A method, medium and system for assessing the risk level of Spartina alterniflora invasion

Abstract

The present invention provides a Spartina alterniflora invasion risk level assessment method, medium and system, belonging to the technical field of Spartina alterniflora invasion risk assessment. The present invention establishes intertidal zone environmental monitoring equipment to obtain multidimensional environmental data within a tidal cycle, constructs a tidal inundation impact matrix and a tidal emergent growth matrix, adopts a matrix multivariate decomposition technology combined with a minimum spanning tree algorithm to extract key environmental factors and growth factors, utilizes a multi-layer neural network to construct a risk assessment model with an integrated gated weight adjustment mechanism, dynamically adjusts network parameters according to a tidal change influence coefficient, realizes accurate quantitative assessment of Spartina alterniflora invasion risk, establishes a risk early warning mechanism with differentiated monitoring frequencies, realizes continuous improvement of the risk assessment model based on iterative optimization of the monitoring frequency adjustment coefficient, and solves the technical problem of insufficient accuracy in risk assessment of Spartina alterniflora invasion under a tidal environment.

Description

Technical Field

[0001] The present invention belongs to the technical field of Spartina alterniflora invasion risk assessment, and in particular relates to a Spartina alterniflora invasion risk level assessment method, medium and system. Background Art

[0002] Risk assessment of Spartina alterniflora invasion is an important technical means of marine ecological protection. Traditional risk assessment methods are mainly based on static environmental parameter monitoring and single growth indicator analysis. By setting up fixed monitoring points to collect data such as vegetation coverage and soil physical and chemical properties, risk levels are divided based on expert evaluation or simple mathematical models. These methods have been widely used in coastal wetland ecological monitoring, marine environmental protection, alien species prevention and control and other fields.

[0003] Traditional technologies have obvious flaws, mainly manifested in the inability to accurately capture the dynamic impact of tidal periodic changes on the growth of Spartina alterniflora, ignoring the complex interaction of environmental factors during tidal inundation and emergence, and lacking precise quantitative analysis of plant growth status under different tidal conditions, resulting in large deviations between risk assessment results and actual invasion situations.

[0004] In current intertidal ecological monitoring, due to the high dynamism and complexity of the tidal environment, traditional static monitoring methods find it difficult to establish an accurate correlation between environmental factors and the risk of biological invasion, and cannot achieve accurate prediction and dynamic assessment of the invasion process of Spartina alterniflora. Summary of the Invention

[0005] In view of this, the present invention provides a method, medium and system for assessing the risk level of Spartina alterniflora invasion, which can solve the technical problem of insufficient accuracy in risk assessment of Spartina alterniflora invasion under tidal environment in the prior art.

[0006] The present invention is achieved as follows: In a first aspect, the present invention provides a method for assessing the risk level of Spartina alterniflora invasion, comprising: establishing an intertidal zone environmental monitoring device, obtaining water level change data, salinity data, temperature data, and light intensity data within a tidal cycle, forming a tidal inundation spatiotemporal data set, and constructing a tidal inundation impact matrix; obtaining vegetation coverage data, soil moisture data, root depth data, and stem density data in the Spartina alterniflora distribution area during the emergence period after the tide recedes, and establishing a tidal emergence growth matrix; performing matrix multivariate decomposition on the tidal inundation impact matrix and the tidal emergence growth matrix, and constructing a factor correlation graph, with environmental factors and growth factors as nodes of the graph, and correlation coefficients between factors as edges. The weights of the factors were calculated, and the minimum spanning tree algorithm was used to solve the optimal factor connection path, remove redundant factor connections, retain the correlation between key factors, extract the main environmental factors and growth factors, calculate the weight coefficients and correlation coefficients of each factor, and form the environmental factor feature vector and the growth factor feature vector; a multi-layer neural network was used to construct a risk assessment network model, and the environmental factor feature vector and the growth factor feature vector were input. The network parameters were dynamically adjusted through the gated weight adjustment mechanism, and the tidal change influence coefficient was used as the input parameter of the gated weight adjustment mechanism to output the risk probability value of Spartina alterniflora invasion; the invasion risk level was divided according to the risk probability value; a risk early warning mechanism was established, and differentiated monitoring frequencies were implemented for areas with different risk levels.

[0007] Among them, the tidal inundation impact matrix is ​​specifically a multidimensional data matrix constructed based on water level changes and environmental parameter changes during the tidal cycle. The rows of the matrix represent different tidal height levels, the columns represent different environmental parameter types, and the matrix element values ​​represent the degree of influence of the corresponding environmental parameters on the growth of Spartina alterniflora at the corresponding tidal height. The degree of influence is represented by a standardized numerical value ranging from 0 to 1, where 0 represents no influence and 1 represents complete inhibition of growth.

[0008] Among them, the tidal emergence growth matrix is ​​specifically a data matrix that records the growth status of Spartina alterniflora in different emergence time periods after the tide recedes. The rows of the matrix represent different emergence time periods, the columns represent different growth index parameters, and the matrix element values ​​represent the measured values ​​of the corresponding growth indicators in the corresponding emergence time periods. The growth indicators include vegetation coverage, soil moisture content, root depth, and stem density. All indicators are converted into a numerical range of 0 to 1 through standardization.

[0009] Among them, the factor correlation graph is specifically an undirected weighted graph constructed with environmental factors and growth factors as nodes and the correlation coefficients between factors as edge weights. Each node in the graph represents a factor variable, and the weight of the edge represents the strength of the correlation between the two factors. The minimum weight path connecting all factor nodes is solved by the minimum spanning tree algorithm to remove redundant factor connection relationships.

[0010] Among them, the tidal change impact coefficient is specifically a comprehensive impact index calculated based on the water level change amplitude, change frequency, and change rate within the tidal cycle, which is used to quantify the impact of tidal changes on the growth environment of Spartina alterniflora. The calculation formula is the water level change amplitude multiplied by 0.4 plus the change frequency multiplied by 0.3 plus the change rate multiplied by 0.3 plus the tidal impact calculation error term.

[0011] Among them, the specific structure of the risk assessment network model is a multi-layer perceptron network including an input layer, a hidden layer and an output layer. The input layer receives the environmental factor feature vector and the growth factor feature vector with a dimension of 128 dimensions. The first hidden layer contains 256 neurons, the second hidden layer contains 128 neurons, the third hidden layer contains 64 neurons, and the output layer contains 1 neuron to output the risk probability value.

[0012] Among them, the step of dividing the intrusion risk level according to the risk probability value is specifically to set the low risk threshold to 0.3, the medium risk threshold to 0.6, and the high risk threshold to 0.8, that is, when the risk probability value ∈ [0, 0.3), it is determined to be a low risk level, when the risk probability value ∈ [0.3, 0.6), it is determined to be a medium risk level, when the risk probability value ∈ [0.6, 0.8), it is determined to be a high risk level, and when the risk probability value ∈ [0.8, 1], it is determined to be an extremely high risk level.

[0013] Among them, after the step of establishing the risk warning mechanism, it also includes recalculating the monitoring frequency adjustment coefficient based on the newly obtained monitoring data, inputting the monitoring frequency adjustment coefficient into the gating weight function to recalculate the gating parameter adjustment coefficient, using the updated gating parameter adjustment coefficient to adjust the parameters of the risk assessment network model, re-executing to obtain the updated Spartina alterniflora invasion risk probability value, and re-dividing the risk level based on the updated risk probability value.

[0014] Among them, the matrix multivariate decomposition is a mathematical method that uses the singular value decomposition method combined with the minimum spanning tree algorithm to reduce the dimension of the tidal inundation impact matrix and the tidal emergence growth matrix and optimize the selection of factors. By decomposing the original matrix into the product form of three matrices, the main environmental factors and growth factors are extracted, and the eigenvector after dimensionality reduction retains more than 90% of the information of the original data.

