Ecological asset dynamic monitoring and evaluation method

Abstract

The application provides an ecological asset dynamic monitoring and evaluation method, which comprises the following steps: based on an ecological asset dynamic monitoring model, dynamically collecting attribute characteristics of a target ecological asset and forming a dynamic attribute characteristic sequence according to the attribute characteristics; based on the dynamic attribute characteristic sequence, constructing a dynamic attribute knowledge graph; and based on an ecological asset evaluation model, dynamically evaluating the asset value of the target ecological asset according to the dynamic attribute knowledge graph to generate a dynamic value sequence point and add the dynamic value sequence point to a dynamic value tree. In the application, the attribute changes of the ecological asset in the time and space dimensions are comprehensively captured, the changes of the ecological asset can be reflected in real time through a dynamic construction mode, the asset value of the ecological asset is dynamically evaluated based on the ecological asset evaluation model and according to the dynamic attribute knowledge graph, the dynamic change characteristics of the ecological asset value with time, environment and various factors can be fully considered, the scientificity and practicability of the ecological asset value evaluation are significantly improved, and strong support is provided for ecological protection and management decision-making.

Description

Technical Field

[0001] This invention relates to the field of intelligent processing technology, specifically to a method for dynamic monitoring and evaluation of ecological assets. Background Technology

[0002] In the field of dynamic monitoring and assessment of ecological assets, traditional technologies face numerous bottlenecks as the demands for ecological protection and resource management continue to increase.

[0003] Currently, most ecological asset monitoring methods rely on a combination of decentralized sensor networks and manual inspections. While sensor networks can collect basic environmental data, they lack a systematic data fusion and analysis mechanism. Data from different types of sensors are independent of each other, making it difficult to construct a comprehensive profile of ecological assets. Manual inspections suffer from low efficiency, strong subjectivity, and limited coverage, failing to achieve real-time, dynamic, and comprehensive monitoring of the attributes and characteristics of ecological assets. In the data processing stage, existing technologies typically employ conventional data mining algorithms to analyze monitoring data. These algorithms are mostly based on surface features and simple correlations in the data, making it difficult to uncover complex nonlinear relationships and potential patterns in ecosystems, and failing to effectively address the high dimensionality, multi-source nature, and dynamic characteristics of ecological data.

[0004] In constructing knowledge graphs of ecological assets, traditional methods often employ manual input or automated extraction based on fixed rules. Manual input is extremely inefficient and prone to human error, making it difficult to meet the processing needs of large-scale ecological data. Automated extraction based on fixed rules relies on pre-defined templates and patterns, and struggles to accurately extract entity, relationship, and attribute information when faced with the diverse semantic expressions and complex terminology in the ecological field. The resulting knowledge graph lacks flexibility and dynamic updating capabilities, failing to reflect changes in ecological assets in a timely manner.

[0005] In the valuation of ecological assets, existing valuation techniques mostly employ static valuation index systems and fixed valuation models. For example, the market value approach is used to assess the economic value of ecological assets, and the alternative cost approach is used to estimate the value of ecosystem service functions. However, these methods do not fully consider the dynamic evolution of ecosystems and ignore the characteristics of ecological asset value changing with time, environment, and human activities. Furthermore, the weighting of various influencing factors during the valuation process is often based on subjective experience or simple statistics of historical data, failing to accurately quantify the complex relationships between ecological asset value and various factors. This results in valuation results lacking timeliness and accuracy, making it difficult to meet the practical needs of dynamic management and scientific decision-making regarding ecological assets. Summary of the Invention

[0006] To address the aforementioned technical problems, this application provides a method for dynamic monitoring and assessment of ecological assets, which can at least solve or alleviate the problems existing in the prior art.

[0007] To achieve the above objectives, according to one aspect of this application, a method for dynamic monitoring and assessment of ecological assets is provided, comprising:

[0008] A method for dynamic monitoring and assessment of ecological assets, characterized by including:

[0009] The model for dynamic monitoring of ecological assets is used to dynamically collect the attribute characteristics of target ecological assets and form a dynamic attribute characteristic sequence accordingly.

[0010] Construct a dynamic attribute knowledge graph based on dynamic attribute feature sequences;

[0011] Based on the ecological asset assessment model, the target ecological asset is dynamically valued according to the dynamic attribute knowledge graph, so as to generate dynamic value sequence points and add them to the dynamic value tree.

[0012] The technical solution in this application has at least the following technical advantages:

[0013] ① Traditional monitoring methods are fragmented and lack systematicity. This application constructs a dynamic monitoring model for ecological assets, which can dynamically collect the attribute characteristics of target ecological assets and form a dynamic attribute characteristic sequence. This model can comprehensively capture the attribute changes of ecological assets in time and space. Compared with traditional fragmented and isolated monitoring methods, it achieves dynamic and comprehensive collection of ecological asset attribute characteristics, effectively avoids data gaps and biases, significantly improves the accuracy and completeness of monitoring, and provides a reliable data foundation for subsequent assessment.

[0014] ② Traditional knowledge graph construction relies on manual input or fixed rule extraction, which is inefficient and inaccurate. This application constructs a dynamic attribute knowledge graph based on dynamic attribute feature sequences, which can automatically and accurately extract entity, relationship, and attribute information related to ecological assets. Through dynamic construction, changes in ecological assets can be reflected in real time. Compared with traditional static and passive knowledge graph construction methods, it has greater flexibility and adaptability, and can quickly and accurately integrate ecological asset information to form a complete and dynamic knowledge graph system.

[0015] ③ Existing ecological asset valuation methods employ static indicator systems and fixed models, failing to reflect dynamic changes in value. This application, based on an ecological asset valuation model and utilizing a dynamic attribute knowledge graph, conducts dynamic asset valuation, fully considering the dynamic changes in ecological asset value over time, environmental factors, and various other influences. By generating dynamic value sequence points and adding them to a dynamic value tree, the dynamic evolution process and causal relationships of ecological asset value are visually displayed, providing scientific, real-time, and dynamic valuation results for ecological resource management. This effectively addresses the problems of lagging, lack of timeliness, and inaccuracy in traditional valuation methods, significantly improving the scientific rigor and practicality of ecological asset valuation and providing strong support for ecological protection and management decisions. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a method for dynamic monitoring and assessment of ecological assets, as described in an embodiment of this application. Detailed Implementation

[0017] like Figure 1 As shown, an embodiment of this application provides a method for dynamic monitoring and assessment of ecological assets, which includes:

[0018] The model for dynamic monitoring of ecological assets is used to dynamically collect the attribute characteristics of target ecological assets and form a dynamic attribute characteristic sequence accordingly.

[0019] Construct a dynamic attribute knowledge graph based on dynamic attribute feature sequences;

[0020] Based on the ecological asset assessment model, the target ecological asset is dynamically valued according to the dynamic attribute knowledge graph, so as to generate dynamic value sequence points and add them to the dynamic value tree.

[0021] Preferably, in a scenario, the above-mentioned technical implementation is described in an alternative or preferred manner.

[0022] 1. Based on the dynamic monitoring model of ecological assets, dynamically collect the attribute characteristics of target ecological assets and form a dynamic attribute characteristic sequence.

[0023] 1.1 Construction of Dynamic Monitoring Model for Ecological Assets

[0024] Suppose the target ecological asset is a specific wetland ecosystem. At time step t (t = 1, 2, ..., T), for the i-th attribute feature x... i (t), whose calculation comprehensively considers multi-scale spatiotemporal factors, autocorrelation effects, and external disturbances.

