A forest ecological environment monitoring system based on data analysis

Abstract

The application discloses a kind of forest ecological environment monitoring systems based on data analysis, comprising: data processing module, for collecting environmental parameters, and pre-processing;Heterogeneous modeling module, for building heterogeneous graph structure, and using Gaussian kernel function to calculate edge weight value;Convolution residual module, for performing multilayer graph convolution operation, and residual output is applied in next layer graph convolution;Pooling mapping module, for executing global pooling and multilayer perceptron mapping to node feature representation, generates state feature vector;Ecological clustering module, for performing clustering analysis using DENCLUE clustering algorithm, and calculating the ecological index value of each region;Monitoring analysis module, for comparing and analyzing the clustering result with historical environmental state data, and generating forest ecological risk monitoring result.The application realizes the structured modeling and dynamic risk intelligent identification of forest ecological environment, improves the precision of ecological monitoring and the timeliness of early warning.

Description

Technical Field

[0001] This invention relates to the field of environmental monitoring technology, and in particular to a forest ecological environment monitoring system based on data analysis. Background Technology

[0002] With global climate change, biodiversity loss, and overexploitation of forest resources, the health of forest ecosystems faces unprecedented challenges. Forests are not only an important component of terrestrial ecosystems but also a key ecological barrier for regulating carbon cycles, conserving water resources, preventing soil erosion, and maintaining regional climate stability. Accurate, continuous, and efficient monitoring of the forest ecological environment is of great significance for achieving sustainable management and protection of forest resources. Traditional forest ecological monitoring methods mainly rely on manual surveys and sampling at fixed observation points. While these methods provided basic ecological data accumulation in the early stages, they have significant limitations in terms of data timeliness, spatial coverage, environmental adaptability, and dynamic perception of ecological changes.

[0003] In recent years, with the development of IoT technology, some studies have begun to explore the use of environmental sensor nodes to collect real-time data on parameters such as air temperature and humidity, light intensity, soil conditions, and water quality in forests, transmitting this data wirelessly to a central processing platform. This approach has improved the automation and data acquisition efficiency of monitoring to some extent. However, most existing monitoring methods are still limited to single-parameter, isolated node data display and storage, lacking in-depth analysis of the correlations between multidimensional environmental variables. Furthermore, forest ecosystems themselves are characterized by multi-source heterogeneity, high-dimensional sparsity, and dynamic nonlinearity; simple parameter threshold comparisons and linear trend analyses are insufficient to accurately identify potential anomalies and systemic risks in the ecological state.

[0004] In existing research, some systems have incorporated machine learning models for data processing, such as using support vector machines or decision trees for data classification or anomaly detection. However, most of these models rely on rule-driven approaches or training sample selection, making them ill-suited for high-dimensional data environments involving large scales, multiple time periods, and numerous ecological factors. Furthermore, current technologies generally lack graph structure modeling capabilities, failing to effectively express the spatial interactions between ecological factors and limiting the depth of modeling the intrinsic evolutionary mechanisms of complex ecosystems. Especially in large-scale forest scenarios, the heterogeneity of ecological responses caused by differences in topography, climate, and vegetation types between different regions is difficult to model and identify using traditional methods.

[0005] Furthermore, current ecological monitoring methods for clustering and regional division often rely on manual rules, failing to incorporate high-dimensional ecological state characteristics for intelligent zoning and indicator extraction. This results in biased and unadaptable ecological assessments. In anomaly detection, existing methods mostly employ fixed threshold alarm mechanisms, neglecting the crucial role of historical evolution data in trend analysis and making it difficult to provide early warnings of systemic and gradual ecological degradation risks. Ecological indicator calculations primarily rely on weighted averages of raw data, lacking comprehensive strategies such as cluster center modeling and feature fluctuation modeling, resulting in insufficient accuracy and sensitivity of the final output ecological indicators.

[0006] Therefore, how to provide a forest ecological environment monitoring system based on data analysis that can realize graph structure modeling of multi-source ecological data, deep representation learning of node states, intelligent regional division and ecological indicator extraction based on density models, and further integrate historical data for dynamic comparison and abnormal trend identification is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] One objective of this invention is to propose a forest ecological environment monitoring system based on data analysis. This invention fully utilizes multi-source environmental sensing technology, weighted heterogeneous graph construction method, graph convolutional neural network, density clustering analysis, and time series comparison algorithm. It describes in detail the entire process from forest ecological multi-parameter collection, graph structure modeling, node state representation learning, ecological region clustering to ecological indicator calculation and risk trend identification. It has the advantages of high precision in ecological state characterization, strong spatial change perception capability, and timely and accurate ecological risk early warning.

[0008] A forest ecological environment monitoring system based on data analysis according to an embodiment of the present invention includes:

[0009] The data processing module is used to collect environmental parameters through sensors deployed in the forest area and to perform preprocessing.

[0010] The heterogeneous modeling module is used to construct a heterogeneous graph structure. It uses sensors as nodes, spatial correlations between different environmental parameters as edge weights, and uses a Gaussian kernel function to calculate the edge weights to generate a weighted heterogeneous graph representation.

[0011] The convolutional residual module is used to perform multi-layer graph convolution operations on the weighted heterogeneous graph representation to obtain node feature representations. After each layer of graph convolution operation, residual connections are performed on the node features, and the residual output is applied to the next layer of graph convolution.