[0015] Among them, the gating weight adjustment mechanism specifically calculates the gating threshold based on the tidal change influence coefficient, the variance and mean of the current input feature. When the gating threshold is less than 0.4, the linear weight adjustment function is used; when the gating threshold is between 0.4 and 0.7, the sigmoid weight adjustment function is used; when the gating threshold is greater than 0.7, the tanh weight adjustment function is used.

[0016] Among them, the differentiated monitoring frequency is specifically: low-risk areas are monitored once a month, medium-risk areas are monitored once a week, high-risk areas are monitored once a day, and extremely high-risk areas are monitored in real time.

[0017] Among them, the monitoring frequency adjustment coefficient is specifically an adjustment parameter calculated based on the current risk level and historical monitoring results, which is used to quantify the impact of changes in monitoring frequency on the risk assessment model. The calculation formula is the current risk level value multiplied by 0.6 plus the historical monitoring accuracy multiplied by 0.4. The monitoring frequency adjustment coefficient serves as the input parameter of the gating weight function.

[0018] Among them, the gating weight function specifically adjusts the parameters of the neural network gating mechanism. The input includes the information entropy of the current layer input data, the variance ratio of the current layer input data, the correlation coefficient of the current layer input data, the tidal change influence coefficient, and the monitoring frequency adjustment coefficient. The output is the gating parameter adjustment coefficient.

[0019] Among them, the gating weight function calculates the segmentation value based on five input parameters. The segmentation value calculation formula is the information entropy multiplied by 0.3 plus the variance ratio multiplied by 0.25 plus the correlation coefficient multiplied by 0.25 plus the tidal change influence coefficient multiplied by 0.1 plus the monitoring frequency adjustment coefficient multiplied by 0.1. When the segmentation value is less than 0.25, a conservative weight adjustment function is used. When the segmentation value is between 0.25 and 0.65, a balanced weight adjustment function is used. When the segmentation value is greater than 0.65, an aggressive weight adjustment function is used.

[0020] The iterative process is repeated until the risk probability value changes by less than 0.02 for three consecutive iterations, thereby forming a stable dynamic risk assessment report.

[0021] Among them, before building the risk assessment network model, the step of establishing a training data set for the risk assessment network model is also included. Specifically, the distribution data of Spartina alterniflora in 30 typical intertidal areas are collected, and each area is continuously monitored for 24 months to obtain tidal cycle data, environmental parameter data, and vegetation growth data to form an original monitoring data set. The original data is quality checked and outlier processing is performed to construct a training data set containing 50,000 samples.

[0022] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions. When the program instructions are run in a computer, the program instructions are used to execute the above-mentioned method for assessing the risk level of invasion by Spartina alterniflora.

[0023] The third aspect of the present invention provides a system for assessing the risk level of invasion of Spartina alterniflora, comprising the above-mentioned computer-readable storage medium, wherein the system is any one of a computer, a server, and a single-chip microcomputer, and the computer-readable storage medium is arranged in the system, and the system is provided with a microprocessor for executing program instructions stored in the computer-readable storage medium.

[0024] The present invention adopts a dual-matrix data acquisition method combining a tidal inundation impact matrix and a tidal emergence growth matrix, extracts key environmental factors and growth factors through matrix multivariate decomposition technology, and constructs a multi-layer neural network risk assessment model based on a gated weight adjustment mechanism, which solves the technical defect that traditional methods cannot accurately quantify the dynamic impact of tides. The present invention realizes the all-round dynamic perception of the growth environment of Spartina alterniflora by establishing a multidimensional environmental parameter monitoring system within the tidal cycle, combines the singular value decomposition and minimum spanning tree algorithm to optimize factor selection, effectively removes data redundancy and noise interference, and improves the feature extraction quality of risk assessment and the reliability of model input. The gated weight adjustment mechanism established by the present invention can dynamically adjust network parameters according to the tidal change influence coefficient, realizes the accurate quantitative assessment of the risk of Spartina alterniflora invasion under different tidal conditions, and solves the technical problem of insufficient accuracy of Spartina alterniflora invasion risk assessment under tidal environment through iterative optimization of dynamic monitoring frequency adjustment mechanism. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 is a flow chart of the method of the present invention.

[0026] Figure 2 This is a schematic diagram of the neural network structure for intrusion risk assessment involved in the present invention. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0028] like Figure 1 FIG. 1 is a flow chart of a method for assessing the risk level of Spartina alterniflora invasion provided by the first aspect of the present invention. The method comprises the following steps:

[0029] S01. Establish intertidal zone environmental monitoring equipment to obtain water level change data, salinity data, temperature data, and light intensity data within the tidal cycle to form a tidal inundation spatiotemporal dataset and construct a tidal inundation impact matrix, which records the inundation state and environmental parameter changes of the Spartina alterniflora growth area at different tidal heights;

[0030] S02. During the emplacement period after the tide recedes, obtaining vegetation coverage data, soil moisture data, root depth data, and stem density data of the Spartina alterniflora distribution area, and establishing a tidal emplacement growth matrix, wherein the tidal emplacement growth matrix reflects the growth state and diffusion characteristics of Spartina alterniflora during different emplacement time periods;

[0031] S03. Perform matrix multivariate decomposition on the tidal inundation impact matrix and the tidal emergence growth matrix to construct a factor correlation graph. The graph uses environmental factors and growth factors as nodes and the correlation coefficients between factors as edge weights. A minimum spanning tree algorithm is used to find the optimal factor connection path. Redundant factor connections are removed, while retaining the key factor correlation relationships. The main environmental factors and growth factors are extracted, and the weight coefficients and correlation coefficients of each factor are calculated to form the environmental factor feature vector and the growth factor feature vector.

[0032] S04. A risk assessment network model is constructed using a multi-layer neural network. The environmental factor feature vector and the growth factor feature vector are input. The network parameters are dynamically adjusted through a gated weight adjustment mechanism. The tidal change influence coefficient is used as an input parameter of the gated weight adjustment mechanism, and the Spartina alterniflora invasion risk probability value is output.

[0033] S05. Divide the intrusion risk level according to the risk probability value, and set the low risk threshold to 0.3, the medium risk threshold to 0.6, and the high risk threshold to 0.8. That is, when the risk probability value ∈ [0, 0.3), it is judged as a low risk level; when the risk probability value ∈ [0.3, 0.6), it is judged as a medium risk level; when the risk probability value ∈ [0.6, 0.8), it is judged as a high risk level; when the risk probability value ∈ [0.8, 1], it is judged as an extremely high risk level;

[0034] S06. Establish a risk warning mechanism and implement differentiated monitoring frequencies for areas with different risk levels. Low-risk areas are monitored once a month, medium-risk areas are monitored once a week, high-risk areas are monitored once a day, and extremely high-risk areas are monitored in real time. The monitoring frequency adjustment coefficient is recalculated based on the newly obtained monitoring data, and the monitoring frequency adjustment coefficient is input into the gating weight function to recalculate the gating parameter adjustment coefficient. The updated gating parameter adjustment coefficient is used to adjust the parameters of the risk assessment network model, and step S04 is re-executed to obtain an updated risk probability value of Spartina alterniflora invasion. Based on the updated risk probability value, step S05 is re-executed to perform risk level classification, and the iterative process is repeated until the risk probability value changes by three consecutive iterations is less than 0.02, forming a stable dynamic risk assessment report.

[0035] The tidal inundation impact matrix is ​​a multidimensional data matrix constructed based on water level changes and environmental parameter changes within a tidal cycle. The rows of the matrix represent different tidal height levels, the columns represent different environmental parameter types, and the matrix element values ​​represent the degree of influence of the corresponding environmental parameters on the growth of Spartina alterniflora at the corresponding tidal height. The degree of influence is represented by a standardized numerical value ranging from 0 to 1, where 0 represents no influence and 1 represents complete inhibition of growth.

