbiotic competition mechanism to improve the pertinence and effectiveness of ecological restoration and to improve the ecological environment of natural oyster reefs is a problem that those skilled in the art need to solve. SUMMARY

[0006] Therefore, the present application provides a natural oyster reef ecological environment restoration method and system based on a symbiotic competition mechanism to solve the problem that natural oyster reefs and natural spat areas are threatened by factors such as high-density cultivation, overfishing, water pollution, and coastal development, and lack of systematic investigation data support, making it difficult to protect, restore, and sustainably use them. At the same time, it is committed to improving the pertinence and effectiveness of ecological restoration to improve the ecological environment of natural oyster reefs. In order to achieve the above purpose, the present application adopts the following technical solutions:

[0007] A natural oyster reef ecological environment restoration method based on a symbiotic competition mechanism, comprising:

[0008] Collecting time series data, constructing a symbiotic competition relationship graph based on the collected data and a graph neural network, calculating node degree centrality and intermediate centrality, and identifying potential key factors;

[0009] Introduce the random forest algorithm, take the potential key factor data as the input, take the ecosystem health index as the output label, train multiple decision trees and fuse the results, screen out the sensitive and easily monitored indicator factors based on the feature importance analysis of the random forest model;

[0010] Calculate the ratio of the target area factor to the mean value of the health group, generate a factor characteristic curve, judge the degree of deviation of the factor from the healthy state through the time domain characteristics, and formulate basic repair measures according to the indicator factor threshold;

[0011] Introduce the reinforcement learning algorithm, take the repair measures as the action strategy, take the improvement degree of the indicator factor as the reward signal, and dynamically optimize the measure combination through real-time interaction with the ecosystem.

[0012] Optionally, the collection of time series data includes: on-site sampling and laboratory analysis of water quality, plankton, swimming animals and oyster populations in the target area to collect time series data.

[0013] Optionally, the water quality of the target area includes: pH value, dissolved oxygen, nitrogen and phosphorus content, plankton includes: species, quantity, dominant species, swimming animals includes: habitat, oyster population includes: density, growth state, collect time series data for n consecutive years, and select the same type of healthy oyster reef ecosystem as the reference group, collect the same period data to generate a health mean fluctuation curve.

[0014] Optionally, the construction of the symbiotic competition relationship graph based on the collected data and the graph neural network includes: based on the survey data, taking water quality factors and biological populations as complex network nodes, taking predation, symbiosis and competition relationships as edges, constructing a symbiotic competition relationship graph through a graph neural network, calculating node degree centrality and intermediate centrality, and identifying potential key factors.

[0015] Optionally, the identification of potential key factors includes:

[0016] The water quality factors and biological populations in the ecological survey data are defined as nodes in the complex network;

[0017] The interaction relationship between nodes is defined as an edge, which is represented by an adjacency matrix;

[0018] Each node is assigned a feature vector to form a feature matrix;

[0019] The features of the node and its neighbors are aggregated through the graph convolution layer of the GNN;

[0020] With the ecosystem health index as the label, the model parameters are optimized through back propagation, and after training, the node feature importance score is used to identify potential key factors that significantly affect the stability of the ecosystem.

[0021] Optionally, the introduced random forest algorithm takes the potential key factor data as input and the ecosystem health index as output label, trains multiple decision trees and fuses the results, including:

[0022] The training set is generated by sampling from the preprocessed data set through the Bootstrap method;

[0023] Select water quality factors, plankton, swimming animals, and oyster population ecological factors as input variables to form an input feature matrix. The health status of the target area is quantified as a health index based on the health group's ecosystem state as the benchmark, and the model's output label is the health index;

[0024] For each training set, a decision tree is trained;

[0025] For new samples, the health index is predicted by all decision trees, and the average value is taken as the final output;

[0026] For each decision tree, calculate the sum of the Gini impurity reduction caused by each ecological factor in all split nodes. Sum the impurity reduction of all decision trees and normalize to get the impact weight of the factor. The larger the value, the more significant the factor's impact on the stability of the ecosystem.

[0027] Optionally, the feature importance analysis based on the random forest model includes: normalizing the importance scores of all factors, and selecting the top 30% of factors as candidate sensitive factors; for the candidate sensitive factors, combined with the feasibility of field monitoring, finally screening out the indicator factors with high sensitivity and easy monitoring.

[0028] Optionally, the ratio of the target area factor to the health group mean is calculated to generate a factor feature curve, and the degree of deviation of the factor from the healthy state is determined by the time domain characteristics, including:

[0029] The ratio of the target area factor to the health group mean is calculated to generate the basic data of the factor feature curve;

[0030] Draw the factor feature curve with years as the horizontal axis and the ratio as the vertical axis;

[0031] Calculate the annual slope of the factor feature curve to reflect the rate of change of the deviation of the factor;

[0032] Linear fitting is performed on the factor feature curve to obtain a trend line, and the long-term trend is determined by the slope;

[0033] The size of the combination ratio, the slope and the trend are combined to quantify the degree of deviation, and the greater the degree of deviation of the factor from the healthy state.