[0012] The pooling mapping module is used to sequentially perform global pooling and multilayer perceptron mapping on the node feature representations output by the last layer graph convolution operation to generate the state feature vectors of each node.

[0013] The ecological clustering module is used to perform cluster analysis on the state feature vector using the DENCLUE clustering algorithm, identify ecological regions with similar state characteristics, and calculate the ecological index values ​​of each region based on the cluster centers.

[0014] The monitoring and analysis module is used to compare and analyze clustering results with historical environmental status data, identify abnormal trends in ecological indicators, and generate forest ecological risk monitoring results.

[0015] Optionally, modules can be integrated using the following methods:

[0016] S1. Collect environmental parameters through sensors deployed in the forest area and perform preprocessing;

[0017] S2. Using sensors as nodes and the spatial correlation between different environmental parameters as edge weights, construct a heterogeneous graph structure, and use the Gaussian kernel function to calculate the edge weights to generate a weighted heterogeneous graph representation.

[0018] S3. Perform multi-layer graph convolution operations on the weighted heterogeneous graph representation to obtain node feature representations. After each layer of graph convolution operations, perform residual connections on the node features and apply the residual output to the next layer of graph convolution.

[0019] S4. For the node feature representations output by the convolution operation of the last layer graph, perform global pooling and multilayer perceptron mapping in sequence to generate the state feature vectors of each node.

[0020] S5. The DENCLUE clustering algorithm is used to perform cluster analysis on the state feature vectors to identify ecological regions with similar state characteristics, and the ecological index values ​​of each region are calculated based on the cluster centers.

[0021] S6. Compare and analyze the clustering results with historical environmental status data to identify abnormal trends in ecological indicators and generate forest ecological risk monitoring results.

[0022] Optionally, the environmental parameters include air temperature, air humidity, light intensity, soil moisture, soil pH, carbon dioxide concentration, PM2.5 concentration, wind speed, and wind direction.

[0023] Optionally, the preprocessing includes missing value imputation, outlier removal, timestamp alignment, format conversion, and data standardization.

[0024] Optionally, S2 specifically includes:

[0025] S21. Construct a heterogeneous graph structure, treating the sensors deployed in the forest area as nodes in the graph structure, denoted as the node set. ,in This indicates the total number of sensor nodes deployed in the forest area;

[0026] S22. Based on the preprocessed environmental parameters, calculate the nodes. and Spatial correlation measure among parameter values ​​collected within a given time window Construct an edge set :

[0027] ;

[0028] in, Representing nodes respectively At any moment Environmental parameter observations Represents a node In the time window The average observed value within, Indicates the length of the time window;

[0029] S23. Calculate nodes using a Gaussian kernel function. and Edge weights between The formula is as follows:

[0030] ;

[0031] in, The bandwidth parameter of the Gaussian kernel function is used to adjust the sensitivity of the edge weight distribution. The weighting coefficient representing the distance factor. This represents the natural exponential function with base e. Represents a node and Manhattan distance between them;

[0032] S24. Set the nodes With edge set and all edge weights The adjacency matrix formed Combining to construct a weighted heterogeneous graph structure It outputs a weighted heterogeneous graph representation.

[0033] Optionally, S3 specifically includes:

[0034] S31. Represent the input graph as a weighted heterogeneous graph convolutional neural network, and let the initial node feature matrix be... ,in Represents a node The initial feature vector, This indicates the total number of sensor nodes deployed in the forest area;

[0035] S32. Perform graph convolution operation on each layer, and calculate the node feature update results based on the normalized edge weights:

[0036] ;

[0037] in, Indicates the first The node feature matrix after layer graph convolution. Indicates the first The node feature matrix after layer graph convolution. It is an adjacency matrix. This is a node degree matrix, with elements on the main diagonal. , For the first The trainable weight matrix of the layer, It is a fully connected mapping matrix. The cross-channel attention weight matrix is ​​generated by global average pooling and normalized by the Sigmoid function. This represents the Hadamard element-wise product. For the residual connection mapping matrix, Represents a non-linear activation function;

[0038] S33. After convolving each layer of the graph, the node feature matrix of the previous layer is... With the current layer feature matrix Perform residual fusion and use the fusion result as input for the next layer. The formula for calculating the fusion result is as follows:

[0039] ;

[0040] in, The features of the merged nodes, The fusion ratio coefficient, with a value between 0 and 1, represents a trainable parameter;

[0041] S34. Repeat steps S32 to S33 until the preset number of convolutional layers is reached. Output node feature matrix .

[0042] Optionally, S4 specifically includes:

[0043] S41, where the node feature matrix Each row vector in the dataset undergoes global pooling along the channel dimension to obtain a compressed feature representation:

[0044] ;

[0045] in, For nodes The compressed feature vector, Represents a node The graph convolutional feature vector, express Dimensions It is a weighted nonlinear transformation function used to enhance the channel response. For vectors The One portion, It is a trainable linear mapping matrix;

[0046] S42. Input the compressed feature representation into a... A multilayer perceptron network consisting of layers, where each layer performs linear transformation, normalization, nonlinear activation, and residual mapping, outputs as follows:

[0047] ;

[0048] in, For nodes After the first The feature vector output by the layer, For nodes After the first The feature vector output by the layer satisfies ,and , For the first The linear weight matrix of the layer, For the channel normalization matrix, For bias vectors, Represents a non-linear activation function. For the first The residual connectivity coefficient matrix of the layer;

[0049] S43, the multilayer perceptron network completes... After layer transformation mapping, nodes are obtained. State feature vector The state feature vector is used to perform cluster analysis and ecological indicator assessment.