[0036] The tidal emergence growth matrix is ​​a data matrix that records the growth status of Spartina alterniflora in different emergence time periods after the tide recedes. The rows of the matrix represent different emergence time periods, the columns represent different growth index parameters, and the matrix element values ​​represent the measured values ​​of the corresponding growth indicators in the corresponding emergence time periods. The growth indicators include vegetation coverage, soil moisture content, root depth, and stem density. All indicators are converted into a numerical range of 0 to 1 through standardization.

[0037] The matrix multivariate decomposition is a mathematical method that uses the singular value decomposition method combined with the minimum spanning tree algorithm to perform dimensionality reduction and factor optimization selection on the tidal inundation impact matrix and the tidal emergent growth matrix. By decomposing the original matrix into the product form of three matrices, the main environmental factors and growth factors are extracted. The eigenvector after dimensionality reduction retains more than 90% of the information of the original data, while removing redundant information and noise interference, providing high-quality input features for the subsequent risk assessment network model.

[0038] The factor correlation graph is an undirected weighted graph constructed with environmental factors and growth factors as nodes and the correlation coefficients between factors as edge weights. Each node in the graph represents a factor variable, and the weight of the edge represents the strength of the correlation between the two factors. The minimum spanning tree algorithm is used to solve the minimum weight path connecting all factor nodes, remove redundant factor connection relationships, and retain the most important factor association structure.

[0039] The tidal change impact coefficient is a comprehensive impact index calculated based on the water level change amplitude, change frequency, and change rate within the tidal cycle. It is used to quantify the impact of tidal changes on the growth environment of Spartina alterniflora. The calculation formula is the water level change amplitude multiplied by 0.4 plus the change frequency multiplied by 0.3 plus the change rate multiplied by 0.3. The tidal change impact coefficient is used as an input parameter of the gated weight adjustment mechanism.

[0040] like Figure 2As shown, the specific structure of the risk assessment network model is a multi-layer perceptron network including an input layer, a hidden layer and an output layer. The input layer receives the environmental factor feature vector and the growth factor feature vector, with a dimension of 128 dimensions. The first hidden layer contains 256 neurons, the second hidden layer contains 128 neurons, the third hidden layer contains 64 neurons, and the output layer contains 1 neuron to output the risk probability value. The network integrates a gating weight adjustment mechanism, which calculates the gating threshold based on the tidal change influence coefficient, the variance and mean of the current input feature. When the gating threshold is less than 0.4, a linear weight adjustment function is used. When the gating threshold is between 0.4 and 0.7, a sigmoid weight adjustment function is used. When the gating threshold is greater than 0.7, a tanh weight adjustment function is used. Different weight adjustment functions are used to adapt to feature patterns of different complexities.

[0041] The steps for establishing the training data set of the risk assessment network model specifically include collecting the distribution data of Spartina alterniflora in 30 typical intertidal areas, continuously monitoring each area for 24 months, obtaining tidal cycle data, environmental parameter data, and vegetation growth data to form an original monitoring data set, performing quality inspection and outlier processing on the original data, eliminating data records with missing values ​​exceeding 20%, standardizing and feature engineering the remaining data, extracting time series features, spatial distribution features, and environmental association features, and constructing a training data set containing 50,000 samples, of which 70% are used for model training, 20% are used for verification, and 10% are used for testing. Each sample contains an input feature vector and a corresponding risk level label.

[0042] The steps of training the risk assessment network model specifically include using the Adam optimizer to optimize the model parameters, setting the learning rate to 0.001, the batch size to 32, and the maximum number of training rounds to 1000 rounds, and using the cross-entropy loss function to calculate the error between the predicted value and the true label. In each training batch, the statistics of the input features of the current batch are first calculated, including the mean, variance, maximum value, and minimum value. The gating threshold is calculated based on the statistics and the tidal change influence coefficient, and the corresponding weight adjustment function is selected according to the gating threshold. The network parameters are updated, and the model performance is evaluated on the validation set. When the validation set loss no longer decreases for 10 consecutive rounds, the training is stopped in advance, and the optimal model parameters are saved. The accuracy of the finally trained model on the test set reaches more than 92%, and the average error of the risk level prediction is less than 0.05.

[0043] The weight adjustment function is used to dynamically adjust the weight of the neural network based on the input feature statistics and the tidal change influence coefficient. The input includes the mean of the input feature vector, the variance of the input feature vector, the skewness of the input feature vector, the kurtosis of the input feature vector, and the tidal change influence coefficient, and the output is the neural network weight adjustment coefficient. The weight adjustment function first calculates the comprehensive statistical value as the weighted average of the five input parameters, with weights of 0.25, 0.3, 0.15, 0.1, and 0.2 respectively. When the comprehensive statistical value is less than 0.3, a linear adjustment function is used, and the adjustment coefficient is the comprehensive statistical value multiplied by 1.5. When the comprehensive statistical value is between 0.3 and 0.7, a quadratic adjustment function is used, and the adjustment coefficient is the square of the comprehensive statistical value multiplied by 2.0. When the comprehensive statistical value is greater than 0.7, an exponential adjustment function is used, and the adjustment coefficient is the exponential function value with a natural constant as the base and the comprehensive statistical value as the exponent divided by 10. Different adjustment functions are used to adapt to input data with different distribution characteristics.

[0044] The gating weight function is used to adjust the parameters of the neural network gating mechanism, and the input includes the information entropy of the current layer input data, the variance ratio of the current layer input data, the correlation coefficient of the current layer input data, the tidal change influence coefficient, and the monitoring frequency adjustment coefficient, and the output is the gating parameter adjustment coefficient; the gating weight function calculates the segmentation value based on the five input parameters, and the segmentation value calculation formula is the information entropy multiplied by 0.3 plus the variance ratio multiplied by 0.25 plus the correlation coefficient multiplied by 0.25 plus the tidal change influence coefficient multiplied by 0.1 plus the monitoring frequency adjustment coefficient multiplied by 0.1. When the segmentation value is less than 0.25, a conservative weight adjustment function is adopted, and the gating parameter is multiplied by an adjustment coefficient of 0.8. When the segmentation value is between 0.25 and 0.65, a balanced weight adjustment function is adopted, and the gating parameter is multiplied by an adjustment coefficient of 1.0. When the segmentation value is greater than 0.65, an aggressive weight adjustment function is adopted, and the gating parameter is multiplied by an adjustment coefficient of 1.2. Different weight adjustment functions are used to optimize the opening and closing state of the gating mechanism under different data complexities.

[0045] The monitoring frequency adjustment coefficient is an adjustment parameter calculated based on the current risk level and historical monitoring results, and is used to quantify the impact of changes in monitoring frequency on the risk assessment model. The calculation formula is the current risk level value multiplied by 0.6 plus the historical monitoring accuracy multiplied by 0.4. The monitoring frequency adjustment coefficient is used as an input parameter of the gating weight function to achieve iterative optimization of the monitoring frequency and model parameters.

[0046] The specific implementation of the above steps is described in detail below.

[0047] The specific implementation of step S01 involves establishing intertidal environmental monitoring equipment and constructing a tidal inundation impact matrix. First, a multi-type sensor array, including water level sensors, salinity sensors, temperature sensors, and light intensity sensors, is deployed in the target intertidal zone. The sensor density is no less than 10 monitoring points per square kilometer, and the data acquisition frequency is set to once every 5 minutes. Next, a data acquisition system is established to transmit the sensor data in real time to a data processing center via a wireless transmission network. A time series database is used to store continuous monitoring data over the tidal cycle. Data preprocessing is then performed to denoise the raw monitoring data and interpolate missing values. A sliding average filter algorithm is used to remove noise interference, and the filter window size is set to 7 data points. Subsequently, a tidal inundation impact matrix is ​​constructed based on the preprocessed data. This matrix is ​​dimensioned by tidal height levels as rows and environmental parameter types as columns. Matrix element values ​​are represented by standardized numerical values ​​representing the degree of impact, with a range of values ​​set from 0 to 1, where 0 represents no impact and 1 represents complete growth inhibition. The purpose of this step is to establish a comprehensive environmental monitoring system to provide accurate basic data support for subsequent risk assessments.