[0034] Optionally, the basis repair measures include:

[0035] The health threshold range of each indicator factor is determined according to the mean value of the indicator factor of the healthy group and the ecological safety threshold value, and the corresponding repair measures are triggered according to the deviation of the current value of the indicator factor of the target area from the threshold value.

[0036] Optionally, a natural oyster reef ecological environment repair system based on a symbiotic competition mechanism comprises:

[0037] The collection module is used for collecting time series data.

[0038] The potential key factor identification module is used for constructing a symbiotic competition relationship graph based on the collected data and a graph neural network, calculating node degree centrality and intermediate centrality, and identifying potential key factors.

[0039] The screening module is used for introducing a random forest algorithm, taking the potential key factor data as input and taking the ecosystem health index as output label, training multiple decision trees and fusing the results, screening out sensitive and easily monitored indicator factors based on feature importance analysis of the random forest model.

[0040] The preparation module is used for calculating the ratio of the factor of the target area to the mean value of the healthy group, generating a factor feature curve, judging the degree of deviation of the factor from the healthy state through time domain features, and preparing basis repair measures according to the threshold value of the indicator factor.

[0041] The optimization module is used for introducing a reinforcement learning algorithm, taking the repair measures as action strategies and taking the improvement degree of the indicator factor as a reward signal, and dynamically optimizing the combination of measures through real-time interaction with the ecosystem.

[0042] According to the above technical solution, compared with the prior art, the present application provides a natural oyster reef ecological environment repair method and system based on a symbiotic competition mechanism, which has the following beneficial effects:

[0043] The application provides a natural oyster reef ecological environment restoration method based on a symbiotic competition mechanism, which comprises the following steps: collecting time sequence data, constructing a symbiotic competition relationship graph based on the collected data and a graph neural network, calculating node degree centrality and intermediate centrality, and identifying potential key factors; introducing a random forest algorithm, taking the potential key factor data as input and taking an ecological system health index as output label, training multiple decision trees and fusing the results, screening out sensitive and easily monitored indicator factors based on feature importance analysis of the random forest model; calculating the ratio of the target area factor to the mean value of the health group, generating a factor feature curve, judging the degree of deviation of the factor from the healthy state through time domain characteristics, and formulating basic restoration measures according to the indicator factor threshold; introducing a reinforcement learning algorithm, taking the restoration measures as action strategies and taking the improvement degree of the indicator factor as a reward signal, and dynamically optimizing the measure combination through real-time interaction with the ecological system. The application quantifies the factor influence weight through the random forest algorithm, improves the identification accuracy of the indicator factor, combines reinforcement learning with real-time monitoring, adjusts the restoration scheme adaptively according to the ecological changes, improves the efficiency, compares the time sequence data of consecutive n years with the health group, solves the problem of insufficient data in traditional restoration, and provides support for cross-ecosystem restoration. BRIEF DESCRIPTION OF DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0045] Figure 1 A natural oyster reef ecological environment restoration method based on a symbiotic competition mechanism is provided.

[0046] Figure 2 A natural oyster reef ecological environment restoration system structure framework diagram based on a symbiotic competition mechanism is provided. DETAILED DESCRIPTION

[0047] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all 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.

[0048] The embodiments of the present application disclose a natural oyster reef ecological environment restoration method based on a symbiotic competition mechanism, as shown in Figure 1 The method comprises the following steps.

[0049] Collecting time series data, constructing symbiotic competition relationship graph based on collected data and graph neural network, calculating node degree centrality and intermediate centrality, and identifying potential key factors;

[0050] Introducing a random forest algorithm, taking potential key factor data as input and taking ecosystem health index as output label, training multiple decision trees and fusing results, and screening out sensitive and easily monitored indicator factors based on feature importance analysis of the random forest model;

[0051] Calculating the ratio of target area factors to the mean value of the healthy group, generating factor characteristic curves, judging the degree of factor deviation from the healthy state through time domain characteristics, and formulating basic repair measures according to the indicator factor threshold;

[0052] Introducing a reinforcement learning algorithm, taking repair measures as action strategies and taking the improvement degree of indicator factors as reward signals, and dynamically optimizing the combination of measures through real-time interaction with the ecosystem.

[0053] Further, the collection of time series data includes on-site sampling and laboratory analysis of water quality, plankton, swimming animals and oyster populations in the target area.

[0054] Further, the water quality of the target area includes pH value, dissolved oxygen, nitrogen and phosphorus content, plankton includes species, quantity, dominant species, swimming animals includes habitat, oyster population includes density, growth state, collects time series data for n consecutive years, and uses zero filling processing for less than n years, and selects the same healthy oyster reef ecosystem as the reference group, collects the same period data to generate the health mean fluctuation curve.

[0055] In the specific embodiment, the collection of time series data specifically includes:

[0056] 1. Determine the sampling time domain and range

[0057] Sampling time domain setting: taking the current year as the reference, the information collection time domain is determined as the continuous 5 years before the current year (such as 2020-2024), if the historical data of the target area or the reference group is less than 5 years, the missing data is supplemented with 0 value.