[0050] Optionally, S5 specifically includes:

[0051] S51. Construct a state feature space, and collect the state feature vectors of all nodes in the state feature space. Calculate the density estimate of each node using a kernel density function:

[0052] ;

[0053] in, Represents a node The density estimate, The dimension of the state feature vector. The normalization factor of the kernel function is defined as follows: , Pi is a constant. The kernel density function bandwidth parameter, Represents the state feature vector In the Components in each dimension Represents the state feature vector In the Components in each dimension This indicates the total number of sensor nodes deployed in the forest area;

[0054] S52. Perform the DENCLUE clustering algorithm in the state feature space. Based on the density estimate of each node, use the density gradient tracking method to iteratively move the position of each node according to the local density increment direction until convergence to the node position with the maximum local density. Group the state feature vectors with the same convergence result into the same cluster to obtain the cluster set. ;

[0055] S53. Let the cluster set be... ,in This represents the total number of clusters, with each cluster corresponding to an ecological region. Cluster centers are calculated for the state feature vectors of all nodes within each cluster, and the ecological index vector of the region is obtained.

[0056] ;

[0057] in, For the first The state feature centers of each cluster The ecological indicator vectors for the ecological region are normalized to obtain the corresponding indicator values. The feature transformation matrix, For ecological compensation matrix, Represents clusters The number of nodes included.

[0058] Optionally, the DENCLUE clustering algorithm starts with the density estimate of each node, adopts a density gradient tracking approach to introduce a weighted iterative update mechanism, introduces a node density weighting adjustment factor and a dynamic constraint term of feature variance in each position iteration, performs position update operations through the local density increment direction, and constructs a cluster set after all nodes have converged. .

[0059] Optionally, S6 specifically includes:

[0060] S61. Based on the ecological indicator vector of each region, construct the regional ecological indicator sequence for the current period, and retrieve the ecological indicator sequence of the same region that has been stored in the corresponding historical period as a reference benchmark.

[0061] S62. Perform time series change alignment analysis on the current ecological indicator sequence and the historical ecological indicator sequence. By calculating the difference between the indicator values ​​at each time point, extract the trend indicators that have shifted in each ecological dimension and generate a regional indicator change trajectory matrix.

[0062] S63. Based on the offset, offset speed and offset direction of each indicator dimension in the change trajectory matrix, combined with the preset threshold rules, ecological areas with abnormal indicator values ​​are identified, and the degree of abnormality is classified into levels, forming a forest ecological risk monitoring result that includes regional index, cluster number, abnormality level, ecological indicator value and change parameter set.

[0063] The beneficial effects of this invention are:

[0064] First, this invention overcomes the shortcomings of traditional monitoring methods that cannot express the spatial interaction between ecological factors by constructing a weighted heterogeneous graph structure and introducing a Gaussian kernel function to model the spatial correlation between sensor nodes. It realizes the structured expression of forest multi-source ecological data and the modeling of regional ecological connections, effectively improving the data organization efficiency and representation ability.

[0065] Secondly, this invention introduces a multi-layer graph convolutional neural network to deeply extract node features and maintains cross-layer information flow through a residual connection mechanism. It combines global pooling and multi-layer perceptron mapping to generate high-dimensional ecological state feature vectors. While preserving the original ecological information of nodes, it enhances the ability to capture complex ecological change patterns, provides high-quality expressions for subsequent clustering and index evaluation, and significantly improves the sensitivity and recognition accuracy of the monitoring system to micro-anomalies in the forest ecological environment.

[0066] Finally, this invention employs the DENCLUE clustering algorithm based on the density function to perform regional division of the state feature vector, calculates ecological index values ​​through dynamic cluster centers, and performs time series comparison with historical state data to accurately identify abnormal trends in ecological index values. This enables intelligent monitoring and hierarchical early warning of forest ecological risks, and solves the technical bottlenecks of existing methods in dynamic evolution detection, spatial heterogeneity identification, and ecological security early warning. Attached Figure Description

[0067] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0068] Figure 1This is a module structure diagram of a forest ecological environment monitoring system based on data analysis proposed in this invention;

[0069] Figure 2 This is a flowchart of a forest ecological environment monitoring system based on data analysis proposed in this invention.

[0070] Figure 3 This is a schematic diagram of the clustering process of an algorithm for a forest ecological environment monitoring system based on data analysis proposed in this invention. Detailed Implementation

[0071] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0072] refer to Figure 1 A data analysis-based forest ecological environment monitoring system includes:

[0073] The data processing module is used to collect environmental parameters through sensors deployed in the forest area and to perform preprocessing.

[0074] The heterogeneous modeling module is used to construct a heterogeneous graph structure. It uses sensors as nodes, spatial correlations between different environmental parameters as edge weights, and uses a Gaussian kernel function to calculate the edge weights to generate a weighted heterogeneous graph representation.