[0048] The specific embodiment of step S02 is to obtain Spartina alterniflora growth data and set up the tide-emerging growth matrix in the period of embracing after the tide recedes.First determine that the tide emerges in the period, based on tide forecast data and real-time water level monitoring result, identify the time window in which the Spartina alterniflora distribution area is completely exposed to the water, and the period of embracing judgment threshold value is set to below the regional water level and lower than average low tide level 0.2m. Next, carry out on-site growth data collection, adopt remote sensing image analysis technology to obtain vegetation coverage data, image resolution is not less than 0.5m, adopt soil moisture sensor to measure soil moisture content, and measuring depth range is 0 to 30cm, adopt root detector to measure root depth, adopt quadrat survey method statistics stalk density.Then the growth data collected are carried out quality control, exclude outlier and missing data, adopt quantile method to detect outlier, set outlier judgment threshold value and be the 25th percentile deducting 1.5 times of interquartile ranges or the 75th percentile plus 1.5 times of interquartile ranges. A tidal emergence growth matrix was then constructed, with the emergence time periods as the row dimension and growth indicator parameters as the column dimension. All indicators were converted to a numerical range of 0 to 1 using the minimum-maximum normalization method. The purpose of this step was to quantify the growth status of Spartina alterniflora under different emergence conditions and provide a data basis for identifying key growth factors.

[0049] The specific implementation of step S03 involves performing matrix multivariate decomposition on the tidal inundation impact matrix and the tidal emergence growth matrix and constructing a factor correlation graph. First, the singular value decomposition method is used to reduce the dimensionality of the two data matrices. By decomposing the original matrices into the product of a left singular vector matrix, a singular value diagonal matrix, and a right singular vector matrix, the key feature information is extracted, retaining the principal components with a cumulative contribution rate of more than 90%. Next, the correlation coefficient between the environmental factors and the growth factors is calculated. The Pearson correlation coefficient method is used to quantify the strength of the linear relationship between the factors. The calculation of the correlation coefficient involves the ratio of the covariance of each pair of factors to the product of their respective standard deviations. A factor correlation graph is then constructed, with the environmental factors and growth factors as the nodes of the graph and the absolute values ​​of the correlation coefficients as the edge weights, forming an undirected weighted graph structure. The minimum spanning tree algorithm is then used to solve the optimal factor connection path. This algorithm uses a greedy strategy, starting with the edge with the largest weight, and gradually constructing a tree structure connecting all nodes with the largest sum of weights, removing redundant factor connections. Finally, we extracted the main environmental and growth factors and calculated the weight coefficients for each factor to form the environmental and growth factor feature vectors, with the feature vector dimension set to 128. The purpose of this step was to identify the key factors that most significantly impacted the risk of Spartina alterniflora invasion, reduce the data dimensionality, and improve the computational efficiency of subsequent models.

[0050] The specific implementation of step S04 is to use a multi-layer neural network to construct a risk assessment network model. First, a neural network architecture is designed, using a multi-layer perceptron structure, including an input layer, multiple hidden layers, and an output layer. The input layer receives environmental factor feature vectors and growth factor feature vectors, and the dimension is set to 128 dimensions. The first hidden layer contains 256 neurons, the second hidden layer contains 128 neurons, the third hidden layer contains 64 neurons, and the output layer contains 1 neuron for outputting the risk probability value. Next, a gated weight adjustment mechanism is integrated. This mechanism dynamically adjusts network parameters based on the tidal change influence coefficient. The tidal change influence coefficient is calculated by multiplying the water level change amplitude by 0.4, the change frequency by 0.3, and the change rate by 0.3. Then, an adaptive weight adjustment function is designed, selecting different adjustment strategies based on the statistical characteristics of the current input features. When the gate threshold is less than 0.4, a linear weight adjustment function is used; when the gate threshold is between 0.4 and 0.7, a sigmoid weight adjustment function is used; when the gate threshold is greater than 0.7, a hyperbolic tangent weight adjustment function is used. The network then performs forward propagation calculations, where the input feature vectors undergo linear transformations and nonlinear activation functions in each hidden layer. Ultimately, the output layer generates a probability value for the risk of Spartina alterniflora invasion, ranging from 0 to 1. The goal of this step is to establish an intelligent risk assessment model that can dynamically adapt to environmental changes and accurately predict the risk of Spartina alterniflora invasion.

[0051] The specific implementation of step S05 involves categorizing invasion risk levels based on risk probability values. First, a risk level threshold system is established. Based on historical invasion case analysis and expert experience, a low-risk threshold of 0.3, a medium-risk threshold of 0.6, and a high-risk threshold of 0.8 are determined, forming a four-level risk assessment system. Next, a risk level assessment is performed. When the model outputs a risk probability value less than or equal to 0.3, the risk level is determined to be low, indicating a low likelihood of Spartina alterniflora invasion in the area. When the risk probability value is between 0.3 and 0.6, the risk level is determined to be medium, indicating a moderate invasion risk requiring attention. When the risk probability value is between 0.6 and 0.8, the risk level is determined to be high, indicating a significant invasion risk requiring preventive measures. When the risk probability value is greater than 0.8, the risk level is determined to be extremely high, indicating a significant invasion risk requiring immediate action. A risk level visualization system is then established, using different colors to identify areas of different risk levels: green indicates low risk, yellow indicates medium risk, orange indicates high risk, and red indicates extremely high risk. A risk assessment report is then generated, containing risk probability values, risk levels, major influencing factors, recommended measures, etc. The purpose of this step is to convert continuous probability values ​​into intuitive and easy-to-understand risk levels, providing a clear risk grading basis for management decisions.

[0052] The specific implementation of step S06 is to establish a dynamic risk warning mechanism and implement iterative optimization of model parameters. First, a differentiated monitoring strategy is formulated. According to the urgency of the invasion in different risk levels, corresponding monitoring frequencies are set. Low-risk areas implement routine monitoring once a month, medium-risk areas implement enhanced monitoring once a week, high-risk areas implement intensive monitoring once a day, and extremely high-risk areas implement continuous monitoring with real-time monitoring. Next, a monitoring data feedback mechanism is established. Newly acquired monitoring data is used to calculate the monitoring frequency adjustment coefficient. This coefficient is calculated based on the current risk level value multiplied by 0.6 and the historical monitoring accuracy multiplied by 0.4. Then, the gate weight function parameters are updated. The monitoring frequency adjustment coefficient is used as an input parameter. The new gate parameter adjustment coefficient is calculated based on the information entropy, variance ratio, correlation coefficient, and tidal change influence coefficient of the current layer input data. When the segment value is less than 0.25, a conservative adjustment strategy is adopted; when the segment value is between 0.25 and 0.65, a balanced adjustment strategy is adopted; when the segment value is greater than 0.65, an aggressive adjustment strategy is adopted. The risk assessment calculation was then re-run using the updated parameters to obtain an updated Spartina alterniflora invasion risk probability value and to reclassify the risk level. Finally, an iterative convergence determination mechanism was established. When the risk probability value changed by less than 0.02 over three consecutive iterations, the model was considered to have reached a stable state, iterations were terminated, and a final dynamic risk assessment report was generated. The purpose of this step was to establish an adaptive dynamic monitoring and early warning system to continuously optimize the risk assessment model and enhance early warning capabilities.