[0025]

[0026] Where: K represents the number of different categories of influencing factors, such as local ecological environment, regional climate patterns, and different levels of human activities. α ik It is the global weight coefficient of the k-th influencing factor on the i-th attribute feature. And 0 < α ik <1, determined using a combination of multi-round iterative analytic hierarchy process (AHP) and principal component analysis (PCA) to comprehensively consider the relative importance of each factor. S is the spatial extent of the ecosystem, and s represents a point in space. β ik (s) is the local weight function of the k-th influencing factor at spatial point s for the i-th attribute feature. It is a spatial weight distribution based on the Gaussian kernel function, reflecting the difference in influence at different spatial locations. Where μ ik It is the spatial center location where the k-th type of influencing factor has the greatest impact on the i-th attribute characteristic, σ ik It represents the standard deviation of the influence range in the control space. γ ik (t-τ ik (s) is a time lag function, τ ik (s) is the time lag of the influence of the k-th type of influencing factor on the i-th attribute at spatial point s. It is determined based on autocorrelation analysis and Granger causality test of historical data. γ ik (t-τ ik (s) uses the exponential decay function γ ik (t-τ ik (s))=exp(-λ ik |t-τ ik (s)|), where λ ik It is the attenuation coefficient. f ik It is a special nonlinear function that describes the relationship between the k-th influencing factor at spatial point s and the i-th attribute characteristic, the ecosystem state S(t), the external environmental factor E(t), and the human activity factor H(t). It is trained using a deep convolutional neural network (DCNN) combined with a long short-term memory network (LSTM), with inputs S(t), E(t), H(t), and s, and outputting the relationship between x and s. i The influence value of (t). ∈ i (t) is the random error term, which follows a mixed normal distribution. Where L is the amount of the mixture components, ω il It is the weight of the l-th mixture component. Used to describe the combined effects of measurement errors from multiple sources and unconsidered factors.

[0027] 1.2 Formation of Dynamic Attribute Feature Sequences

[0028] Suppose there are n attribute features, and the attribute feature vector at time t is... The dynamic attribute feature sequence is {X(1),X(2),…,X(T)}. To consider the correlation and dynamic trends among the attribute features, this application further processes the sequence to obtain a weighted dynamic attribute feature sequence. in: W(t) is an n×n time-varying weight matrix, whose elements w pq (t) represents the influence weight of the p-th attribute feature on the q-th attribute feature at time t. pq (t) is learned through a dynamic Bayesian network (DBN) and dynamically adjusted according to the causal and temporal dependencies between attribute features.

[0029] 2. Construct a dynamic attribute knowledge graph based on dynamic attribute feature sequences.

[0030] 2.1 Entity Recognition and Classification

[0031] To identify entities from dynamic attribute feature sequences, a bidirectional gated recurrent unit (BiGRUAttention) model based on an attention mechanism is employed. Let the input dynamic attribute feature sequence be... For the input at time step t After passing through the BiGRU layer, the hidden state h is obtained. t :

[0032] Then, the attention weight α at each time step is calculated using an attention mechanism. t :

[0033] Among them W a b a and v are learnable parameters. Finally, the weighted hidden state is obtained. Entity classification using a fully connected layer: Where y is the probability distribution of entity classification, W c and b c These are learnable parameters.

[0034] 2.2 Relation Extraction

[0035] To determine the semantic relationships between entities, a joint model based on Graph Convolutional Network (GCN) and multi-head attention mechanism is adopted. Let the entity set be E = {e1, e2, ..., e...} m}, each entity e j The corresponding eigenvector is f j An adjacency matrix A is constructed to represent the initial connection relationships between entities. After passing through the GCN layer, the updated entity feature vector f is obtained. j′ :

[0036] Among them W gcn and b gcn These are the learnable parameters of the GCN layer. Then, the strength of the relationships between entities is calculated using a multi-head attention mechanism.

[0037] Where Q = W q F′, K = W k F′, V=W v F′, F′=[f 1′ ,f 2′ ,…,f m′ ] T W q W k and W v It is a learnable parameter, d k This is the dimension of the key vector. After multi-head attention, we obtain the relation strength matrix R, whose elements r jk Represents entity e j and e k The strength of the relationship between them.

[0038] 2.3 Knowledge Graph Construction

[0039] The identified entities and extracted relationships are integrated into a knowledge graph. The knowledge graph is represented as a graph G = (V, E), where V is the set of nodes, corresponding to entities; and E is the set of edges, corresponding to the relationships between entities. To account for the dynamic nature of the knowledge graph, this application introduces a time dimension, resulting in a dynamic knowledge graph G(t) = (V(t), E(t)), where V(t) and E(t) change dynamically with time t. The attribute updates of nodes and edges are optimized using a reinforcement learning algorithm to adapt to the dynamic changes in the characteristics of ecological asset attributes.

[0040] 3. Based on the ecological asset assessment model, dynamic asset value assessment of the target ecological asset is performed according to the dynamic attribute knowledge graph to generate dynamic value sequence points and add them to the dynamic value tree.

[0041] 3.1 Construction of Ecological Asset Assessment Model

[0042] Suppose the total value V(t) of the target ecological asset consists of multiple levels of value, and a hierarchical weighted evaluation model is adopted: Where: L represents the number of value levels, such as direct economic value, indirect ecosystem service value, and potential value. ω l (t) is the weight of the l-th value level at time t, which is dynamically adjusted using Dynamic Hierarchical Analysis (DAHP) combined with expert opinions and real-time data. l It represents the number of value components at the l-th value level. il(t) is the weight of the i-th value component at time t under the l-th value level, determined by a combined weighting method based on entropy weighting and grey relational analysis. il (t) is the value of the i-th value component at time t under the l-th value level. Its calculation considers the dynamic attribute knowledge graph G(t) and the attribute feature sequence. Where η il (τ) is the time discount function, using the hyperbolic discount function. k il It is a discount factor, determined based on the time sensitivity of different value components. g il It is a special nonlinear function, trained using the Deep Deterministic Policy Gradient (DDPG) algorithm in deep reinforcement learning, with inputs G(τ) and The output is v il (τ).

[0043] 3.2 Generation and Addition of Dynamic Value Sequence Points to the Dynamic Value Tree

[0044] At each time step t, the calculated total value V(t) constitutes the dynamic value sequence points. In this application, the dynamic value tree is a multi-branch tree structure T = (N, E) T ), where N is the set of nodes, and each node n corresponds to the value V(t) of a time step and related attribute information (such as value composition, influencing factors, etc.); E T It is a set of edges, where edges represent temporal order and value transfer relationships. The dynamic value sequence point V(t) at each time step and its related information are added to the dynamic value tree using a tree insertion algorithm, while simultaneously updating the tree structure and node attributes to reflect the dynamic changes and causal relationships of ecological asset value.

[0045] Therefore, the above-mentioned alternative or preferred technical implementations have the following technical advantages.

[0046] 1. Traditional ecological asset monitoring often employs simple linear regression models or static analysis methods based on a single data source. These methods tend to consider only a few factors and neglect the spatiotemporal complexity and dynamic changes of ecosystems. For example, predicting vegetation growth based solely on average temperature and precipitation data over a specific period fails to account for local environmental differences across regions and time lag effects. Furthermore, traditional methods handle measurement errors and unconsidered factors in a simplistic manner, typically assuming a single normal distribution, which cannot accurately describe specific sources of error.

[0047] In this application, A spatial weighting function β is introduced. ik (s) and time lag function γ ik (t-τ ik(s)) can comprehensively consider the impact of different spatial locations and time delays on the characteristics of ecological assets. In wetland ecosystems, local environmental factors such as hydrological conditions and soil types vary greatly in different regions. Spatial weighting functions can more accurately capture these differences. At the same time, ecosystem responses often exhibit time lags, and time lag functions reflect this delay effect, improving monitoring accuracy. A special nonlinear function f is used. ik This method describes the relationships between ecosystem state, external environmental factors, and human activities. Trained using a deep convolutional neural network (DCNN) combined with a long short-term memory network (LSTM), it learns the complex nonlinear relationships between these factors. Compared to traditional linear models, it better reflects the actual operating patterns of ecosystems. (Random error term ∈...) i (t) The use of a mixed normal distribution can more accurately describe the combined effects of measurement errors from multiple sources and unconsidered factors, thus improving the robustness of the model.