[0058] Spatial range demarcation: clearly define the boundary of the target degraded oyster reef area, such as using the tidal line and geographical markers as boundaries, and select a healthy oyster reef area similar to the target area in environmental conditions such as water temperature, salinity and bottom type as the reference group, the spatial distance between the two should not exceed 100 km in principle, to reduce the interference of environmental differences.

[0059] 2. Sampling point distribution of target area and reference group

[0060] Sampling principle: Grid sampling method + key area encryption method, divide 100m x 100m grid in target area and reference group, set 1 sampling point in each grid; increase 50% sampling points in oyster larvae dense area, water quality fluctuation area, such as near the estuary.

[0061] Sampling quantity: The number of sampling points in the target area and the reference group is not less than 30, which ensures the coverage of different microhabitats, such as intertidal zone and shallow water area.

[0062] 3. Field sampling operation

[0063] Water quality sampling:

[0064] Use professional marine water sampler, such as Niskin water sampler, to collect 0.5m depth of surface and 0.5m water sample from bottom, respectively into 500mL polyethylene bottle. Pretreatment: rinse 3 times with water sample; on-site determination of pH value, dissolved oxygen using portable water quality detector, the rest of the water sample is stored at low temperature, within 24 hours to send laboratory analysis of chemical oxygen demand, nitrogen and phosphorus content and other indicators.

[0065] Plankton sampling:

[0066] Use 25 plankton net, aperture 0.064mm, horizontal drag net at 0.5m below water surface, drag speed 1.5 knots, last for 3 minutes, collect phytoplankton sample; use 13 plankton net, aperture 0.112mm, collect zooplankton, sample into 50mL centrifuge tube, add 4% formaldehyde solution for fixation.

[0067] Swimming animal sampling:

[0068] Use multi-mesh gillnet, mesh size 2-10cm, set at each sampling point at night, choose 18:00-6:00 the next day, record the species, number, body length and weight of captured swimming animals; at the same time, observe their habitat habits such as habitat water layer and activity frequency through underwater camera.

[0069] Oyster population sampling:

[0070] Set 1m x 1m quadrat at each sampling point, count the number of oyster individuals in the quadrat and calculate the density; randomly select 30 individuals, measure shell length, shell height, soft body weight growth state indicators, and record the type of attached base and survival state.

[0071] 4. Laboratory analysis process

[0072] Water quality index analysis: chemical oxygen demand is determined by potassium dichromate method, total nitrogen is determined by alkaline potassium persulfate digestion-ultraviolet spectrophotometry, and total phosphorus is determined by ammonium molybdate spectrophotometry.

[0073] Plankton analysis: Phytoplankton samples are identified under a microscope and counted to calculate cell density and determine dominant species with a dominance > 0.02. Zooplankton samples are classified and counted under a dissecting microscope to determine species composition and quantity.

[0074] 5. Data processing and curve generation

[0075] Time series data integration:

[0076] Summarize the water quality and biological indicator data of the target area by year to form a 5-year time series dataset, and fill in 0 for years less than 5 years. The reference group is also processed with the same period data.

[0077] Fluctuation curve generation:

[0078] Draw the index fluctuation curve of the target area, such as dissolved oxygen fluctuation curve and oyster density fluctuation curve, with years as the horizontal axis and index values as the vertical axis. Calculate the average value of each year's index of the reference group to generate a healthy mean fluctuation curve for subsequent deviation analysis.

[0079] Further, the method comprises: based on the survey data, regarding the water quality factors and biological populations as nodes in a complex network, regarding the predation, symbiosis, and competition relationships as edges, constructing a symbiotic and competitive relationship graph through a graph neural network, calculating node degree centrality and intermediate centrality, and identifying potential key factors.

[0080] Further, the method comprises:

[0081] Regarding the water quality factors and biological populations in the ecological survey data as nodes in a complex network;

[0082] Defining the interaction relationship between the nodes as edges, and representing the edges by an adjacency matrix;

[0083] Assigning a feature vector to each node to form a feature matrix;

[0084] Aggregating the features of the nodes and their neighbors through the graph convolution layer of the GNN;

[0085] Taking the ecosystem health index as a label, optimizing the model parameters through back propagation, and after the training is completed, identifying the potential key factors that have a significant impact on the stability of the ecosystem through node feature importance scoring.

[0086] In the specific implementation, the specific steps of identifying the potential key factors comprise:

[0087] I. Graph structure construction

[0088] 1. Node Definition: Water quality factors such as pH, dissolved oxygen, and nitrogen and phosphorus content, and biological populations such as phytoplankton, zooplankton, nematodes, and oyster populations from ecological survey data are defined as nodes in a complex network, denoted as the set V = {v1, v2, ..., v...}. n}, where n is the total number of factors.