[0053] The detailed structure of the risk assessment network model utilizes a deep feedforward neural network architecture, with the overall framework based on the multilayer perceptron design concept. The input layer receives feature vectors of environmental factors and growth factors that have undergone feature engineering. The input dimension is fixed at 128, with each dimension corresponding to a standardized feature parameter. The first hidden layer contains 256 neurons and uses the rectified linear unit activation function. This layer performs preliminary nonlinear transformation and feature combination on the input features. The second hidden layer contains 128 neurons and also uses the rectified linear unit activation function. This layer further extracts high-order feature patterns. The third hidden layer contains 64 neurons and also uses the rectified linear unit activation function, responsible for final feature abstraction and pattern recognition. The output layer contains one neuron and uses the sigmoid activation function to ensure that the output risk probability value is between 0 and 1. The core innovation of this model is the gated weight adjustment mechanism integrated into the network. This mechanism dynamically adjusts the weight matrices of each layer based on parameters such as the tidal influence coefficient and input feature statistics, enabling the model to adaptively respond to varying environmental conditions. The implementation of the gating mechanism adopts the design idea of ​​the attention mechanism. The importance weight of each neuron is calculated through the gating function, and then the weight is multiplied by the output value of the neuron to enhance important features and suppress redundant information.

[0054] The detailed steps for establishing the training dataset begin with data collection. Thirty representative coastal intertidal zones were selected as monitoring sites, covering diverse geographic locations, climatic conditions, tidal patterns, and ecological environments. Each monitoring site was monitored continuously for 24 months to ensure that the data covered seasonal and interannual variations. Monitoring content included three categories: tidal cycle data, environmental parameter data, and vegetation growth data. Tidal cycle data covered indicators such as water level changes, tidal frequency, and tidal amplitude. Environmental parameter data included factors such as temperature, salinity, light intensity, and soil properties. Vegetation growth data included characteristics such as cover, density, biomass, and distribution range. Next, data quality control was performed, with a systematic quality check of the raw monitoring data. Data records with missing values ​​exceeding 20% ​​were removed, and outliers caused by sensor failure, human error, extreme weather, and other factors were identified and addressed. Data standardization was then implemented, using zero-mean unit variance normalization for all numerical variables and one-hot encoding for categorical variables. Feature engineering was then performed to extract derived features from the raw data, including time series features, spatial distribution features, and environmental correlation features. Sliding window techniques were used to extract short-term and long-term trend features, spatial interpolation methods were used to generate spatial distribution features, and correlation analysis was used to construct environmental correlation features. Finally, a complete training dataset consisting of 50,000 samples was constructed, with 70%, 20%, and 10% used for model training, validation, and testing, respectively. Each sample contained a 128-dimensional input feature vector and a corresponding risk level label. The labels used a four-category system to represent low, medium, high, and extremely high risk levels.

[0055] It should be noted that the key technical ideas of the present invention are mainly reflected in the following three aspects:

[0056] The construction of a dual data structure of the tidal inundation impact matrix and the tidal emergent growth matrix is ​​the first core technical idea. Traditional invasion risk assessment methods usually only focus on a single time node or static environmental parameters. The present invention realizes full-cycle dynamic monitoring of the growth environment of Spartina alterniflora by constructing a tidal inundation impact matrix and a tidal emergent growth matrix. The tidal inundation impact matrix captures the inhibitory or promoting effects of environmental parameters on plant growth at different tidal heights, while the tidal emergent growth matrix records the actual growth response characteristics of plants after the tide recedes. This dual matrix structure can fully describe the complex dynamic process of the intertidal ecosystem. Compared with the limitations of traditional methods that rely only on a single indicator or instantaneous data, it significantly improves the accuracy of characterizing the relationship between environmental factors and biological responses.

[0057] The dynamic neural network model based on the gated weight adjustment mechanism constitutes the second key technical idea. Traditional risk assessment models usually use fixed-weight machine learning algorithms, which cannot adapt to the high dynamics and complexity of the intertidal environment. The gated weight adjustment mechanism introduced in the present invention can dynamically adjust the network parameters according to the tidal change influence coefficient and the statistical characteristics of the input features, realizing the model's adaptive learning of characteristic patterns under different environmental conditions. This dynamic adjustment mechanism enables the network to automatically select the most suitable weight adjustment function when faced with environmental data of different complexities, thereby significantly improving the sensitivity to environmental changes and prediction accuracy while maintaining the stability of the model.

[0058] The third core technical approach is to use a differentiated monitoring frequency and iterative optimization mechanism based on risk levels. Traditional monitoring methods typically use a uniform monitoring frequency, which wastes resources and can miss key risk signals. The differentiated monitoring mechanism established by this invention can automatically adjust the monitoring frequency according to the risk level and feed back the newly acquired monitoring data into the gating weight function through the monitoring frequency adjustment coefficient, thereby achieving continuous optimization of the model parameters. This closed-loop feedback mechanism ensures that the risk assessment system can continuously improve itself as the environment changes and new data accumulates, forming a truly intelligent dynamic monitoring system.

[0059] These three core technical approaches form a complete intelligent risk assessment ecosystem. The dual-matrix data structure provides high-quality structured input for the dynamic neural network. The gated weight adjustment mechanism ensures the model's adaptive processing of complex environmental data. The differentiated monitoring and iterative optimization mechanism enables the system's self-learning and continuous improvement. This synergistic effect enables the entire system to maintain stable and reliable predictive performance despite the high complexity and uncertainty of intertidal ecosystems. Compared to traditional static assessment methods, it achieves a qualitative leap in predictive accuracy, resource utilization efficiency, and environmental adaptability.

[0060] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions. When the program instructions are run in a computer, the program instructions are used to execute the above-mentioned method for assessing the risk level of invasion by Spartina alterniflora.

[0061] The third aspect of the present invention provides a system for assessing the risk level of invasion of Spartina alterniflora, comprising the above-mentioned computer-readable storage medium, wherein the system is any one of a computer, a server, and a single-chip microcomputer, and the computer-readable storage medium is arranged in the system, and the system is provided with a microprocessor for executing program instructions stored in the computer-readable storage medium.

[0062] Specifically, the present invention addresses the technical issue of insufficiently accurate Spartina alterniflora invasion risk assessment under tidal conditions. Its fundamental principle lies in establishing a precise mathematical mapping between the tidal dynamics and the risk of plant invasion. The tidal inundation impact matrix quantifies the inhibitory effect of tidal inundation on Spartina alterniflora growth by recording changes in environmental parameters at different tidal heights, while the tidal emergence growth matrix captures the plant's growth recovery process after the tide recedes. The combination of these two matrices provides a complete description of the Spartina alterniflora growth state throughout the tidal cycle.

[0063] The application of matrix multivariate decomposition technology addresses the issues of redundant environmental factors and chaotic correlations in traditional methods. Singular value decomposition retains over 90% of the original data's information. The minimum spanning tree algorithm ensures accurate identification of key factor relationships, providing high-quality feature input for subsequent risk assessment. The gated weight adjustment mechanism is the core innovation of this invention. By dynamically adjusting the neural network weights based on the tidal influence coefficient, the model can adaptively respond to environmental changes under varying tidal conditions.

[0064] The key to the logicality of the technical solution of the present invention lies in its hierarchical and progressive processing architecture. From the bottom-level tidal environment data acquisition to the middle-level feature extraction and optimization, and then to the upper-level intelligent risk assessment, each level is designed according to the particularity of the tidal environment. In particular, the gated weight adjustment mechanism selects different weight adjustment functions according to the complexity of tidal changes, ensuring the stability and accuracy of the model under various tidal conditions. The iterative optimization mechanism achieves continuous improvement and accuracy enhancement of the risk assessment model through feedback adjustment of the monitoring frequency adjustment coefficient.

[0065] A specific embodiment 1 of the present invention is provided below. The specific implementation of each step in this embodiment 1 is described in detail as follows.

[0066] The specific implementation of step S01 is to establish intertidal zone environmental monitoring equipment and construct a tidal inundation impact matrix. First, a multi-type sensor array is deployed in the target intertidal zone area, with a sensor density of no less than 10 monitoring points per square kilometer and a data collection frequency set to once every 5 minutes. The construction of the tidal inundation impact matrix is ​​expressed as follows:

[0067] ;

[0068] Where, is the tidal inundation impact matrix; For the Tidal height level The degree of influence of environmental parameters on the growth of Spartina alterniflora; is the total number of tidal height levels, ranging from 8 to 12; The total number of environment parameter types is 4.