[0048] 2. Traditional entity recognition and relation extraction methods are mostly based on rules or simple machine learning algorithms, such as Naive Bayes classifiers and support vector machines. These methods have limited ability to understand specific texts and semantics, and struggle to handle the rich terminology and unique semantic relationships in ecological domains. Furthermore, traditional methods often ignore the time-series and contextual information of the data, resulting in lower accuracy in entity recognition and relation extraction.

[0049] In this application, a Bi-Gated Recurrent Unit (BiGRUAttention) model based on an attention mechanism is employed for entity recognition. The attention mechanism automatically focuses on important parts of the input sequence, and when processing dynamic attribute feature sequences of ecological assets, it can better capture key information at different time steps, improving the accuracy of entity recognition. For example, when identifying different biological species in a wetland ecosystem, it can more accurately classify them based on their growth cycle, environmental adaptability, and other dynamic characteristics. A joint model based on Graph Convolutional Networks (GCN) and multi-head attention mechanisms is used for relation extraction. GCN can effectively process graph-structured data, representing entities and relations in the ecological asset knowledge graph as a graph structure, and updating the feature representation of entities through graph convolution operations. The multi-head attention mechanism can capture the strength of relationships between entities from multiple perspectives, improving the accuracy and robustness of relation extraction. In wetland ecosystems, it can more accurately identify complex relationships such as competition and symbiosis among biological species.

[0050] 3. Traditional ecological asset valuation models often employ static valuation methods, such as the market value approach and the replacement cost approach. These methods typically only consider the current state and direct economic value of ecological assets, neglecting the dynamic changes in the ecosystem and the indirect value of ecosystem services. Furthermore, traditional methods often use fixed weights and simple linear combinations to calculate value, failing to reflect the dynamic characteristics of ecological asset value changing over time and with the environment. This application addresses these issues by... A hierarchical weighted assessment model was implemented, dividing the value of ecological assets into multiple levels, such as direct economic value, indirect ecosystem service value, and potential value. The weights of each level and component were dynamically adjusted using the Dynamic Analytic Hierarchy Process (DAHP) and a combined weighting method. This approach can more comprehensively consider the value composition of ecological assets and dynamically adjust weights according to different time and environmental conditions, improving the accuracy and flexibility of the assessment. Value components v il The calculation of (t) takes into account the dynamic attribute knowledge graph G(t) and the attribute feature sequence. And introduce the time discount function η il (τ). The nonlinear function g trained by the Deep Deterministic Policy Gradient (DDPG) algorithm in deep reinforcement learning. il It allows us to learn the complex relationships between the value of ecological assets and various factors. The time discount function can reflect the discounting effect of future value, which is more in line with the needs of practical economic decision-making.

[0051] 4. Traditional ecological asset assessment results are usually presented in static reports or simple numerical forms, which cannot intuitively show the dynamic changes and causal relationships of ecological asset value over time. When making decisions, decision-makers find it difficult to obtain comprehensive information from these static results, and are also unable to conduct effective predictions and analyses.

[0052] In this application, the dynamic value tree is a multi-branch tree structure, where each node corresponds to the value and related attribute information at a given time step, and edges represent the temporal order and value transfer relationships. By adding dynamic value sequence points and their related information to the dynamic value tree, the dynamic changes and causal relationships of ecological asset value can be intuitively displayed. Decision-makers can traverse the dynamic value tree to understand the value composition, influencing factors, and trends of value changes at different time points, thereby making more scientific and rational decisions. For example, in the protection and management of wetland ecosystems, decision-makers can adjust protection strategies and resource allocation based on the information from the dynamic value tree to maximize the value of ecological assets.

[0053] Optionally, dynamic monitoring is carried out by dynamically collecting the attribute characteristics of the target ecological assets based on the dynamic monitoring model of ecological assets and forming a dynamic attribute characteristic sequence accordingly. This includes:

[0054] The attribute feature carriers collected by the attribute sensors are acquired, and feature extraction is performed on the attribute feature carriers to obtain the attribute features of the target ecological asset.

[0055] Preferably, in a scenario, the above-mentioned technical implementation is described in an alternative or preferred manner.

[0056] 1. Acquire attribute feature carriers collected by attribute sensors

[0057] 1.1 Property Sensors and Data Acquisition

[0058] In ecological asset monitoring, various types of attribute sensors work together to collect data. Taking a marine ecosystem as an example, the sensors involved include: water quality sensors: used to measure water quality parameters such as pH, dissolved oxygen (DO), chemical oxygen demand (COD), and salinity; biosensors: monitoring biological indicators such as plankton abundance and chlorophyll concentration; and meteorological sensors: collecting meteorological information such as sea surface temperature, wind speed, wind direction, and sunlight. These sensors collect data at different locations and depths, and the data is recorded in time-series format and stored in a distributed database. Each data record includes a sensor ID, acquisition time t, acquisition location p = (x, y, z) (three-dimensional coordinates), and measured value v.

[0059] 1.2 Spatiotemporal Distribution of Attribute Feature Carriers

[0060] Let S be the set of sensors, and s∈S represent a sensor. For sensor s, the data collected at time t and location p constitutes an attribute feature carrier d(s,t,p). Considering the spatiotemporal characteristics of sensor networks, data acquisition exhibits spatiotemporal correlation. This application uses the spatiotemporal covariance function C(t1,p1,t2,p2) to describe this correlation:

[0061] Where: σ 2 θ represents the overall variance of the data. t and θ p These are temporal and spatial smoothing parameters, respectively, which control the rate of decay of spatiotemporal correlation. λ t and λ p These are the time and space scale parameters, which determine the range of spatiotemporal correlation.

[0062] 2. Extract features from attribute feature carriers.

[0063] 2.1 Feature Extraction Model

[0064] Let D be the set of attribute feature carriers. For the i-th attribute feature x i The feature extraction process is represented as follows:

[0065]

[0066] Where: S is the set of sensors. t' is the historical time, from time 1 to the current time t. Ω is the set of locations within the monitoring area. α i (s,t′,p) represents the weight coefficients of the i-th attribute feature at sensor s, time t′, and location p, dynamically determined using a Spatiotemporal Weighted Regression (STWR) algorithm combined with a Kalman filter. STWR considers the spatiotemporal nonstationarity of the data, while the Kalman filter is used to update the weights in real time. i It is a special feature transformation function that considers the collected data d(s,t′,p), the time difference t-t′, and the location distance ||p-p0|| (p0 is the location of the target monitoring point). Taking the extraction of biodiversity characteristics of marine ecosystems as an example, f i It is a deep learning-based convolutional recurrent neural network (CNNRNN) model. The input is biological data collected by different sensors at different spatiotemporal points, and the output is an estimate of biodiversity. i (t) is the random error term, which follows a mixture Gaussian distribution. Where K is the amount of the mixture components, ω ik It is the weight of the k-th mixture component. It is used to describe errors from multiple sources, such as sensor noise and model uncertainty.

[0067] 2.2 Feature Transformation Function

[0068] Using CNNRNN-based biodiversity feature extraction function f i For example: First, the CNN part is used to extract spatial features. Let the input biological data matrix be X. s,t′ (Data from sensor s at time t') is processed through l convolutional layers to obtain the feature map. Among them: W l This is the convolution kernel of the l-th layer, where * indicates the convolution operation. b l This is the bias term. ReLU is the activation function, ReLU(x) = max(0,x). Then, the feature map output by the CNN is flattened and input into the RNN part. The hidden state update formula of the RNN is: h s,t =tanh(W h h s,t′-1 +W x F s,t′ +b h ), where: h s,t′ It is the hidden state of time t'; W h and W xThese are the weight matrices for the hidden state and the input, respectively. h This is the bias term. Finally, the estimated biodiversity is obtained through a fully connected layer: f i (d(s,t′,p),tt′,||p-p0||)=W o h s,t′ +b o , where: W o This is the weight matrix of the output layer. o It is a bias term.