[0075] The convolutional residual module is used to perform multi-layer graph convolution operations on the weighted heterogeneous graph representation to obtain node feature representations. After each layer of graph convolution operation, residual connections are performed on the node features, and the residual output is applied to the next layer of graph convolution.

[0076] The pooling mapping module is used to sequentially perform global pooling and multilayer perceptron mapping on the node feature representations output by the last layer graph convolution operation to generate the state feature vectors of each node.

[0077] The ecological clustering module is used to perform cluster analysis on the state feature vector using the DENCLUE clustering algorithm, identify ecological regions with similar state characteristics, and calculate the ecological index values ​​of each region based on the cluster centers.

[0078] The monitoring and analysis module is used to compare and analyze clustering results with historical environmental status data, identify abnormal trends in ecological indicators, and generate forest ecological risk monitoring results.

[0079] This invention integrates data processing, heterogeneous modeling, convolutional residuals, pooling mapping, ecological clustering, and monitoring analysis into a unified structure, enabling full-process monitoring of the forest ecological environment from data collection to risk output. It has the advantages of clear structure, closed-loop function, and strong scalability.

[0080] refer to Figure 2-3 In this embodiment, the modules are interconnected using the following method:

[0081] S1. Collect environmental parameters through sensors deployed in the forest area and perform preprocessing;

[0082] S2. Using sensors as nodes and the spatial correlation between different environmental parameters as edge weights, construct a heterogeneous graph structure, and use the Gaussian kernel function to calculate the edge weights to generate a weighted heterogeneous graph representation.

[0083] S3. Perform multi-layer graph convolution operations on the weighted heterogeneous graph representation to obtain node feature representations. After each layer of graph convolution operations, perform residual connections on the node features and apply the residual output to the next layer of graph convolution.

[0084] S4. For the node feature representations output by the convolution operation of the last layer graph, perform global pooling and multilayer perceptron mapping in sequence to generate the state feature vectors of each node.

[0085] S5. The DENCLUE clustering algorithm is used to perform cluster analysis on the state feature vectors to identify ecological regions with similar state characteristics, and the ecological index values ​​of each region are calculated based on the cluster centers.

[0086] S6. Compare and analyze the clustering results with historical environmental status data to identify abnormal trends in ecological indicators and generate forest ecological risk monitoring results.

[0087] This invention clarifies the logical flow between each functional module, realizing a step-by-step implementation path from raw data preprocessing, graph construction, feature calculation to final risk assessment, thereby enhancing the stability and controllability of the system and effectively improving the consistency and intelligence of data processing at each stage.

[0088] In this embodiment, the environmental parameters include air temperature, air humidity, light intensity, soil moisture, soil pH, carbon dioxide concentration, PM2.5 concentration, wind speed, and wind direction.

[0089] This invention achieves comprehensive coverage of multi-dimensional elements of forest ecology by collecting environmental data of multiple parameters such as air temperature, humidity, light intensity, soil moisture, pH value, CO2 concentration, PM2.5 concentration, wind speed and wind direction, thereby enhancing the system's ability to perceive subtle changes in ecological conditions.

[0090] In this embodiment, the preprocessing includes missing value imputation, outlier removal, timestamp alignment, format conversion, and data standardization.

[0091] This invention improves the quality of raw environmental data through operations such as missing value imputation, outlier removal, timestamp alignment, format conversion, and data standardization, thereby enhancing the accuracy and stability of subsequent modeling processes and providing a reliable input foundation for high-quality graph construction and feature calculation.

[0092] In this embodiment, S2 specifically includes:

[0093] S21. Construct a heterogeneous graph structure, treating the sensors deployed in the forest area as nodes in the graph structure, denoted as the node set. ,in This indicates the total number of sensor nodes deployed in the forest area;

[0094] S22. Based on the preprocessed environmental parameters, calculate the nodes. and Spatial correlation measure among parameter values ​​collected within a given time window Construct an edge set :

[0095] ;

[0096] in, Representing nodes respectively At any moment Environmental parameter observations Represents a node In the time window The average observed value within, Indicates the length of the time window;

[0097] S23. Calculate nodes using a Gaussian kernel function. and Edge weights between The formula is as follows:

[0098] ;

[0099] in, The bandwidth parameter of the Gaussian kernel function is used to adjust the sensitivity of the edge weight distribution. The weighting coefficient representing the distance factor. This represents the natural exponential function with base e. Represents a node and Manhattan distance between them;

[0100] S24. Set the nodes With edge set and all edge weights The adjacency matrix formed Combining to construct a weighted heterogeneous graph structure It outputs a weighted heterogeneous graph representation.

[0101] This invention constructs a weighted heterogeneous graph with sensors as nodes and parameter correlation as edge weights, and uses a Gaussian kernel function to calculate the edge weights, thereby realizing quantitative modeling of ecological data in spatial structure and improving the ability to express the spatial coupling relationship between ecological factors.