[0069] Matrix elements The standardized calculation formula is:

[0070] ;

[0071] Where, is the original monitoring value; For the The minimum observed value of an environmental parameter; For the The maximum observed value of an environmental parameter.

[0072] The parameter acquisition method is: Real-time monitoring is performed using sensors, including step 1: deploying water level sensors, salinity sensors, temperature sensors, and light intensity sensors; step 2: setting the data collection frequency to once every 5 minutes; and step 3: establishing a wireless transmission network to transmit data to the processing center in real time. and Obtained through statistical analysis of historical monitoring data, with a statistical period of no less than 24 months.

[0073] The specific implementation of step S02 is to obtain the growth data of Spartina alterniflora during the emergence period after the tide recedes and establish a tidal emergence growth matrix. The construction of the tidal emergence growth matrix is ​​expressed as follows:

[0074] ;

[0075] Where, For the tidal emergence growth matrix; For the The first Standardized values ​​of growth indicators; is the total number of time periods to be floated, ranging from 6 to 10; the number of matrix columns is fixed at 4, corresponding to vegetation coverage, soil moisture content, root depth, and stem density, respectively.

[0076] The growth index normalization formula is the same as the normalization method in step S01 and will not be described in detail here.

[0077] The specific implementation of step S03 is to perform matrix multivariate decomposition on the tidal inundation impact matrix and the tidal emergence growth matrix and construct a factor correlation graph. The singular value decomposition method is used to reduce the matrix dimension. The decomposition formula is expressed as follows:

[0078] ;

[0079] Where, is the original matrix to be decomposed; is the left singular vector matrix; is a diagonal matrix of singular values; is the transpose of the right singular vector matrix.

[0080] The calculation formula for the correlation coefficient between factors is:

[0081] ;

[0082] Where, Factor and factors Pearson correlation coefficient between ; and Factors and factors No. observations; and is the mean of the corresponding factor; is the total number of observed samples.

[0083] The parameter acquisition method is: and It is obtained by extracting the eigenvector after singular value decomposition; and Calculated using the arithmetic mean method; The value range is 1000 to 5000.

[0084] The specific implementation of step S04 is to use a multi-layer neural network to construct a risk assessment network model. The calculation formula for the tidal change influence coefficient is:

[0085] ;

[0086] Where, is the tidal change influence coefficient; is the amplitude of water level change; is the frequency of tidal changes; is the water level change rate; Calculate the error term for tidal effects, the default .

[0087] The calculation formula for the comprehensive statistical value of the weight adjustment function is:

[0088] ;

[0089] Where, Adjust the comprehensive statistics for weights; is the mean of the input feature vector; is the variance of the input eigenvector; is the input eigenvector skewness; is the kurtosis of the input eigenvector.

[0090] The segmented calculation formula of the weight adjustment coefficient is:

[0091] when hour: ;

[0092] when hour: ;

[0093] when hour: ;

[0094] Where, is the weight adjustment coefficient; is a natural constant.

[0095] The parameter acquisition method is: Use water level sensor for continuous monitoring, with measurement accuracy of no less than 1cm; Through tidal cycle analysis and calculation, the number of complete tidal cycles within 24 hours is counted; Calculation using the time derivative of water level change; The range is 0.02 to 0.05. 、 、 、 Calculated using statistical methods and updated in real time based on current batch input data.

[0096] The specific implementation of step S05 is the same as the aforementioned risk level threshold classification method, and will not be described in detail here.

[0097] The specific implementation of step S06 is to establish a dynamic risk warning mechanism and implement iterative optimization of model parameters. The calculation formula of the monitoring frequency adjustment coefficient is:

[0098] ;

[0099] Where, To monitor the frequency adjustment factor; is the current risk level value; For historical monitoring accuracy; Adjust the error term for the monitoring frequency.

[0100] The calculation formula for the segment value of the gate weight function is:

[0101] ;

[0102] Where, is the segment value of the gate weight; The information entropy of the input data of the current layer; The variance ratio of the input data for the current layer; Correlation coefficient of the input data for the current layer.

[0103] The information entropy calculation formula is:

[0104] ;

[0105] Where, For the The probability of the occurrence of a eigenvalue; is the total number of eigenvalue categories.

[0106] The segmented calculation formula of the gate parameter adjustment coefficient is:

[0107] when hour: ;

[0108] when hour: ;

[0109] when hour: ;

[0110] Where, is the adjusted gating parameter; are the original gating parameters.

[0111] The formula for determining the convergence of risk probability value is:

[0112] ;

[0113] Where, For the The risk probability value of the iteration; For the The risk probability value of the iteration; is the convergence threshold, set to 0.02.

[0114] The parameter acquisition method is: According to the risk level classification result of step S05, low risk, medium risk, high risk, and very high risk correspond to values ​​0.2, 0.5, 0.7, and 0.9, respectively; The calculation is made by comparing the historical forecast results with the actual situation, and the statistical period is 12 months; The range is 0.01 to 0.03. Calculation is done using information theory methods; The ratio of the current layer input data variance to the historical variance; The Pearson correlation coefficient method is used to calculate the internal correlation of the current input data. Obtained through the histogram statistical method, the continuous values ​​are discretized into 10 to 20 intervals for probability statistics. is the original gating parameter of the current layer of the neural network, which is determined by the network training process and has an initial value range of 0.1 to 0.9.

[0115] During the iterative optimization process, the model stops when convergence conditions are met for three consecutive iterations, resulting in a final, stable, dynamic risk assessment. Using this mathematical model and calculation formula, the entire assessment system achieves quantitative assessment and dynamic monitoring and early warning of the invasion risk of Spartina alterniflora.

[0116] It should be noted that the tidal flooding impact matrix A multidimensional matrix structure is used to systematically organize the variations in environmental parameters at different tidal heights. This matrix, based on the ecological theory of environmental gradients, uses rows and columns to characterize the interactions between tidal variations and environmental factors. Compared to existing single-parameter analysis methods, this matrix simultaneously captures the synergistic effects of multiple environmental factors under varying tidal conditions, enabling a comprehensive quantitative description of complex intertidal ecosystems and significantly improving the accuracy and systematic nature of environmental impact assessments.

[0117] Matrix element normalization formula Based on the minimum-maximum normalization principle, environmental parameters with different dimensions and numerical ranges are converted to unified dimensionless values. This normalization method is more robust than traditional mean normalization techniques and can effectively handle extreme values ​​commonly found in environmental monitoring data. It ensures that all environmental factors have equal weighting in the risk assessment process and avoids assessment bias caused by differences in numerical magnitude.

[0118] Tidal Emergence Growth Matrix This matrix represents the growth status of Spartina alterniflora at different emergence time periods. The design principle is based on the time-dependent nature of plant growth. Compared to traditional static growth indicator evaluation methods, this matrix structure dynamically captures the temporal changes in plant growth. Through comprehensive analysis of multiple time periods and indicators, it accurately predicts the spread potential of invasive species, providing a scientific basis for developing targeted prevention and control strategies.

[0119] Singular Value Decomposition Based on the theory of matrix decomposition in linear algebra, this method decomposes a high-dimensional, complex data matrix into the product of three lower-dimensional matrices. Compared to traditional principal component analysis techniques, this decomposition method offers improved numerical stability and computational efficiency. It significantly reduces the data dimension while preserving the primary information of the original data, effectively removing redundant information and noise interference. This provides high-quality feature input for subsequent machine learning models, improving the generalization capabilities of risk assessment models.

[0120] Pearson correlation coefficient Based on the principle of linear correlation measurement in statistics, this method quantifies the strength of the linear relationship between two variables. Compared to simple empirical judgments, this correlation analysis method has a rigorous mathematical foundation and can objectively identify the intrinsic connections between environmental factors and growth factors. By constructing a factor correlation network, it can accurately identify key influencing pathways in complex ecosystems, providing scientific guidance for optimizing monitoring plans and improving early warning efficiency.