[0069] 3. Obtain the attribute characteristics of the target ecological asset.

[0070] Through the above feature extraction process, this application obtains n attribute features of the target ecological asset at time t, which constitute an attribute feature vector:

[0071]

[0072] The dynamic attribute feature sequence is {X(1),X(2),…,X(T)}, where T is the total number of monitoring time steps. This sequence will serve as the foundational data for subsequent construction of a dynamic attribute knowledge graph and ecological asset assessment, providing strong support for a comprehensive and accurate evaluation of the value and health status of ecological assets.

[0073] Therefore, the above-mentioned alternative or preferred technical implementations have the following technical advantages.

[0074] 1. Traditional ecological asset monitoring data acquisition often fails to consider the spatiotemporal correlation of sensor data. Different sensors operate independently, and data processing frequently employs simple averaging or statistical methods, neglecting to delve into the intrinsic temporal and spatial relationships within the data. For example, in marine ecological monitoring, traditional methods only analyze water quality data collected by sensors at different locations, ignoring the mutual influence between data from different locations and at different time points. Furthermore, traditional methods handle sensor errors simplistically, generally assuming a single normal distribution, which fails to accurately describe specific error sources.

[0075] This application introduces a spatiotemporal covariance function C(t1,p1,t2,p2) to describe the spatiotemporal correlation of sensor data. In marine ecosystems, water quality, biological data, and other information at different locations influence each other with changes in time and space. This function allows for a more accurate capture of these spatiotemporal variation patterns, improving data utilization efficiency. For example, when predicting the amount of plankton in a certain area, combining relevant data from surrounding areas at different times makes the prediction more consistent with reality. Using a Gaussian mixture distribution to describe the random error term εe(t) more accurately reflects errors from multiple sources, such as sensor noise and model uncertainty. Compared to the traditional assumption of a single normal distribution, the Gaussian mixture distribution better fits specific error distributions, improving data reliability and the accuracy of subsequent analysis.

[0076] 2. Traditional feature extraction methods are mostly based on simple linear models or empirical formulas, which cannot handle particularly nonlinear relationships. In ecosystems, various ecological indicators have unique interactions, and traditional methods struggle to accurately describe these relationships. For example, when assessing marine biodiversity, traditional methods only consider a simple combination of a few biological indicators, ignoring the complex impact of environmental factors (such as water quality and meteorology) on biodiversity. Furthermore, traditional methods typically do not consider the spatiotemporal dynamics of data, employing static feature extraction methods that cannot adapt to the dynamic characteristics of ecosystems. In this application, a feature transformation function f_{i} based on CNNRNN is used for feature extraction. The CNN part automatically extracts the spatial features of the data, while the RNN part processes the time-series information of the data. In marine ecological monitoring, this combination can fully exploit the spatial and temporal characteristics of biological data, more accurately estimating ecological indicators such as biodiversity. For example, CNN identifies the spatial distribution patterns of biological communities in different regions, while RNN tracks the changing trends of biological communities over time, thereby improving the accuracy and reliability of feature extraction. Weight coefficient α i The values ​​(s,t′,p) are dynamically determined using a combination of Spatiotemporal Weighted Regression (STWR) and Kalman filtering. STWR considers the spatiotemporal nonstationarity of the data and can adjust the weights according to the characteristics of the data at different spatiotemporal locations; Kalman filtering updates the weights in real time to adapt to the dynamic changes in the ecosystem. In marine ecological monitoring, the ecological environment varies greatly in different seasons and regions. Dynamic weight adjustment makes the feature extraction process more flexible and more accurately reflects the actual situation of the ecosystem.

[0077] 3. Traditional methods for monitoring and assessing ecological assets often separate data collection, feature extraction, and assessment, lacking systematicity and holistic approach. Insufficient information transfer and interaction between these stages lead to biased assessment results. Furthermore, traditional methods struggle to monitor and accurately predict dynamic changes in ecosystems in real time, failing to meet the practical needs of ecological protection and management. This application organically integrates data collection and feature extraction, fully considering the spatiotemporal characteristics and complex relationships of the data. From the data collection stage, spatiotemporal correlation is considered, and a special model and dynamic weight adjustment are used in the feature extraction stage. The entire process forms a complete system that can more comprehensively and accurately reflect the attributes and characteristics of ecological assets. For example, in marine ecosystems, this systematic processing allows for a more accurate assessment of the health status and value of marine ecological assets. Due to the use of dynamic models and real-time updated weights, the new method can track ecosystem changes in real time and make more accurate predictions of future trends. In marine ecological protection and management, decision-makers can adjust protection strategies promptly based on real-time monitoring and prediction results, taking effective measures to address ecological problems and improving the efficiency and effectiveness of ecological protection.

[0078] Optionally, the attribute feature carriers collected by the attribute sensors are acquired, and feature extraction is performed on the attribute feature carriers to obtain the attribute features of the target ecological asset, including:

[0079] The acquired attribute feature carriers are subjected to quantum state fusion to generate a quantum state hybrid dataset;

[0080] The quantum state hybrid dataset is mapped as dynamic spatiotemporal graph structure data;

[0081] The dynamic spatiotemporal graph structure data is cross-domain mapped, and cross-domain semantic information is given to the dynamic spatiotemporal graph structure data based on graph attention to generate a core feature subset with evaluation value.

[0082] Input a subset of core features into the encoder to learn the probability distribution representation of the features and generate a latent feature vector;

[0083] The generator generates enhanced feature samples based on the latent feature vectors, while the discriminator distinguishes between the real core features and the generated feature samples.

[0084] The variational autoencoder reconstructs and optimizes the real core features and generated feature samples to generate the target ecological asset attribute features.

[0085] Preferably, in a scenario, the above-mentioned technical implementation is described in an alternative or preferred manner.

[0086] 1. Perform quantum state fusion on the acquired attribute feature carriers to generate a quantum state hybrid dataset.

[0087] 1.1 Fusion of higher-order quantum states

[0088] Let the obtained attribute feature carrier set be Each d i Mapping to the quantum state space yields the density matrix ρ i To consider the fusion of higher-order quantum states, a higher-order fusion operator is introduced. The generated quantum state hybrid dataset ρ mix Represented as:

[0089] Where: k is the order of fusion, k≥1, the higher the order, the higher the complexity of fusion and the degree of information integration. These are higher-order weighting coefficients, satisfying... and These weighting coefficients can be determined by combining quantum genetic algorithms with multi-objective optimization, taking into account factors such as the reliability of attribute feature carriers, their importance to the overall evaluation, and the degree of quantum entanglement between them. It is a higher-order fusion operator, for example, for k=2, it is defined as in It is the tensor product, [·,·] is the commutator, and λ is an adjustment parameter used to balance the contribution of the tensor product and the commutator.

[0090] 2. Map the quantum state hybrid dataset to dynamic spatiotemporal graph structure data.

[0091] 2.1 Construction of Dynamic Spatiotemporal Graphs Based on Fractal Dimension

[0092] The dynamic spacetime graph G = (V, E, T) is given, where the node set V is mapped from elements of a quantum state hybrid dataset. For node v... j and v k Edge weight e jk The calculation takes into account fractal dimension and multi-scale spatiotemporal correlation.

[0093] Let the spacetime point (t) j ,p j ) and (t k ,p k The multi-scale spatiotemporal correlation function is C. ms (t j ,p j ,t k ,p k ), defined as:

[0094] Where: L is the scale number, β l It is the weight coefficient of the l-th scale. And β l ≥0. C l(t j ,p j ,t k ,p k ) is the spatiotemporal covariance function at the l-th scale, for example Parameters at different scales different.