[0102] In this embodiment, S3 specifically includes:

[0103] S31. Represent the input graph as a weighted heterogeneous graph convolutional neural network, and let the initial node feature matrix be... ,in Represents a node The initial feature vector, This indicates the total number of sensor nodes deployed in the forest area;

[0104] S32. Perform graph convolution operation on each layer, and calculate the node feature update results based on the normalized edge weights:

[0105] ;

[0106] in, Indicates the first The node feature matrix after layer graph convolution. Indicates the first The node feature matrix after layer graph convolution. It is an adjacency matrix. This is a node degree matrix, with elements on the main diagonal. , For the first The trainable weight matrix of the layer, It is a fully connected mapping matrix. The cross-channel attention weight matrix is ​​generated by global average pooling and normalized by the Sigmoid function. This represents the Hadamard element-wise product. For the residual connection mapping matrix, Represents a non-linear activation function;

[0107] S33. After convolving each layer of the graph, the node feature matrix of the previous layer is... With the current layer feature matrix Perform residual fusion and use the fusion result as input for the next layer. The formula for calculating the fusion result is as follows:

[0108] ;

[0109] in, The features of the merged nodes, The fusion ratio coefficient, with a value between 0 and 1, represents a trainable parameter;

[0110] S34. Repeat steps S32 to S33 until the preset number of convolutional layers is reached. Output node feature matrix .

[0111] This invention employs multi-layer graph convolution operations and introduces a residual connection mechanism to deeply extract high-dimensional features of nodes in heterogeneous graph structures and retain cross-layer information, thereby enhancing the richness and stability of node representation and providing accurate representation for subsequent state recognition.

[0112] In this embodiment, S4 specifically includes:

[0113] S41, where the node feature matrix Each row vector in the dataset undergoes global pooling along the channel dimension to obtain a compressed feature representation:

[0114] ;

[0115] in, For nodes The compressed feature vector, Represents a node The graph convolutional feature vector, express Dimensions It is a weighted nonlinear transformation function used to enhance the channel response. For vectors The One portion, It is a trainable linear mapping matrix;

[0116] S42. Input the compressed feature representation into a... A multilayer perceptron network consisting of layers, where each layer performs linear transformation, normalization, nonlinear activation, and residual mapping, outputs as follows:

[0117] ;

[0118] in, For nodes After the first The feature vector output by the layer, For nodes After the first The feature vector output by the layer satisfies ,and , For the first The linear weight matrix of the layer, For the channel normalization matrix, For bias vectors, Represents a non-linear activation function. For the first The residual connectivity coefficient matrix of the layer;

[0119] S43, the multilayer perceptron network completes... After layer transformation mapping, nodes are obtained. State feature vector The state feature vector is used to perform cluster analysis and ecological indicator assessment.

[0120] This invention constructs state feature vectors through global pooling compression and multilayer perceptron mapping, realizing nonlinear feature abstraction of node ecological state, improving the separability and expression dimension of node information, and helping to identify similar ecological feature regions in the subsequent clustering process.

[0121] In this embodiment, S5 specifically includes:

[0122] S51. Construct a state feature space, and collect the state feature vectors of all nodes in the state feature space. Calculate the density estimate of each node using a kernel density function:

[0123] ;

[0124] in, Represents a node The density estimate, The dimension of the state feature vector. The normalization factor of the kernel function is defined as follows: , Pi is a constant. The kernel density function bandwidth parameter, Represents the state feature vector In the Components in each dimension Represents the state feature vector In the Components in each dimension This indicates the total number of sensor nodes deployed in the forest area;

[0125] S52. Perform the DENCLUE clustering algorithm in the state feature space. Based on the density estimate of each node, use the density gradient tracking method to iteratively move the position of each node according to the local density increment direction until convergence to the node position with the maximum local density. Group the state feature vectors with the same convergence result into the same cluster to obtain the cluster set. ;

[0126] S53. Let the cluster set be... ,in This represents the total number of clusters, with each cluster corresponding to an ecological region. Cluster centers are calculated for the state feature vectors of all nodes within each cluster, and the ecological index vector of the region is obtained.

[0127] ;

[0128] in, For the first The state feature centers of each cluster The ecological indicator vectors for the ecological region are normalized to obtain the corresponding indicator values. The feature transformation matrix, For ecological compensation matrix, Represents clusters The number of nodes included.

[0129] This invention employs the DENCLUE clustering algorithm to perform density clustering on state feature vectors, thereby achieving intelligent division of forest ecological regions. It also combines cluster centers to calculate ecological indicators, improving the rationality of region division and the representativeness of indicator evaluation.

[0130] In this embodiment, the DENCLUE clustering algorithm starts with the density estimate of each node and introduces a weighted iterative update mechanism using density gradient tracking. In each position iteration, a node density weighting adjustment factor and a dynamic constraint term for feature variance are introduced. Position update operations are performed through the local density increment direction, and a cluster set is constructed after all nodes have converged. .

[0131] This invention optimizes the node iteration path and convergence process by introducing a density-weighted adjustment factor and a dynamic constraint term for feature variance in DENCLUE clustering, thereby improving the robustness of the clustering results and its adaptability to complex distributions of ecological data.

[0132] In this embodiment, S6 specifically includes:

[0133] S61. Based on the ecological indicator vector of each region, construct the regional ecological indicator sequence for the current period, and retrieve the ecological indicator sequence of the same region that has been stored in the corresponding historical period as a reference benchmark.