[0121] Tidal change influence coefficient Based on the principle of multi-factor weighted fusion, the combined impacts of water level fluctuations, including amplitude, frequency, and rate, on the ecosystem are comprehensively considered. This weighted fusion approach is more comprehensive and accurate than single tidal indicator assessments. By setting different weight coefficients to reflect the relative importance of each factor, it can fully reflect the comprehensive impact of tidal changes on the growth environment of Spartina alterniflora, providing a reliable basis for dynamically adjusting risk assessment model parameters.

[0122] Weight adjustment comprehensive statistics Based on the principles of multivariate statistical analysis, a weighted fusion of input feature statistics, such as mean, variance, skewness, and kurtosis, is performed. Compared to traditional single-statistic analysis, this statistical fusion method has stronger data description capabilities and can comprehensively characterize the distribution characteristics and variation patterns of input data. By combining it with the tidal influence coefficient, it achieves adaptive recognition of different data patterns and significantly improves the adaptability of the neural network model to complex environmental conditions.

[0123] The piecewise calculation formula for the weight adjustment coefficient utilizes the principle of piecewise function design, selecting appropriate adjustment strategies based on the varying ranges of the comprehensive statistical value. This piecewise adjustment mechanism offers greater flexibility and adaptability than traditional fixed-weight methods, dynamically selecting the most appropriate adjustment function based on the complexity of the data features. By combining linear, quadratic, and exponential functions, it achieves precise adjustment for input data with varying distribution characteristics, effectively improving the model's prediction accuracy under diverse environmental conditions.

[0124] Monitoring frequency adjustment coefficient Based on the adaptive adjustment principle that combines historical feedback with current status, a weighted fusion of the current risk level and historical monitoring accuracy is performed. Compared to fixed monitoring frequency strategies, this adaptive adjustment method has higher resource utilization efficiency and can dynamically optimize monitoring plans based on risk changes and monitoring results. This achieves the rational allocation of monitoring resources and the continuous improvement of the early warning system, significantly improving the timeliness and accuracy of intrusion risk monitoring.

[0125] Gating weight segment value Based on a multidimensional evaluation principle that combines information theory and statistics, this method comprehensively considers multiple dimensions of data, including information entropy, variance ratio, and correlation. Compared to traditional single-metric gating mechanisms, this multidimensional evaluation method offers greater discriminative power and can comprehensively assess the complexity of input data from multiple perspectives. Through precise segmented threshold setting, it enables intelligent adjustment of the gating mechanism, effectively improving the stability and accuracy of neural networks when processing complex, time-varying data.

[0126] In order to better understand and implement the present invention, Example 2 of a specific application scenario of the present invention is provided below: A certain marine ecological research team received a task and needed to comprehensively assess the risk of Spartina alterniflora invasion in a certain typical intertidal zone along the coast. This regional area is approximately 15 square kilometers and has typical semi-diurnal tide characteristics. Historically, small-scale Spartina alterniflora invasion events have occurred, but there is a lack of systematic risk assessment and early warning mechanisms. The research team decided to adopt the technical solution of the present invention to establish a risk early warning system based on multi-factor dynamic assessment.

[0127] The research team first selected 6 representative monitoring sites in the target area, which covered different terrain features, water depth conditions and tidal influence levels. Each site deployed comprehensive monitoring equipment including water level sensors, salinity sensors, temperature sensors and light intensity sensors. The sensor deployment density reached 12 monitoring points per square kilometer, exceeding the design requirement of 10 points. The data acquisition system is set to automatically collect all environmental parameters every 5 minutes and transmit them to the data processing center in real time via a 4G wireless network. After 3 consecutive months of data collection, the research team obtained a basic monitoring data set covering the complete seasonal changes. The basic information of the monitoring area is shown in Table 1:

[0128] Table 1 Basic information of monitoring area

[0129]

[0130] Based on the collected original monitoring data, the research team constructed a tidal inundation impact matrix in accordance with the technical solution of the present invention. For water level data, tidal height levels are used to represent the tidal height, which is divided into 8 levels, from low tide level -1.5m to high tide level 2.5m, with each level interval of 0.5m. Environmental parameters include salinity, temperature, light intensity and dissolved oxygen concentration, of which dissolved oxygen concentration is optional data. By denoising and standardizing the original data, the complete tidal inundation impact matrix data was obtained. Some key data of the tidal inundation impact matrix are shown in Table 2:

[0131] Table 2 Tidal flooding impact matrix data

[0132]

[0133] During the ebb and flow period, the research team used a combination of remote sensing image analysis, field sampling, and instrumental measurement to obtain growth status data for the potential distribution areas of Spartina alterniflora. The ebb and flow period was divided into eight periods, each lasting three hours, covering the complete cycle from the complete ebb of the tide to the arrival of the next tide. Growth indicators include four parameters: vegetation coverage, soil moisture content, root depth, and stem density. Through continuous monitoring and data standardization, a tidal ebb and flow growth matrix was constructed. The specific data of the tidal ebb and flow growth matrix are shown in Table 3:

[0134] Table 3 Tidal emergent growth matrix data

[0135]

[0136] The research team used the singular value decomposition method to perform multivariate decomposition on the tidal inundation impact matrix and the tidal emergence growth matrix, and extracted the main environmental factors and growth factors. By retaining the principal components with a cumulative contribution rate of more than 90%, the original 32-dimensional feature vector was compressed into a 128-dimensional feature vector. The Pearson correlation coefficient was used to calculate the correlation between each factor, and a factor correlation graph was constructed. The minimum spanning tree algorithm was used to optimize the factor connection relationship and remove redundant factor connections. On this basis, the tidal change impact coefficient was calculated. The weight adjustment comprehensive statistical value is 0.67. It is 0.52.

[0137] Based on the extracted feature vectors, the research team constructed a multi-layer neural network risk assessment model. The model employs a 128-dimensional input layer, three hidden layers with 256, 128, and 64 neurons, respectively, and a 1-dimensional output layer. An integrated gated weight adjustment mechanism dynamically adjusts network parameters based on the tidal influence coefficient. The model was trained using the Adam optimizer with a learning rate of 0.001 and a batch size of 32. Convergence was achieved after 800 training rounds. The prediction accuracy on the test set reached 93.6%, with an average risk level prediction error of 0.031.

[0138] After a complete risk assessment, the research team determined the probability of Spartina alterniflora invasion and the corresponding risk levels for the six monitoring sites. Based on the low-risk threshold of 0.3, the medium-risk threshold of 0.6, and the high-risk threshold of 0.8, the risk assessment results for each site are shown in Table 4:

[0139] Table 4 Risk assessment results

[0140]

[0141] Based on the risk assessment results, the research team established a differentiated dynamic monitoring and early warning mechanism. For the extremely high-risk area of ​​ST06 station, the real-time monitoring mode was activated and the monitoring frequency adjustment coefficient was adjusted. Calculated as 0.76, the gate weight segment value The risk probability value was 0.68, adopting an aggressive weight adjustment strategy. After 15 consecutive days of dynamic monitoring and iterative optimization of model parameters, the change range of the risk probability value stabilized within 0.015, meeting the convergence judgment condition.

[0142] During the implementation process, the research team found that the risk levels of sites ST03 and ST05 changed significantly during the continuous monitoring period. The risk probability value of site ST03 increased from the initial 0.67 to 0.74, and that of site ST05 decreased from 0.72 to 0.68. This dynamic change reflects the sensitivity and accuracy of the technical solution of the present invention in capturing the impact of changes in environmental conditions on invasion risks. By timely adjusting monitoring strategies and early warning measures, the research team successfully predicted the small-scale germination of Spartina alterniflora near site ST06 and took timely removal measures at the early stages of germination.