[0095] Fractal dimension D of node features f (x j The edge weight e is calculated using methods such as box counting. jk for:

[0096] Where f(x) j ,x k ) is a feature correlation function, such as cosine similarity.

[0097] 3. Perform cross-domain mapping on the dynamic spatiotemporal graph structure data, and assign cross-domain semantic information to the dynamic spatiotemporal graph structure data based on graph attention to generate a core feature subset with evaluation value.

[0098] 3.1 Multimodal cross-domain mapping

[0099] Let X be the node feature matrix and A be the edge adjacency matrix of the dynamic spatiotemporal graph G = (V, E, T). Multimodal cross-domain mapping considers various different mapping functions φ. m (m = 1, 2, ..., M), mapping the node feature matrix X to multiple new feature spaces X m′ :X m′ =φ m (X) where phi_m is a deep neural network with different structures, such as convolutional neural network (CNN), recurrent neural network (RNN), etc.

[0100] Then, the results from these different feature spaces are fused to obtain the final cross-domain mapping result X′: Where γ m These are the fusion weight coefficients, determined through an adaptive weight learning algorithm (such as a gradient descent-based weight update algorithm).

[0101] 3.2 Multi-head graph attention mechanism combined with fuzzy logic

[0102] Multi-head graph attention mechanisms are used to imbue dynamic spatiotemporal graph structure data with cross-domain semantic information. For node v i Attention coefficient of the h-th head The calculation formula is as follows:

[0103] Among them W h and a h It is the learnable parameter of the h-th head.

[0104] Introducing fuzzy logic, defining fuzzy attention coefficients in It is a fuzzy membership function, for example Sim(x i ,x j ) represents the node feature similarity, and β and τ are fuzzy control parameters.

[0105] node v i New feature h i for: Where H is the number of heads and σ is the activation function. This method generates a subset of core features with evaluable value.

[0106] 4. Input a subset of core features into the encoder to learn the probability distribution representation of the features and generate latent feature vectors.

[0107] 4.1 Deep Hierarchical Encoder and Gaussian Mixture Variational Inference

[0108] The encoder employs a deep hierarchical structure, consisting of multiple coding layers. Let the l-th layer encoder be... in

[0109] The encoder learns the probability distribution representation of features through Gaussian mixture variational inference. The encoder outputs the mixture Gaussian distribution parameters of the latent feature vector z, i.e., the mean μ of each mixture component. c Standard deviation σ c and weight π c (c = 1, 2, ..., C): The latent eigenvector z is obtained by sampling from a mixture of Gaussian distributions: Where ∈ c It is a random vector sampled from the standard normal distribution N(0,1).

[0110] 5. The generator generates enhanced feature samples based on the latent feature vectors, while the discriminator distinguishes between the real core features and the generated feature samples.

[0111] 5.1 Multi-scale Generative Adversarial Networks (GANs) combined with residual connections

[0112] Generator G θ (z) Employs a multi-scale structure consisting of multiple generative subnetworks. It consists of a network, with each subnetwork responsible for generating features at different scales. Where δ m These are the scale fusion weight coefficients, optimized using a reinforcement learning algorithm. Simultaneously, the generator introduces residual connections; for the nth layer of the generator subnetwork, the output is: Where y0 = z.

[0113] Discriminator It also employs a multi-scale structure to discriminate features at different scales. The loss functions of both the generator and discriminator incorporate an adversarial gradient penalty term L. gp :

[0114]

[0115] Where λ gp It is the gradient penalty coefficient. It is the interpolation distribution of real features and generated features.

[0116] 6. The variational autoencoder reconstructs and optimizes the real core features and generated feature samples to generate the target ecological asset attribute features.

[0117] 6.1 Variational Autoencoder Combined with Quantum Entanglement Regularization

[0118] Variational autoencoders are encoders and generator Composition. The loss function of the variational autoencoder, in addition to reconstruction error and KL divergence, introduces a quantum entanglement regularization term L. qe :

[0119] Where λ qe These are the quantum entanglement regularization coefficients. The quantum entanglement regularization term L... qe It is used to measure the degree of quantum entanglement between potential eigenvectors z, and is calculated using methods such as quantum mutual information.

[0120] 1. Variational Autoencoder (VAE) Loss Function and Optimization

[0121] Variational autoencoders contain encoders and generator Its loss function L VAE as follows:

[0122] Reconstruction error term In this application, this error term measures the feature samples reconstructed by the generator based on the latent feature z. With true core features The differences. In ecological asset monitoring, if the core characteristics... This includes vegetation biomass and species diversity in terrestrial ecosystems, and water quality parameters and biological density in aquatic ecosystems. Desired operation. This is achieved by sampling from the distribution of latent features z output by the encoder. This error term prompts the generator to reconstruct these features as accurately as possible to narrow the gap with the true features.

[0123] KL divergence term In this application, KL divergence is used to measure the distribution of the latent feature z of the encoder output. The difference from the prior distribution p(z) (usually set as a standard normal distribution N(0,I)). In the context of ecological assets, it makes the latent feature distribution learned by the encoder closer to the standard normal distribution, ensuring that the latent feature space has a good structure and that features of different ecological regions or types are reasonably distributed in the latent space, which is beneficial for subsequent generation and analysis.

[0124] Quantum entanglement regularization term λ qe L qe In this application, this term measures the degree of quantum entanglement between potential eigenvectors z, λ qe This is a regulation coefficient. In ecosystems, different ecological characteristics have complex relationships; for example, in marine ecosystems, water quality parameters and biological indicators influence each other. This regularization term prompts the latent feature space to capture these complex relationships, making the generated feature samples more closely match the actual situation of the ecosystem.

[0125] By calculating the loss function L VAE The gradients of the encoder parameters φ and generator parameters θ are then used to update the parameters using an optimization algorithm. Taking the Adam optimization algorithm as an example, its update formula is as follows:

[0126] For encoder parameter φ: For the generator parameter θ:, Where α is the learning rate. and These are the first-order and second-order moment estimates for the encoder parameter φ. and These are the first-order and second-order moment estimates of the generator parameter θ, and ∈ is a small constant to prevent the denominator from being zero.

[0127] 2. Reconstruction Optimization and Generation of Target Ecological Asset Attributes

[0128] During the optimization process, as the number of iterations increases, the loss function L... VAE The reduction in the reconstruction error term indicates an improvement in the generator's ability to reconstruct the true core features. The reduction in the KL divergence term makes the latent feature space more consistent with the prior distribution, and the quantum entanglement regularization term enables the latent features to capture the complex relationships between ecological features. After multiple iterations of optimization, the encoder can effectively map the true core features to the latent feature space, and the generator can generate high-quality feature samples based on the latent features. These samples are the target ecological asset attribute features.

[0129] 3. Parameter representation of the attribute characteristics of the target ecological assets

[0130] 3.1 Terrestrial Ecosystems

[0131] forest ecosystem

[0132] Biomass: including aboveground biomass B a (unit: tons / hectare) and underground biomass B u (Unit: tons / hectare). It can be estimated using remote sensing data combined with biomass models, such as B... a = f(NDVI, LAI, ...), where NDVI is the normalized difference vegetation index and LAI is the leaf area index. Species diversity: represented by the Shannon-Wiener index H. Where S is the total number of species, p i It represents the proportion of individuals of the i-th species out of the total number of individuals. Carbon sink C s (Unit: tons of carbon dioxide / hectare·year): This can be estimated by measuring biomass growth and carbon content, combined with ecological models, such as C s =g(B a B u ,…).