[0134] S62. Perform time series change alignment analysis on the current ecological indicator sequence and the historical ecological indicator sequence. By calculating the difference between the indicator values ​​at each time point, extract the trend indicators that have shifted in each ecological dimension and generate a regional indicator change trajectory matrix.

[0135] S63. Based on the offset, offset speed and offset direction of each indicator dimension in the change trajectory matrix, combined with the preset threshold rules, ecological areas with abnormal indicator values ​​are identified, and the degree of abnormality is classified into levels, forming a forest ecological risk monitoring result that includes regional index, cluster number, abnormality level, ecological indicator value and change parameter set.

[0136] This invention dynamically compares the current ecological indicator sequence with historical data, and combines trajectory offset analysis and threshold identification rules to achieve trend identification and level warning of abnormal forest ecological areas. It has the advantages of strong monitoring timeliness and high risk judgment accuracy.

[0137] Example 1:

[0138] To verify the feasibility of this invention in practice, it was applied to a typical forest ecological protection area. This area has complex terrain, diverse vegetation types, and a rapidly changing ecological environment, making it difficult to achieve large-scale, continuous, and structured data collection and risk identification using traditional manual observation methods. This invention deploys multiple environmental sensors to collect real-time data on nine types of environmental parameters, including air temperature, air humidity, light intensity, soil moisture, soil pH, carbon dioxide concentration, PM2.5 concentration, wind speed, and wind direction. By combining weighted heterogeneous graph modeling, graph convolutional network analysis, denque clustering, and indicator trend assessment, it achieves comprehensive monitoring and intelligent early warning of the ecological status of this forest area.

[0139] In actual deployment, a total of 210 environmental sensor nodes were set up along the main roads, gentle slopes, woodlands, water bodies, and forest edges. The sampling period was set to once every 10 minutes, with a monitoring duration of 45 days. The system preprocessed the data uploaded from these nodes, performing operations such as missing value imputation, outlier removal, and timestamp alignment, ultimately forming a uniformly formatted time-series environmental data structure. Subsequently, using each sensor as a node in the graph, a weighted heterogeneous graph structure was constructed based on the spatial correlation of different environmental factors. The average graph density reached 0.37, and the average shortest path between nodes was 3.12, indicating a reasonable and compact network topology.

[0140] A graph convolutional neural network was used to perform three convolutional operations on the graph structure, with feature dimensions of 64, 128, and 256 for each layer. A channel attention residual mechanism was applied after each layer to generate a 256-dimensional node state feature vector. All node state feature vectors were then clustered using the DENCLUE method, dividing the graph into 14 ecologically homogeneous regions. The variance of the node density estimate within each cluster was less than 0.014, indicating high state similarity within each region and a stable and reliable clustering structure.

[0141] Further time-series comparative analysis was conducted on the daily ecological indicator values ​​calculated for each ecological region over 45 days and compared with historical seasonal data for the same period. The results showed that from day 28 to day 34, the ecological indicator values ​​of cluster 6 exhibited a continuous downward shift, with an average decrease of 5.7% in air humidity, an abnormal decrease of 0.9 in soil pH, and an increase of 27.2% in PM2.5 concentration. System detection and assessment identified this as a medium-level ecological disturbance event, triggering the regional ecological risk early warning mechanism. This early warning information was ultimately exported as a risk report, including an index of the risk-occurring area, the dimensions of the abnormal indicators, the magnitude of the shift, and the rate of trend, providing a precise basis for subsequent management measures.

[0142] The table below shows the statistical results and corresponding trends of ecological indicators in different regions. The data comes from the system output:

[0143] Table 1 Summary of Monitoring Data on Multidimensional Environmental Status and Indicator Changes in Forest Ecological Regions

[0144]

[0145] As can be seen from the above implementation, the present invention has significant advantages in processing large-scale heterogeneous ecological data, identifying ecological regions, and analyzing abnormal trends of indicators. It not only improves the accuracy and timeliness of forest ecological monitoring, but also strengthens the dynamic perception of potential ecological risks, verifying the practical value and operability of the present invention in complex natural environments.