[0089] 2. Edge definition and weight quantification: The interaction relationships between nodes, such as predation, symbiosis, and competition, are defined as edges, using the adjacency matrix A∈R. n×n Indicates. If node v i With v j There is an interaction, A ij A represents the interaction strength (a non-zero value); if there is no interaction, then A... ij =0. Weighting rules: Symbiotic relationships, nutrient exchange between oysters and algae, are assigned positive values; competitive relationships, resource competition among phytoplankton, are assigned negative values. The absolute value indicates the intensity of the interaction, calibrated based on laboratory-measured interaction coefficients or literature data.

[0090] II. Graph Neural Network (GNN) Modeling Steps

[0091] 1. Input Feature Matrix Construction

[0092] For each node v i Assigning feature vector x i Examples of water quality factors include concentration values ​​and biological population density / biomass, forming a feature matrix, X∈R. n×d d is the feature dimension.

[0093] 2. Calculation of graph convolutional layers

[0094] The GNN aggregates the features of nodes and their neighbors through its graph convolutional layers, using the following formula:

[0095]

[0096] in: For node v i Hidden features at layer ι; N(i) represents node v i The neighbor set of node v; deg(i) is the set of nodes v. i The degree (number of edges); W (l) b (l) σ represents the weight matrix and bias term of the l-th layer; σ is the activation function, such as ReLU.

[0097] 3. Model Training and Key Factor Identification

[0098] The ecosystem health index is used as a reference health index for oyster reefs, and the model parameters are optimized through back propagation. After training, the node feature importance score, i.e. the gradient weight of the last layer output of the GNN, is used to identify potential key factors that significantly affect the stability of the ecosystem.

[0099] III. Node centrality index calculation

[0100] 1. Degree Centrality

[0101] The formula for measuring the direct connection strength of a node with other nodes is: where The number of edges of node v i , n-1 is the maximum possible number of connections, and the value is normalized. The higher the value, the more frequent the direct interaction between the node and other factors.

[0102] 2. Betweenness Centrality

[0103] The formula for measuring the bridge role of a node in the shortest path of other nodes is:

[0104]

[0105] where: σ st is the total number of shortest paths from node s to t; σ st (v i ) is the number of shortest paths from s to t that pass through node v i . The higher the value, the more critical the node is in regulating the interaction between other factors. The symbiotic competition relationship between ecological factors is quantified, and the feature learning of GNN and the centrality index are combined to accurately identify the key factors driving the stability of natural oyster reef ecosystems.

[0106] Further, the random forest algorithm is introduced, with potential key factor data as input and ecosystem health index as output label, training multiple decision trees and fusing the results, including:

[0107] The training set is generated by sampling from the preprocessed data set using the Bootstrap method;

[0108] Select water quality factors, plankton, nekton, and oyster population ecological factors as input variables to form the input feature matrix. The health status of the health group is used as the benchmark, and the health status of the target area is quantified as a health index, which is used as the output label of the model;

[0109] For each training set, a decision tree is trained;

[0110] For new samples, predict the health index through all decision trees, and take the average as the final output;

[0111] For each decision tree, calculate the sum of the Gini impurity reduction caused by each ecological factor in all split nodes, sum the impurity reduction of all decision trees, and then normalize to get the impact weight of the factor. The larger the value, the more significant the impact of the factor on the stability of the ecosystem.

[0112] In a specific embodiment, the random forest algorithm specifically includes:

[0113] I. Data preparation and preprocessing

[0114] 1. Input and output variable definition

[0115] The input variables are ecological factor data: select water quality factors: pH value, dissolved oxygen, nitrogen and phosphorus content, etc., plankton: species proportion, density, swimming animals: population, oyster population: density, growth rate, etc. Ecological factors constitute the input feature matrix X = {x1, x2, …, x m}, where m is the number of factors, and each factor contains time series data for 5 consecutive years, with zero filling for less than 5 years.

[0116] 2. The output variable is the ecosystem health index: taking the health group: the ecosystem state of the same health oyster reef as the benchmark, the health status of the target area is quantified as the health index y, with a value range of 0-1, 1 representing complete health, as the output label of the model.

[0117] 3. Data preprocessing

[0118] Missing value processing: use the zero filling method for less than 5 years of data to fill in 0 values for missing time series data;

[0119] Standardization: standardize the input features to eliminate the influence of dimension, the formula is: Where μ i is the mean of the i-th factor, and σ i is the standard deviation.

[0120] II. Random forest model training steps

[0121] 1. Bootstrap sampling to generate training sets: randomly sample N samples with replacement from the preprocessed data set with sample size N to generate the first training set D1; repeat this process K times, K is the number of decision trees, usually take 100-500, get K training sets D1, D2, …, D K .

[0122] 2. Build multiple decision trees

[0123] For each training set D k , train a decision tree T k , the process is as follows:

[0124] For each node of the tree, randomly select factors from all m ecological factors as candidate split features;

[0125] Use Gini impurity as the node split criterion, calculate the Gini index of each candidate feature, and select the feature and split threshold that maximizes the reduction in impurity:

[0126] Where S is the current node sample set, C is the number of health index categories, and p c is the proportion of samples belonging to category c after discretization.