[0133] grassland ecosystem

[0134] Vegetation Coverage V c : Expressed as a percentage, it can be obtained through remote sensing image analysis or field quadrat surveys, such as V c = h(RGB, NIR, ...), where RGB is the visible light band and NIR is the near-infrared band. Grass yield Y g (Unit: kg / ha): Estimated by weighing forage in field-harvested quadrats, and related to factors such as vegetation cover and soil fertility, such as Y g =k(V c OM, ...), where OM is the soil organic matter content. Soil fertility includes soil organic matter content OM (expressed as a percentage), soil nitrogen content N (unit: g / kg), soil phosphorus content P (unit: g / kg), and soil potassium content K (unit: g / kg), determined through soil sampling and chemical analysis.

[0135] 3.2 Aquatic Ecosystem

[0136] Marine ecosystem

[0137] Water quality parameters, such as dissolved oxygen (DO) (unit: mg / L), chemical oxygen demand (COD) (unit: mg / L), and pH, are monitored in real time by water quality sensors.

[0138] Biological parameters: Plankton density D p(Unit: individuals / liter), determined by microscopic observation and counting; Fish resource quantity B f (Unit: tons), estimated using fishery resource survey methods, such as B f =l(D p (temp, ...), where temp is seawater temperature. Ecosystem service function parameter: Fishery output Y f (Unit: tons), obtained through fisheries statistics; coastal protection function can be measured by the coastal zone topographic change rate R. t Evaluation of indicators, such as ΔT is the change in coastal topography over a certain period of time, and T is the initial topographic parameter.

[0139] Freshwater ecosystems (rivers, lakes, etc.)

[0140] Hydrological parameters: flow rate Q (unit: cubic meters per second) and water level H (unit: meters), are monitored in real time by hydrological station measuring equipment.

[0141] Water quality parameters: Total phosphorus (TP) (mg / L), total nitrogen (TN) (mg / L), and transparency (SD) (m), measured by chemical analysis and Seychelles disc. Biological parameters: Benthic fauna diversity was measured using the Simpson Diversity Index (DDI). s express, Aquatic plant coverage V a (Percentages are expressed) and obtained through remote sensing image analysis or field surveys.

[0142] By optimizing the loss function of the variational autoencoder, the real core features and generated feature samples are reconstructed and optimized, ultimately generating the target ecological asset attribute features.

[0143] Therefore, in the field of dynamic monitoring and assessment of ecological assets, this application demonstrates a fundamentally different technical nature from traditional technologies in terms of data processing, feature extraction, and model optimization, bringing significant technical advantages to ecological asset monitoring and assessment scenarios as follows:

[0144] 1. Traditional data fusion methods often employ simple weighted averaging or rule-based fusion strategies, which are insufficient for uncovering complex relationships between data. When constructing spatiotemporal map structures, they typically only consider spatiotemporal correlations at a single scale, making it difficult to reflect the unique spatiotemporal variation characteristics of ecosystems. For example, in forest ecological monitoring, traditional methods simply summarize sensor data from different regions, ignoring the interactions of ecological factors at different vegetation heights and time scales. This application employs a higher-order fusion operator. and multi-order weight coefficients The weights are optimized using a quantum genetic algorithm, fully considering the quantum entanglement characteristics and complex correlations among ecological data. Taking wetland ecosystems as an example, this method can more accurately fuse multi-source data such as water quality, biology, and meteorology, avoiding information loss caused by traditional weighted averaging and improving the depth and accuracy of data fusion. Dynamic spatiotemporal graph construction based on fractal dimension: This involves introducing fractal dimension and a multi-scale spatiotemporal correlation function C. ms Calculating edge weights can capture the self-similarity and spatiotemporal dynamics of ecosystems at different scales. In marine ecological monitoring, it can more meticulously depict the interactions of ecological factors in different sea areas and at different times, and compared with traditional single-scale spatiotemporal maps, it can more realistically reflect the complex structure and dynamic evolution of ecosystems.

[0145] 2. Traditional cross-domain mapping methods typically rely on manually designed mapping rules, lacking the ability to adaptively learn from the inherent characteristics of the data. Regarding graph attention mechanisms, traditional methods often involve single-attention computation, making it difficult to handle the fuzziness and multimodal information in ecological data. For example, in ecological assessments, traditional mapping methods cannot effectively convert ecological monitoring data into features that meet assessment requirements, and their ability to handle complex relationships between ecological factors is limited.

[0146] This application utilizes a variety of different mapping functions φ m (e.g., CNN, RNN) performs feature mapping, and determines the fusion weight γ through an adaptive weight learning algorithm. m It can automatically learn feature representations from different modalities of data, achieving efficient conversion from ecological monitoring data to assessment features. When assessing the value of forest ecological assets, it can simultaneously process multimodal data such as vegetation spectrum, topography, and meteorology, improving the accuracy and applicability of feature conversion. It calculates the importance of features from different perspectives through multi-head attention, combined with fuzzy membership functions. Addressing the ambiguity and uncertainty of ecological data. In wetland ecosystems, it can more accurately identify the complex relationships between different biological populations and environmental factors. Compared with traditional single attention mechanisms, it can more comprehensively capture the semantic information of the ecosystem, providing a richer subset of features for ecological assessment.

[0147] 3. Traditional encoder generator models have simple structures and weak learning capabilities for data probability distributions, resulting in low-quality generated feature samples. In variational autoencoders, traditional methods typically fail to consider the specific relationships within ecological data, and the optimization process lacks specificity. For example, in ecological feature generation, features generated by traditional models do not match the actual ecological situation, failing to provide effective support for ecological assessment. This application employs a deep hierarchical encoder, combined with Gaussian mixture variational inference to learn the probability distribution of features, enabling more accurate capture of the complex distribution characteristics of ecological data. When processing ecological asset features from different regions and times, it can generate potential feature vectors that better reflect the actual distribution, laying the foundation for generating high-quality ecological feature samples. The generator uses a multi-scale structure and residual connections, combined with an adversarial gradient penalty term, improving the diversity and realism of generated samples. When generating ecological asset attribute features, it can generate feature samples at different scales that are closer to the real ecological situation, achieving better generation results compared to traditional generative adversarial networks. A quantum entanglement regularization term L is introduced into the variational autoencoder loss function. qe This enables the model to learn the complex relationships between ecological features. In ecological asset assessment, the generated attribute features can better reflect the interactions within the ecosystem. The optimized feature samples provide more reliable data support for the dynamic monitoring and accurate assessment of ecological assets. Compared with traditional variational autoencoders, it has stronger adaptability and accuracy in ecological application scenarios.

[0148] Optionally, a dynamic attribute knowledge graph is constructed based on the dynamic attribute feature sequence, including: extracting entities, relationships, and attributes from the dynamic attribute feature sequence to obtain the entities of the ecological asset, the semantic relationships between entities, and the attribute information of the entities, respectively; preprocessing the extracted entities and updating the semantic relationships and attribute information based on the preprocessed entities, wherein the preprocessing includes alignment and disambiguation processing; loading the entities, the semantic relationships between entities, and the attribute information of the entities into a graph data structure and loading it into the constructed initial state knowledge graph to form the dynamic attribute knowledge graph of the target ecological asset.

[0149] Preferably, in a scenario, the above-mentioned technical implementation is described in an alternative or preferred manner.

[0150] 1. Extract entity, relation, and attribute data from dynamic attribute feature sequences.

[0151] 1.1 Entity Extraction

[0152] Let the dynamic attribute feature sequence be S = {s1, s2, ..., s}. n}, where s iLet represent the i-th feature in the sequence. The goal of entity extraction is to identify entities related to ecological assets from this sequence. This application uses a Conditional Random Field (CRF) model for entity extraction. The probability formula for the CRF model is: Where: x is the input dynamic attribute feature sequence, and y is the corresponding entity label sequence. Z(x) is the normalization factor. λ k It is the kth characteristic function f k The weights are learned through training data. k (y i-1 ,y i ,x,i) is a feature function used to describe the local features of the label sequence y at position i. For example, when x is a dynamic attribute feature sequence about a forest ecosystem, entities such as "trees", "birds", and "soil" can be identified.