[0146] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A forest ecological environment monitoring system based on data analysis, characterized in that, include: The data processing module is used to collect environmental parameters through sensors deployed in the forest area and to perform preprocessing. The heterogeneous modeling module is used to construct heterogeneous graph structures. Using sensors as nodes and the spatial correlation between different environmental parameters as edge weights, it calculates the edge weights using a Gaussian kernel function to generate a weighted heterogeneous graph representation, including: Construct a heterogeneous graph structure, with sensors deployed in the forest area as nodes in the graph structure, denoted as the node set. ,in This indicates the total number of sensor nodes deployed in the forest area; Based on the preprocessed environmental parameters, the computation node and Spatial correlation measure among parameter values ​​collected within a given time window Construct an edge set : ； in, Representing nodes respectively At any moment Environmental parameter observations Represents a node In the time window The average observed value within, Indicates the length of the time window; Nodes are computed using the Gaussian kernel function. and Edge weights between The formula is as follows: ； in, The bandwidth parameter of the Gaussian kernel function is used to adjust the sensitivity of the edge weight distribution. The weighting coefficient representing the distance factor. This represents the natural exponential function with base e. Represents a node and Manhattan distance between them; Node set With edge set and all edge weights The adjacency matrix formed Combining to construct a weighted heterogeneous graph structure It outputs a weighted heterogeneous graph representation; The convolutional residual module performs multi-layer graph convolution operations on the weighted heterogeneous graph representation to obtain node feature representations. After each graph convolution operation, it performs residual connections on the node features and applies the residual output to the next graph convolution, including: We represent the input graph as a weighted heterogeneous graph convolutional neural network, and let the initial node feature matrix be... ,in Represents a node The initial feature vector, This indicates the total number of sensor nodes deployed in the forest area; For each layer, perform graph convolution operations and calculate node feature updates based on normalized edge weights: ； in, Indicates the first The node feature matrix after layer graph convolution. Indicates the first The node feature matrix after layer graph convolution. It is an adjacency matrix. This is a node degree matrix, with elements on the main diagonal. , For the first The trainable weight matrix of the layer, It is a fully connected mapping matrix. The cross-channel attention weight matrix is ​​generated by global average pooling and normalized by the Sigmoid function. This represents the Hadamard element-wise product. For the residual connection mapping matrix, Represents a non-linear activation function; After convolution of each graph layer, the node feature matrix of the previous layer is... With the current layer feature matrix Perform residual fusion and use the fusion result as input for the next layer. The formula for calculating the fusion result is as follows: ； in, The features of the merged nodes, The fusion ratio coefficient, with a value between 0 and 1, represents a trainable parameter; Repeat the above process until the preset number of convolutional layers is reached. Output node feature matrix ; The pooling mapping module is used to sequentially perform global pooling and multilayer perceptron mapping on the node feature representations output from the last layer graph convolution operation, generating state feature vectors for each node, including: Among them, the node feature matrix Each row vector in the dataset undergoes global pooling along the channel dimension to obtain a compressed feature representation: ； in, For nodes The compressed feature vector, Represents a node The graph convolutional feature vector, express Dimensions It is a weighted nonlinear transformation function used to enhance the channel response. For vectors The One portion, It is a trainable linear mapping matrix; The compressed feature representation is input into a... A multilayer perceptron network consisting of layers, where each layer performs linear transformation, normalization, nonlinear activation, and residual mapping, outputs as follows: ； in, For nodes After the first The feature vector output by the layer, For nodes After the first The feature vector output by the layer satisfies ,and , For the first The linear weight matrix of the layer, For the channel normalization matrix, For bias vectors, Represents a non-linear activation function. For the first The residual connectivity coefficient matrix of the layer; The multilayer perceptron network completes... After layer transformation mapping, nodes are obtained. State feature vector The state feature vector is used to perform cluster analysis and ecological indicator assessment; The ecological clustering module is used to perform cluster analysis on the state feature vector using the DENCLUE clustering algorithm, identify ecological regions with similar state characteristics, and calculate the ecological index values ​​of each region based on the cluster centers. The monitoring and analysis module is used to compare and analyze clustering results with historical environmental status data, identify abnormal trends in ecological indicators, and generate forest ecological risk monitoring results.