[0127] The decision tree grows to the maximum depth or the node sample size is less than the threshold, and pruning is not performed.

[0128] 3. Model fusion, health index prediction

[0129] For a new sample, predict the health index through all K decision trees, and take the average as the final output: Where T k (X) is the prediction result of the kth decision tree for input X.

[0130] Three, ecological factor influence weight

[0131] Through the feature importance evaluation of random forest, the influence weight of each ecological factor on ecosystem stability is quantified, and the steps are as follows:

[0132] 1. Calculate the reduction of node impurity

[0133] For each decision tree T k , calculate the sum of the reduction of Gini impurity caused by each ecological factor x i in all split nodes, denoted as ΔGini k (x i ).

[0134] 2. Feature importance normalization

[0135] Sum the impurity reduction of all decision trees, and then normalize to get the influence weight w i of factor x i : Where w i ∈ [0, 1], and The larger the value, the more significant the factor's impact on ecosystem stability.

[0136] Four, model verification

[0137] Using out-of-bag data, the samples not selected by Bootstrap sampling are used to validate the model performance, and the mean square error (MSE) of the predicted health index and the actual health index is calculated:

[0138]

[0139] Where N is the number of out-of-bag samples, and the smaller the MSE, the higher the prediction accuracy of the model. out

[0140] The random forest algorithm is used to quantify the influence weight of each ecological factor on the stability of the natural oyster reef ecosystem, providing quantitative basis for screening key restoration targets.

[0141] Further, the feature importance analysis based on the random forest model screens sensitive and easily monitored indicator factors, including: normalizing the importance scores of all factors, and selecting the top 30% of factors as candidate sensitive factors, such as dissolved oxygen, phytoplankton dominant species proportion, and oyster larval survival rate; and finally screening indicator factors with high sensitivity and easy monitoring by combining field monitoring feasibility.

[0142] Further, the ratio of the target area factor to the mean value of the health group is calculated to generate a factor feature curve, and the degree of deviation of the factor from the healthy state is determined by the slope and fluctuation trend of the curve in the time domain, including:

[0143] The ratio of the target area factor to the mean value of the health group is calculated to generate a factor feature curve.

[0144] The factor feature curve is plotted with years as the horizontal axis and the ratio as the vertical axis.

[0145] The annual slope of the factor feature curve is calculated to reflect the rate of change of the deviation of the factor.

[0146] The factor feature curve is linearly fitted to obtain a trend line, and the long-term trend is determined by the slope.

[0147] The deviation degree is quantified by combining the ratio, slope, and trend, and the larger the ratio, the more serious the deviation of the factor from the healthy state.

[0148] In the specific implementation, the specific steps of screening indicator factors with high sensitivity and easy monitoring include:

[0149] I. Screening indicator factors based on random forest feature importance

[0150] 1. Feature importance quantification

[0151] ​Based on the trained random forest model, the input is the ecological factor data, and the output is the ecosystem health index. The feature importance score of each ecological factor is calculated. The core index is the reduction of impurity when the decision tree is split, and the formula is:

[0152]

[0153] Where f is the ecological factor, K is the number of decision trees, T k is the kth decision tree, and ΔGini(t) is the reduction of Gini impurity after splitting node t by factor f. The calculation formula is the same as the Gini impurity in the random forest training.

[0154] 2. Screening sensitive factors

[0155] The importance scores of all factors are normalized to make the sum equal to 1, and the top 30% of the scores are selected as candidate sensitive factors.

[0156] 3. Verification of easy monitoring

[0157] For candidate sensitive factors, combined with the feasibility of field monitoring, such as whether it can be collected in real time through sensors, laboratory analysis cost, etc., finally the indicator factors with high sensitivity and easy monitoring are selected, such as dissolved oxygen, dominant species of phytoplankton, etc.

[0158] II. Ratio calculation of target area factors and health group mean and generation of characteristic curve

[0159] 1. Data alignment and mean calculation

[0160] Select the health group of the same period as the target area, with 5 consecutive years, and fill in zero for less than 5 years. The annual mean of the ecological factor data of the health group of the same type of healthy oyster reef is calculated, denoted as μ f (y), f is the factor, y is the year, y = 1, 2,..., 5. Extract the factor data of the same period in the target area, denoted as x f (y).

[0161] 2. Factor ratio calculation

[0162] Calculate the simultaneous domain ratio of the target area factor and the health group mean, and generate the basic data of the factor characteristic curve, the formula is:

[0163] If μ f (y) = 0, then r f (y) = 0);

[0164] Where r f (y) is the ratio of factor f in year y, r f (y) = 1 indicates consistency with the health status, and rf (y) > 1 indicates deviation from the healthy state.

[0165] 3. Generating factor characteristic curve

[0166] Plotting the curve with year y as the horizontal axis and the ratio r f (y) as the vertical axis. That is, the factor characteristic curve.