[0153] 1.2 Relation Extraction

[0154] The purpose of relation extraction is to determine the semantic relationships between entities. This application uses a deep learning-based relation extraction model, such as a convolutional neural network (CNN). Let the input entity pair be (e1, e2), and their corresponding feature vectors be respectively... and They are concatenated into a new vector

[0155] The formula for the convolutional layer of a CNN model is: c j =ReLU(W j v+b j ), where: W j b is the weight matrix of the j-th convolutional kernel. j This is the bias term. ReLU is the activation function, ReLU(x) = max(0,x). After processing by convolutional and pooling layers, a fixed-length feature vector h is obtained, which is then classified through a fully connected layer to predict the relationship r between entity pairs.

[0156] Among them: W h It is the weight matrix of the fully connected layer, b h It is a bias term. The softmax function is used to convert the output into a probability distribution. m is the number of categories of the relationship. For example, in a forest ecosystem, a relationship such as "trees provide habitat for birds" would be identified.

[0157] 1.3 Attribute Extraction

[0158] Attribute extraction is the process of extracting relevant attribute information from an entity. This application uses a combination of rule-based and machine learning methods. Let entity e be the entity, and its corresponding dynamic attribute feature sequence be S. e For each attribute type a, this application defines a rule set R. a And a machine learning model M a .

[0159] First, use rule set R a For S e Perform a match; if a match is found, extract the corresponding attribute value v. a If a rule match fails, a machine learning model M is used. a Prediction is performed. For example, for the entity "tree", the attribute types include "tree height" and "diameter at breast height". Through rule matching or machine learning models, this application extracts attribute information such as "tree height: 10 meters" and "diameter at breast height: 0.5 meters" from the dynamic attribute feature sequence.

[0160] 2. Preprocess the extracted entities, and update the semantic relationships and attribute information based on the preprocessed entities.

[0161] 2.1 Entity Alignment

[0162] The purpose of entity alignment is to merge the same entity from different sources or with different representations. Let the extracted entity set be E = {e1, e2, ..., e...}. m This application uses a similarity-based method for entity alignment. Entity e is defined. i and e j The similarity function between them is sim(e i ,e j For example, using cosine similarity: in and They are entity e i and e j The eigenvectors of sim(e). i ,e j If e is greater than a certain threshold τ, then e is considered to be... i and e j They are the same entity; merge them into one entity e. ij .

[0163] 2.2 Entity Disambiguation

[0164] Entity disambiguation resolves the ambiguity of entity names. Suppose there exists an entity name n, corresponding to multiple distinct entities e. n1 ,e n2 ,…,e nkThis application uses contextual information and a knowledge base for entity disambiguation. A contextual similarity function, sim, is defined. ctx (e ni C), where C is the context in which entity name n appears. Simultaneously, each entity e is retrieved from the knowledge base. ni The relevant information is used to calculate its matching degree with the context (match(e)). ni C). The final disambiguation score (e ni )=αsim ctx (e ni ,C)+(1-α)match(e ni ,C), where α is the weighting coefficient. Select the entity with the highest score as the correct entity.

[0165] 2.3 Update semantic relationships and attribute information

[0166] Based on the preprocessed entities, update the semantic relationships between entities and the attribute information of the entities. If two entities are merged into one entity, then merge the relationships and attribute information between them. If an entity is disambiguated into another entity, then update all related relationships and attribute information.

[0167] 3. Entities, semantic relationships between entities, and entity attribute information are loaded into a graph data structure and then into the constructed initial-state knowledge graph to form a dynamic attribute knowledge graph of the target ecological assets.

[0168] 3.1 Graph Data Structure Representation

[0169] Entities are represented as nodes in a graph, semantic relationships between entities are represented as edges, and attribute information of entities is stored in the attributes of nodes and edges. Let the set of nodes be V, the set of edges be E, and each node v∈V have a set of attributes A. v Each edge e∈E has a set of attributes A e The graph data structure is represented by an adjacency matrix A, where A... ij This indicates whether an edge exists between node i and node j. If an edge exists, then A... ij =1, and store the edge's attribute information in the corresponding location.

[0170] 3.2 Loading into the initial state knowledge graph

[0171] Let the initial knowledge graph be G0 = (V0, E0, A0). New nodes, edges, and attribute information are loaded into the initial knowledge graph to form the dynamic attribute knowledge graph G = (V, E, A) of the target ecological asset. For a new node v ∈ V - V0, it is added to the node set V, and its attribute information is added to A. For a new edge e ∈ E - E0, it is added to the edge set E, and its attribute information is added to A.

[0172] Therefore, the above-mentioned approach, based on the specific scenario of dynamic monitoring and assessment of ecological assets, offers the following technical advantages compared to traditional methods:

[0173] 1. Traditional methods often employ rule-based entity extraction techniques, requiring the manual creation of numerous rules, which struggles to adapt to the complex and ever-changing language expressions and data types in the ecological field. For example, in forest ecological monitoring, rules cannot accurately identify entities for newly emerging species names or descriptions of ecological phenomena. Traditional relation extraction relies on handcrafted features and shallow machine learning models, such as Naive Bayes classifiers. These models struggle to capture the unique semantic information and contextual relationships in ecological data, resulting in low accuracy and recall in relation extraction. For instance, traditional methods are prone to misjudgment or omission when describing symbiotic or predatory relationships among species in an ecosystem. Traditional attribute extraction methods typically involve simple keyword matching, ignoring the association between attribute values ​​and entities and context. In ecological scenarios, the same attribute for different entities can have different value ranges and meanings, making accurate extraction and differentiation difficult with traditional methods. This application utilizes a Conditional Random Field (CRF) model, automatically learning contextual information in sequence data through feature functions, eliminating the need for manually writing numerous rules. In ecological monitoring, it can better adapt to dynamic attribute feature sequences from different sources and in different formats, accurately identifying various ecological entities, such as rare species and ecological regions. In this application, a deep learning-based convolutional neural network (CNN) model can automatically extract features from entity pairs and capture complex semantic information in the data. In ecosystems, it can more accurately identify various relationships between species and between species and their environment, such as "trees provide habitat for birds" and "soil nutrients supply plants," providing more comprehensive information for ecological research and conservation. This application combines rule-based and machine learning methods, utilizing the deterministic nature of rules to handle common attribute extraction while leveraging machine learning models to handle complex situations. In ecological asset monitoring, it accurately extracts various attribute information of entities, such as species growth cycles and environmental indicators of ecological areas, improving the accuracy and flexibility of attribute extraction.

[0174] 2. Traditional entity alignment methods primarily rely on string matching, considering only the similarity of entity names while neglecting semantic and contextual information. In the ecological domain, the same entity may have multiple names or representations, making misalignment prone to occur with traditional methods. Traditional entity disambiguation methods depend on manually constructed knowledge bases and rules, making it difficult to handle large-scale, dynamically changing ecological data. For entities with the same name but different meanings in the ecological domain, such as species with the same name in different regions, traditional methods cannot accurately distinguish them. In this application, the entity alignment method based on a similarity function (such as cosine similarity) considers the entity's feature vector, enabling a more accurate measurement of semantic similarity between entities. In ecological monitoring, it accurately merges the same ecological entity from different sources and with different representations, avoiding data redundancy and errors, and improving the quality of the knowledge graph. The disambiguation method combining contextual similarity and knowledge base matching can fully utilize the contextual information of entity appearance and relevant knowledge in the knowledge base. In ecological scenarios, it accurately resolves the ambiguity of entity names, ensuring the uniqueness and accuracy of entities in the knowledge graph, providing a reliable foundation for subsequent analysis and decision-making.