2. The forest ecological environment monitoring system based on data analysis according to claim 1, characterized in that, The modules are connected in the following way: S1. Collect environmental parameters through sensors deployed in the forest area and perform preprocessing; S2. Using sensors as nodes and the spatial correlation between different environmental parameters as edge weights, construct a heterogeneous graph structure, and use the Gaussian kernel function to calculate the edge weights to generate a weighted heterogeneous graph representation. S2 specifically includes: S21. Construct a heterogeneous graph structure, treating the sensors deployed in the forest area as nodes in the graph structure, denoted as the node set. ,in This indicates the total number of sensor nodes deployed in the forest area; S22. Based on the preprocessed environmental parameters, calculate the nodes. and Spatial correlation measure among parameter values ​​collected within a given time window Construct an edge set : ； in, Representing nodes respectively At any moment Environmental parameter observations Represents a node In the time window The average observed value within, Indicates the length of the time window; S23. Calculate nodes using a Gaussian kernel function. and Edge weights between The formula is as follows: ； in, The bandwidth parameter of the Gaussian kernel function is used to adjust the sensitivity of the edge weight distribution. The weighting coefficient representing the distance factor. This represents the natural exponential function with base e. Represents a node and Manhattan distance between them; S24. Set the nodes With edge set and all edge weights The adjacency matrix formed Combining to construct a weighted heterogeneous graph structure It outputs a weighted heterogeneous graph representation; S3. Perform multi-layer graph convolution operations on the weighted heterogeneous graph representation to obtain node feature representations. After each layer of graph convolution operations, perform residual connections on the node features and apply the residual output to the next layer of graph convolution. S3 specifically includes: S31. Represent the input graph as a weighted heterogeneous graph convolutional neural network, and let the initial node feature matrix be... ,in Represents a node The initial feature vector, This indicates the total number of sensor nodes deployed in the forest area; S32. Perform graph convolution operation on each layer, and calculate the node feature update results based on the normalized edge weights: ； in, Indicates the first The node feature matrix after layer graph convolution. Indicates the first The node feature matrix after layer graph convolution. It is an adjacency matrix. This is a node degree matrix, with elements on the main diagonal. , For the first The trainable weight matrix of the layer, It is a fully connected mapping matrix. The cross-channel attention weight matrix is ​​generated by global average pooling and normalized by the Sigmoid function. This represents the Hadamard element-wise product. For the residual connection mapping matrix, Represents a non-linear activation function; S33. After convolving each layer of the graph, the node feature matrix of the previous layer is... With the current layer feature matrix Perform residual fusion and use the fusion result as input for the next layer. The formula for calculating the fusion result is as follows: ； in, The features of the merged nodes, The fusion ratio coefficient, with a value between 0 and 1, represents a trainable parameter; S34. Repeat steps S32 to S33 until the preset number of convolutional layers is reached. Output node feature matrix ; S4. For the node feature representations output by the convolution operation of the last layer graph, perform global pooling and multilayer perceptron mapping in sequence to generate the state feature vectors of each node. S4 specifically includes: S41, where the node feature matrix Each row vector in the dataset undergoes global pooling along the channel dimension to obtain a compressed feature representation: ； in, For nodes The compressed feature vector, Represents a node The graph convolutional feature vector, express Dimensions It is a weighted nonlinear transformation function used to enhance the channel response. For vectors The One portion, It is a trainable linear mapping matrix; S42. Input the compressed feature representation into a... A multilayer perceptron network consisting of layers, where each layer performs linear transformation, normalization, nonlinear activation, and residual mapping, outputs as follows: ； in, For nodes After the first The feature vector output by the layer, For nodes After the first The feature vector output by the layer satisfies ,and , For the first The linear weight matrix of the layer, For the channel normalization matrix, For bias vectors, Represents a non-linear activation function. For the first The residual connectivity coefficient matrix of the layer; S43, the multilayer perceptron network completes... After layer transformation mapping, nodes are obtained. State feature vector The state feature vector is used to perform cluster analysis and ecological indicator assessment; S5. The DENCLUE clustering algorithm is used to perform cluster analysis on the state feature vectors to identify ecological regions with similar state characteristics, and the ecological index values ​​of each region are calculated based on the cluster centers. S6. Compare and analyze the clustering results with historical environmental status data to identify abnormal trends in ecological indicators and generate forest ecological risk monitoring results.

3. The forest ecological environment monitoring system based on data analysis according to claim 2, characterized in that, The environmental parameters include air temperature, air humidity, light intensity, soil moisture, soil pH, carbon dioxide concentration, PM2.5 concentration, wind speed, and wind direction.

4. A forest ecological environment monitoring system based on data analysis according to claim 2, characterized in that, The preprocessing includes missing value imputation, outlier removal, timestamp alignment, format conversion, and data standardization.

5. A forest ecological environment monitoring system based on data analysis according to claim 2, characterized in that, S5 specifically includes: S51. Construct a state feature space, and collect the state feature vectors of all nodes in the state feature space. Calculate the density estimate of each node using a kernel density function: ； in, Represents a node The density estimate, The dimension of the state feature vector. The normalization factor of the kernel function is defined as follows: , Pi is a constant. The kernel density function bandwidth parameter, Represents the state feature vector In the Components in each dimension Represents the state feature vector In the Components in each dimension This indicates the total number of sensor nodes deployed in the forest area; S52. Perform the DENCLUE clustering algorithm in the state feature space. Based on the density estimate of each node, use the density gradient tracking method to iteratively move the position of each node according to the local density increment direction until convergence to the node position with the maximum local density. Group the state feature vectors with the same convergence result into the same cluster to obtain the cluster set. ; S53. Let the cluster set be... ,in This represents the total number of clusters, with each cluster corresponding to an ecological region. Cluster centers are calculated for the state feature vectors of all nodes within each cluster, and the ecological index vector of the region is obtained. ； in, For the first The state feature centers of each cluster The ecological indicator vectors for the ecological region are normalized to obtain the corresponding indicator values. The characteristic transformation matrix, For ecological compensation matrix, Represents clusters The number of nodes included.

6. A forest ecological environment monitoring system based on data analysis according to claim 5, characterized in that, The DENCLUE clustering algorithm starts with the density estimate of each node and introduces a weighted iterative update mechanism using density gradient tracking. In each position iteration, a node density weighting adjustment factor and a dynamic constraint term for feature variance are introduced. Position updates are performed through local density increment directions, and a cluster set is constructed after all nodes have converged. .

7. A forest ecological environment monitoring system based on data analysis according to claim 2, characterized in that, S6 specifically includes: S61. Based on the ecological indicator vector of each region, construct the regional ecological indicator sequence for the current period, and retrieve the ecological indicator sequence of the same region that has been stored in the corresponding historical period as a reference benchmark. S62. Perform time series change alignment analysis on the current ecological indicator sequence and the historical ecological indicator sequence. By calculating the difference between the indicator values ​​at each time point, extract the trend indicators that have shifted in each ecological dimension and generate a regional indicator change trajectory matrix. S63. Based on the offset, offset speed and offset direction of each indicator dimension in the change trajectory matrix, combined with the preset threshold rules, ecological areas with abnormal indicator values ​​are identified, and the degree of abnormality is classified into levels, forming a forest ecological risk monitoring result that includes regional index, cluster number, abnormality level, ecological indicator value and change parameter set.

Images