[0167] III. Judging the degree of deviation of the factor from the healthy state based on time domain characteristics

[0168] 1. Curve slope calculation

[0169] Calculate the annual slope of the characteristic curve C f , which reflects the change rate of the degree of deviation of the factor, and the formula is:

[0170]

[0171] Where, k f (y) > 0 indicates that the degree of deviation is expanding, and k f (y) < 0 indicates that the degree of deviation is shrinking.

[0172] 2. Fluctuation trend analysis

[0173] Linearly fit the curve C f to obtain the trend line Determine the long-term trend by the slope a f :

[0174] a f > 0: The degree of deviation of the factor from the healthy state shows a long-term expanding trend;

[0175] a f ≈ 0: The degree of deviation is stable;

[0176] a f < 0: The degree of deviation shows a long-term shrinking trend.

[0177] 3. Comprehensive score of the degree of deviation

[0178] Quantify the degree of deviation D f = max(r f (y)) · (1 + |a f |) by combining the size of the ratio, the slope, and the trend.

[0179] Where, max(r f (y)) reflects the maximum deviation amplitude, |a f | reflects the trend intensity, and D fThe larger the value, the more severe the deviation of the factor from a healthy state. Key indicator factors are screened from ecological factors, and their degree of deviation from a healthy state is precisely quantified to provide a targeted basis for subsequent remediation measures.

[0180] Furthermore, the basic remediation measures formulated based on the indicator factor threshold include:

[0181] Based on the mean values ​​of indicator factors and ecological safety thresholds of the healthy group, the health threshold range of each indicator factor is determined. According to the deviation between the current value of the indicator factor in the target area and the threshold, corresponding remediation measures are triggered, including: adjusting the aquaculture density to control it at 60%-80% of the environmental carrying capacity, introducing functional organisms, such as filter-feeding shellfish to regulate plankton, controlling pollution sources and reducing nitrogen and phosphorus inputs, etc.

[0182] Furthermore, this also includes: introducing reinforcement learning algorithms, using remediation measures as action strategies, using the degree of improvement of indicator factors as reward signals, and dynamically optimizing the combination of measures through real-time interaction with the ecosystem, such as adjusting the oxygenation frequency according to dissolved oxygen fluctuations.

[0183] In a specific implementation, the formulation of basic remediation measures specifically includes:

[0184] I. Formulation of Basic Repair Measures

[0185] 1. Determine the health threshold of indicator factors

[0186] Basic restoration measures are formulated based on the health threshold of indicator factors, taking into account the deviation between their current values ​​and the threshold, and then designing targeted adjustment strategies. Referring to the healthy group (i.e., the mean values ​​of indicator factors and ecological safety thresholds of similar healthy oyster reefs), the health threshold range T for each indicator factor is determined. f =[T f,min T f,max ], where f represents an indicator factor, such as dissolved oxygen, nitrogen and phosphorus content, and stocking density.

[0187] Dissolved oxygen (DO) health threshold T DO = [5, 8] mg / L;

[0188] The environmental carrying capacity threshold T for aquaculture density 密度 = [0.6C, 0.8C], where C is the environmental carrying capacity, i.e. the maximum sustainable stocking density of the healthy group;

[0189] Total nitrogen (TN) threshold T TN = [0.5, 1.0] mg / L, Total Phosphorus (TP) Threshold T TP = [0.05, 0.1] mg / L.

[0190] 2. Develop basic measures based on factor bias

[0191] According to the target area indicator factor current value x f deviation from the threshold T f , trigger the corresponding repair measures:

[0192] Measure 1: Adjust the breeding density If the current breeding density D>0.8C, reduce the density to 0.6C~0.8C, the adjustment formula is:

[0193] ΔD=D-min(0.8C,D×0.9); That is, the maximum reduction of 10% at a time, to ensure not less than 0.6C.

[0194] Measure 2: Release functional organisms If the proportion of phytoplankton dominant species P>T P , max, such as the proportion of green algae exceeds 60%, release filter-feeding mollusks, such as Sinonovacula constricata, the release amount is positively correlated with the deviation:

[0195] N=k×(P-T P , max)×S;

[0196] Wherein, k is the proportionality coefficient, such as 0.5ind / m 2 , S is the area of the repair area.

[0197] Measure 3: Control source and intercept sewage If the total nitrogen / total phosphorus x TN / TP >T TN / TP，max , calculate the input amount to be reduced:

[0198] ΔL=max(0,x TN / TP -T TN / TP,max )×V×1.2;

[0199] Wherein, V is the volume of the area water body, and 1.2 is the safety factor to ensure that the reduction is below the threshold.

[0200] In a specific embodiment, the reinforcement learning algorithm specifically includes:

[0201] The repair system is regarded as an intelligent agent, which dynamically optimizes the combination of measures through real-time interaction with the ecosystem and takes the improvement degree of the indicator factor as feedback.

[0202] 1. Definition of the core elements of reinforcement learning

[0203] State (State): A vector s t =[x f1 (t), x f2 (t),..., k f1 (t), k f2 (t),...] composed of the current value and the trend of the indicator factor, wherein x f (t) is the factor value at time t, and k f(t) represents its slope, reflecting the trend of change. The formula is the same as the time-domain characteristic slope mentioned earlier.