[0143] To verify the advantages of the technical solution of the present invention, the research team compared the assessment results with traditional expert assessment methods and single-factor risk assessment methods. Traditional expert assessment methods mainly rely on ecological experts to classify risk levels based on field surveys and empirical judgments, lacking quantitative assessment standards and dynamic adjustment mechanisms. Single-factor risk assessment methods usually only consider one or two major environmental factors, such as salinity or tidal amplitude, and are unable to fully reflect the multi-factor interactions in complex ecosystems. The comparison results are shown in Table 5:

[0144] Table 5 Comparison results of technical solutions

[0145]

[0146] Traditional approaches to addressing the core technical challenge of Spartina alterniflora invasion risk assessment primarily rely on qualitative expert evaluation methods and simple single-factor quantitative analysis. While expert evaluation methods can leverage professional knowledge and experience to make comprehensive judgments, they suffer from subjectivity, poor consistency, and an inability to achieve continuous monitoring. While single-factor assessment methods offer a degree of objectivity, they often overlook the complex interactions among multiple factors, leading to biased and inaccurate assessment results.

[0147] The present invention has brought significant technological progress compared to traditional means. First, the prediction accuracy has been increased from 78.5% of traditional expert evaluation and 82.1% of single-factor evaluation to 93.6%. This improvement is mainly attributed to the application of multi-factor collaborative analysis and dynamic weight adjustment mechanism. Secondly, the warning timeliness has been shortened from 5.8 days and 7.2 days of traditional methods to 1.3 days, which has greatly improved the warning response speed and gained valuable time for timely implementation of prevention and control measures. Third, the false alarm rate has been reduced from 14.7% and 12.4% of traditional methods to 3.8%, significantly improving the reliability and practicality of risk assessment. Fourth, the present invention has achieved a transition from static assessment to dynamic continuous monitoring, which can track the impact of environmental changes on invasion risks in real time and adjust monitoring strategies and warning levels in a timely manner. Fifth, by establishing a standardized mathematical model and algorithm framework, the subjectivity and inconsistency problems in traditional expert evaluations are eliminated, laying a technical foundation for large-scale application and promotion. The combined effect of these technological advances makes the present invention have obvious technical advantages and application value in solving the problem of risk assessment of Spartina alterniflora invasion.

[0148] It should be noted that the detailed explanation of the variables involved in the present invention is shown in Table 6.

[0149] Table 6 Variable Explanation Table

[0150]

[0151] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention.

Claims

1. A method for assessing the risk level of Spartina alterniflora invasion, characterized in that: include: Establish intertidal zone environmental monitoring equipment to obtain water level change data, salinity data, temperature data, and light intensity data within the tidal cycle to form a tidal inundation spatiotemporal data set and construct a tidal inundation impact matrix. During the emergence period after the tide recedes, obtain vegetation coverage data, soil moisture data, root depth data, and stem density data in the distribution area of ​​Spartina alterniflora to establish a tidal emergence growth matrix. Perform matrix multivariate decomposition on the tidal inundation impact matrix and the tidal emergence growth matrix to construct a factor correlation graph. Use environmental factors and growth factors as nodes of the graph, and the correlation coefficients between factors as edge weights. Use the minimum spanning tree algorithm to solve the optimal factor. Connect paths, remove redundant factor connections, retain the correlation between key factors, extract major environmental factors and growth factors, calculate the weight coefficient and correlation coefficient of each factor, and form the environmental factor characteristic vector and the growth factor characteristic vector; use a multi-layer neural network to construct a risk assessment network model, input the environmental factor characteristic vector and the growth factor characteristic vector, dynamically adjust the network parameters through the gated weight adjustment mechanism, use the tidal change influence coefficient as the input parameter of the gated weight adjustment mechanism, and output the Spartina alterniflora invasion risk probability value; divide the invasion risk level according to the risk probability value; establish a risk warning mechanism, and implement differentiated monitoring frequencies for areas with different risk levels.

2. The method for assessing the risk level of Spartina alterniflora invasion according to claim 1, wherein: The tidal inundation impact matrix is ​​specifically a multidimensional data matrix constructed based on water level changes and environmental parameter changes within a tidal cycle. The rows of the matrix represent different tidal height levels, the columns represent different environmental parameter types, and the matrix element values ​​represent the degree of influence of the corresponding environmental parameters on the growth of Spartina alterniflora at the corresponding tidal height. The degree of influence is represented by a standardized numerical value ranging from 0 to 1, where 0 represents no influence and 1 represents complete inhibition of growth.

3. The method for assessing the risk level of Spartina alterniflora invasion according to claim 2, wherein: The tidal emergence growth matrix is ​​specifically a data matrix that records the growth status of Spartina alterniflora in different emergence time periods after the tide recedes. The rows of the matrix represent different emergence time periods, the columns represent different growth index parameters, and the matrix element values ​​represent the measured values ​​of the corresponding growth indicators in the corresponding emergence time periods. The growth indicators include vegetation coverage, soil moisture content, root depth, and stem density. All indicators are converted into a numerical range of 0 to 1 through standardization.

4. The method for assessing the risk level of Spartina alterniflora invasion according to claim 3, wherein: The factor correlation graph is specifically an undirected weighted graph constructed with environmental factors and growth factors as nodes and the correlation coefficients between factors as edge weights. Each node in the graph represents a factor variable, and the weight of the edge represents the strength of the correlation between the two factors. The minimum weight path connecting all factor nodes is solved by the minimum spanning tree algorithm to remove redundant factor connection relationships.

5. The method for assessing the risk level of Spartina alterniflora invasion according to claim 4, wherein: The tidal change impact coefficient is specifically a comprehensive impact index calculated based on the water level change amplitude, change frequency, and change rate within the tidal cycle. It is used to quantify the impact of tidal changes on the growth environment of Spartina alterniflora. The calculation formula is the water level change amplitude multiplied by 0.4 plus the change frequency multiplied by 0.3 plus the change rate multiplied by 0.3 plus the tidal impact calculation error term.

6. The method for assessing the risk level of Spartina alterniflora invasion according to claim 5, wherein: The specific structure of the risk assessment network model is a multilayer perceptron network comprising an input layer, a hidden layer and an output layer. The input layer receives the environmental factor feature vector and the growth factor feature vector with a dimension of 128 dimensions. The first hidden layer contains 256 neurons, the second hidden layer contains 128 neurons, the third hidden layer contains 64 neurons, and the output layer contains 1 neuron to output the risk probability value.

7. The method for assessing the risk level of Spartina alterniflora invasion according to claim 6, wherein: The step of dividing the intrusion risk level according to the risk probability value is specifically to set the low risk threshold to 0.3, the medium risk threshold to 0.6, and the high risk threshold to 0.8, that is, when the risk probability value ∈ [0, 0.3), it is determined to be a low risk level, when the risk probability value ∈ [0.3, 0.6), it is determined to be a medium risk level, when the risk probability value ∈ [0.6, 0.8), it is determined to be a high risk level, and when the risk probability value ∈ [0.8, 1], it is determined to be an extremely high risk level.

8. The method for assessing the risk level of Spartina alterniflora invasion according to claim 7, wherein: After the step of establishing a risk warning mechanism, it also includes recalculating the monitoring frequency adjustment coefficient based on the newly obtained monitoring data, inputting the monitoring frequency adjustment coefficient into the gating weight function to recalculate the gating parameter adjustment coefficient, using the updated gating parameter adjustment coefficient to adjust the parameters of the risk assessment network model, re-executing to obtain an updated Spartina alterniflora invasion risk probability value, and re-dividing the risk level based on the updated risk probability value.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores program instructions, and when the program instructions are executed in a computer, they are used to execute the method for assessing the risk level of Spartina alterniflora invasion according to any one of claims 1 to 8.

10. A system for assessing the risk level of Spartina alterniflora invasion, characterized in that: The computer-readable storage medium according to claim 9 is included, the system is any one of a computer, a server, and a single-chip microcomputer, the computer-readable storage medium is arranged in the system, and the system is provided with a microprocessor for executing program instructions stored in the computer-readable storage medium.

Images