[0204] Action: A combination of remedial measures t = [ΔD(t), N(t), ΔL(t)], which represents the specific values ​​of the aquaculture density adjustment, biological input, and pollution interception, selected from the preset discretized space, such as ΔD∈{0, 5%, 10%}.

[0205] Reward: A quantitative indicator that measures the degree of improvement of the indicator factor, designed based on the change in the deviation between the factor and the threshold.

[0206] R t =∑ f w f ×max(0,(d f,t-1 -d f,t ));

[0207] in, The factor f at time t and the mean of the threshold The deviation, w f The factor importance weights are derived from the feature importance of the random forest, ∑w f =1. If the deviation decreases, R t A positive value indicates a positive deviation; a larger deviation indicates a negative deviation.

[0208] 2. Model Training and Policy Optimization

[0209] Value function (Q-value): represents the state s of a given condition. t Take action a t The expected cumulative reward is denoted as Q(s). t , a t ).

[0210] Q-value update formula:

[0211]

[0212] Where α∈(0,1) is the learning rate, such as 0.3, and γ∈(0,1) is the discount factor, such as 0.9, which balances immediate and future rewards. a′ Q(s t+1 , a′) is the next state s t+1 The maximum expected reward.

[0213] 3. Real-time dynamic adjustment of processes

[0214] 1. Status Monitoring: The system's monitoring module collects indicator factor data every 7 days and updates the status status. t ;

[0215] 2. Action selection: Based on the current Q value, an epsilon-greedy strategy is used to select the action, with a probability of 1-epsilon to select the optimal action and epsilon probability to explore randomly, such as epsilon = 0.1;

[0216] 3. Reward calculation: After performing the action, the reward R of the next time t+1 is calculated t+1 ;

[0217] 4. Strategy update: The Q table is iteratively optimized using the Q value update formula until the strategy converges, such as the reward fluctuation is less than 5% in the last 10 iterations. Through the above steps, the basic repair measures can quickly respond based on the indicator factor threshold, while the reinforcement learning dynamically optimizes the measure combination to better adapt to the real-time changes of the ecosystem, improving the repair efficiency.

[0218] In a specific embodiment, a natural oyster reef ecological environment repair system based on a symbiotic competition mechanism, as shown in Figure 2 , includes:

[0219] Collection module: for collecting time series data;

[0220] Potential key factor identification module: for constructing a symbiotic competition relationship graph based on collected data and graph neural networks, calculating node degree centrality and intermediate centrality, and identifying potential key factors;

[0221] Screening module: for introducing a random forest algorithm, taking potential key factor data as input and ecosystem health index as output label, training multiple decision trees and fusing results, and screening out sensitive and easily monitored indicator factors based on random forest model feature importance analysis;

[0222] Formulation module: for calculating the ratio of target area factors to health group mean, generating factor feature curves, judging the degree of factor deviation from healthy state through time domain features, and formulating basic repair measures based on indicator factor threshold;

[0223] Optimization module: for introducing a reinforcement learning algorithm, taking repair measures as action strategy and indicator factor improvement degree as reward signal, and dynamically optimizing measure combination through real-time interaction with the ecosystem.

[0224] In a specific embodiment, taking a certain degraded oyster reef area as an example:

[0225] Collect water quality and biological data from 2018 to 2022 in this area, and select a healthy oyster reef 50 km away as a reference group;

[0226] Analysis using the random forest algorithm found that dissolved oxygen (weight 35%), phytoplankton dominant species proportion (weight 25%), and oyster larvae density (weight 20%) are key indicator factors;

[0227] Initial scheme set the culture density to reduce 20%, and put filter-feeding shellfish;

[0228] After 3 months, monitoring showed that the dissolved oxygen was still 15% lower than the healthy group, the system automatically triggered the oxygenation equipment, and adjusted the amount of shellfish put in;

[0229] After 1 year of continuous repair, the survival rate of oysters in the target area increased by 50%, and the water quality indicators returned to 85% of the level of the healthy group.

[0230] The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other. For the device disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method part.

[0231] The above description of the disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to the embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for repairing the ecological environment of natural oyster reef based on the symbiotic competition mechanism, characterized in that, The method comprises the following steps: Collecting time series data, constructing a symbiotic competition relationship graph based on the collected data and a graph neural network, calculating node degree centrality and intermediate centrality, and identifying potential key factors; Introducing a random forest algorithm, taking the potential key factor data as input and taking the ecosystem health index as output label, training multiple decision trees and fusing the results, and screening out sensitive and easily monitored indicator factors based on the feature importance analysis of the random forest model; Calculate the ratio of the target area factor to the mean value of the health group to generate a factor characteristic curve, determine the degree of deviation of the factor from the healthy state through time domain characteristics, and develop basic repair measures according to the indicator factor threshold. Introducing a reinforcement learning algorithm, taking the repair measures as action strategies and taking the improvement degree of the indicator factor as the reward signal, and dynamically optimizing the measure combination through real-time interaction with the ecosystem.

2. The method for ecological environment restoration of natural oyster reef based on the symbiotic competition mechanism according to claim 1, characterized in that, The collected time series data includes on-site sampling and laboratory analysis of water quality, plankton, swimming animals and oyster populations in the target area.

3. The method according to claim 2, wherein, The water quality of the target area includes pH value, dissolved oxygen, nitrogen and phosphorus content, plankton includes species, quantity, dominant species, swimming animals includes habitat, oyster population includes density and growth state, and continuous n-year time series data is collected.

4. The method for ecological environment restoration of natural oyster reef based on mutualistic competition mechanism according to claim 1, characterized in that, Based on the collected data and a graph neural network, a symbiotic competition relationship graph is constructed, node degree centrality and intermediate centrality are calculated, and potential key factors are identified.

5. The method for ecological environment restoration of natural oyster reef based on mutualistic competition mechanism according to claim 4, characterized in that, The identification of potential key factors includes: Defining water quality factors and biological populations in ecological survey data as nodes in a complex network; The interaction relationship between nodes is defined as an edge, represented by an adjacency matrix; Assign a feature vector to each node to form a feature matrix; Through the graph convolution layer of GNN, the features of the node and its neighbors are aggregated; Taking the ecosystem health index as the label, the model parameters are optimized through back propagation, and after training, the node feature importance score is used to identify potential key factors that have a significant impact on the stability of the ecosystem.

6. The method for ecological environment restoration of natural oyster reef based on mutualistic competition mechanism according to claim 1, characterized in that, The introduction of the random forest algorithm, taking the potential key factor data as input and taking the ecosystem health index as output label, training multiple decision trees and fusing the results includes: Bootstrap method is used to sample training set from preprocessed data set; Select water quality factors, plankton, swimming animals and oyster population ecological factors as input variables to form an input feature matrix, and take the health status of the target area as the health index as the output label of the model; For each training set, train a decision tree; For new samples, predict the health index through all decision trees and take the average value as the final output. For each decision tree, the sum of the Gini impurity reduction caused by each ecological factor in all split nodes is calculated, the impurity reduction of all decision trees is summed up, and the influence weight of the factor is normalized to obtain the influence weight of the factor. The greater the value, the more significant the influence of the factor on the stability of the ecosystem.

7. The method according to claim 1, wherein the method is characterized by, The feature importance analysis based on the random forest model screens sensitive and easily monitored indicator factors, including: normalizing the importance scores of all factors, and selecting the top 30% of factors with the highest scores as candidate sensitive factors; for the candidate sensitive factors, combined with the feasibility of field monitoring, finally screening out indicator factors with high sensitivity and easy monitoring.

8. The method for ecological environment restoration of natural oyster reef based on mutualistic competition mechanism according to claim 1, characterized in that, The calculation of the ratio of the target area factor to the mean value of the healthy group generates a factor feature curve, which reflects the degree of deviation of the factor from the healthy state through time domain characteristics, including: Calculating the ratio of the target area factor to the mean value of the healthy group in the same time domain to generate the basic data of the factor feature curve; Taking the year as the horizontal axis and the ratio as the vertical axis, the factor feature curve is drawn; Calculate the annual slope of the factor feature curve to reflect the change rate of the deviation degree of the factor; Linear fitting is performed on the factor feature curve to obtain a trend line, and the long-term trend is judged by the slope; Combined with the ratio, the slope and the trend, the deviation degree is quantified, and the greater the value, the more serious the deviation of the factor from the healthy state.

9. The method for ecological environment restoration of natural oyster reef based on mutualistic competition mechanism according to claim 1, characterized in that, The development of basic restoration measures according to the indicator factor threshold includes: Referring to the mean value of the indicator factor of the healthy group and the ecological safety threshold, the health threshold range of each indicator factor is determined, and the corresponding restoration measures are triggered according to the deviation of the current value of the indicator factor of the target area from the threshold.

10. A natural oyster reef ecological environment restoration system based on a symbiotic competition mechanism, characterized in that, It includes: Data acquisition module: used for collecting time series data; Potential key factor identification module: used for constructing a symbiotic competition relationship graph based on collected data and graph neural network, calculating node degree centrality and intermediate centrality, and identifying potential key factors; Screening module: used for introducing a random forest algorithm, taking potential key factor data as input and taking ecosystem health index as output label, training multiple decision trees and fusing results, screening sensitive and easily monitored indicator factors based on random forest model feature importance analysis; Development module: used for calculating the ratio of the target area factor to the mean value of the healthy group, generating a factor feature curve, judging the degree of deviation of the factor from the healthy state through time domain characteristics, and developing basic restoration measures according to the indicator factor threshold; Optimization module: used for introducing a reinforcement learning algorithm, taking restoration measures as action strategy and taking indicator factor improvement degree as reward signal, and dynamically optimizing the combination of measures through real-time interaction with the ecosystem.

Images