[0175] 3. Traditional graph data structure representation methods are relatively simple, typically focusing only on the connections between nodes and edges, ignoring the attribute information of nodes and edges. In ecological asset monitoring, ecological entities and relationships have rich attributes, which traditional methods cannot fully store and represent. Traditional knowledge graph loading methods are static and difficult to adapt to the dynamic changes in ecological data. In ecosystems, entities, relationships, and attributes change with time and the environment, and traditional methods cannot update the knowledge graph in a timely manner.

[0176] This application comprehensively stores entity, relation, and attribute information in a graph data structure, using adjacency matrices and attribute sets to represent the graph's structure and attributes. In ecological monitoring, it fully records various attributes of ecological entities (such as species biological characteristics and geographical information of ecological regions) and attributes of relationships between entities (such as relationship strength and duration), providing richer information for ecological research. This application dynamically loads new nodes, edges, and attribute information into the initial state knowledge graph, enabling timely reflection of ecosystem changes. In dynamic monitoring of ecological assets, it updates the knowledge graph in real time, providing the latest and most accurate information for ecological decision-making, improving the efficiency and scientific rigor of ecological management.

[0177] Optionally, entity extraction, relation extraction, and attribute extraction are performed on the dynamic attribute feature sequence to obtain the entities of the ecological assets, the semantic relationships between entities, and the attribute information of the entities, including:

[0178] The dynamic attribute feature sequence is converted into a superposition state representation sequence;

[0179] The superposition state representation sequence is decomposed at multiple scales to obtain a set of quantum features containing multi-scale detailed features;

[0180] Each feature element in the quantum feature set is abstracted into a molecular unit;

[0181] Semantic feature aggregation is performed between molecular units to generate entity candidate clusters;

[0182] Spatiotemporal evolution pattern recognition is performed on the candidate entity clusters to generate relationship pre-mining results;

[0183] Based on the connectivity and path constraints of the constructed topology graph, targeted entity extraction, relation extraction, and attribute extraction are performed on the entity candidate clusters and the relation pre-mining results to obtain the entities of the ecological assets, the semantic relationships between entities, and the attribute information of the entities, respectively.

[0184] Preferably, in a scenario, the above-mentioned technical implementation is described in an alternative or preferred manner.

[0185] 1. Transform the dynamic attribute feature sequence into a superposition state representation sequence.

[0186] Let the dynamic attribute feature sequence be S = {s1, s2, ..., s}. n}, where s i =(s i1 ,s i2 ,…,s im ) is an m-dimensional vector representing the values ​​of the i-th dynamic attribute feature across different dimensions. In ecological asset monitoring scenarios, if monitoring a forest ecosystem, these dimensions correspond to tree height, diameter at breast height (DBH), crown width, and indicators such as soil moisture and light intensity in the area. This application uses a quantum entanglement-based mapping function to transform it into a superposition state representation sequence Q = {q1, q2, ..., q...} n For each s i The mapping process is as follows: Where: |ψ jk > represents the quantum ground state, and 2^m represents the total number of quantum state combinations, because each dimension of a feature has multiple superpositions of states in quantum representation. α ijk It is the probability amplitude corresponding to the quantum ground state, satisfying The probability amplitude is calculated using a special neural network. To achieve this, the network uses s i The input is α, and the output is α. ijk : The neural network here It is a deep convolutional neural network (DCNN) whose structure includes multiple convolutional layers, pooling layers and fully connected layers. It is trained with a large amount of ecological data to learn the mapping relationship between features and probability amplitudes.

[0187] 2. Perform multi-scale decomposition on the superposition state representation sequence to obtain a set of quantum features containing multi-scale detailed features.

[0188] Employing a multi-scale decomposition operator based on quantum wavelet transform The superposition state representation sequence Q is processed. The multi-scale decomposition process is represented as a recursive form: Q (0) =Q, where l represents the scale level. Specific multi-scale decomposition operators. The definition is as follows: Among them: U pr This is a set of unitary transformation matrices, the parameters of which are optimized using a quantum genetic algorithm to adapt to the characteristics of ecological data. Unitary transformation matrix U pr Used to rotate and transform quantum states, thereby extracting feature information at different scales. pr These are the corresponding weighting coefficients, satisfying... The weight coefficients are determined by an adaptive weight adjustment algorithm, which dynamically adjusts the weights based on the importance of the features at different scales.

[0189] 3. Abstract each feature element in the quantum feature set into a molecular unit.

[0190] Define an abstract function based on quantum state projection and feature fusion. The quantum feature set Q (l) Each feature element in Abstracted into molecular units Among them: Π s It is a quantum state projection operator used to project quantum states onto a specific subspace to extract feature information from different aspects. s These are predefined feature vectors, representing different semantic feature dimensions. ⊙ represents element-wise multiplication. β s It is a fusion weight, which is learned through an attention-based neural network that dynamically assigns weights based on the importance of features.

[0191] 4. Perform semantic feature aggregation on molecular units to generate entity candidate clusters.

[0192] Define a semantic similarity function based on quantum entanglement similarity and graph neural networks.

[0193] Where: h iand h j It is a molecular unit and The feature vector is encoded using a Graph Neural Network (GNN). The input to the GNN is a graph of molecular units, where nodes represent molecular units and edges represent the initial relationships between them. W is a learnable weight matrix. σ is an activation function, such as the sigmoid function. A hybrid clustering algorithm based on density and hierarchical clustering is used to cluster the molecular units, generating candidate clusters C = {C1, C2, ..., C}. p This algorithm combines the density clustering concept of the DBSCAN algorithm with the hierarchical structure construction capability of hierarchical clustering. The specific steps are as follows: First, the DBSCAN algorithm is used to perform preliminary clustering of molecular units based on semantic similarity (sim), obtaining a set of density-connected core points and boundary points. Then, the hierarchical clustering algorithm is used to further merge and divide these sets, forming the final entity candidate clusters.

[0194] 5. Perform spatiotemporal evolution pattern recognition on entity candidate clusters to generate relation pre-mining results.

[0195] Let C_t represent the set of entity candidate clusters at time t. Define a spatiotemporal evolution pattern function f(C_{t1}, C_t) based on a Spatiotemporal Dynamic Graph Convolutional Network (STGCN): h t =ST-GCN(C t-1 C t ), where h t This represents the features at time t. The spatiotemporal dynamic graph convolutional network combines the ideas of graph convolutional networks and temporal convolutional networks, and can simultaneously handle the spatial relationships and temporal evolution between entity candidate clusters.

[0196] Through a classifier For h t The results of the relationship pre-mining are obtained by classification. pre : Classifier It is a fully connected neural network whose output is a probability distribution of different relation types.

[0197] 6. Based on the connectivity and path constraints of the constructed topology graph, targeted entity extraction, relation extraction, and attribute extraction are performed on the entity candidate clusters and relation pre-mining results.

[0198] Construct a topology graph G = (V, E), where V is the set of nodes, corresponding to candidate clusters of entities; and E is the set of edges, corresponding to the pre-mined results of relationships.

[0199] Define the connectivity function conn(u,v) as:

[0200] The path constraint function path(u,v) is implemented using a path search algorithm based on reinforcement learning. This algorithm uses a topological graph G as the environment, where the agent searches for a path from node u to node v, satisfying certain constraints such as path length and edge weights.

[0201] In the topological graph G, targeted entity extraction, relation extraction, and attribute extraction are performed based on connectivity and path constraints. Targeted entity extraction is achieved through a selector based on a graph attention mechanism. To achieve: in This is the extracted set of entities. Relation extraction determines the semantic relationships between entities by analyzing the attributes of edges that satisfy connectivity and path constraints. Attribute extraction obtains entity attribute information by combining additional information from nodes and edges through a reasoning model based on knowledge graph embedding. This reasoning model maps information from the topological graph to the knowledge graph and uses the semantic information of the knowledge graph for attribute reasoning.

[0202] Therefore, the aforementioned alternative or preferred technologies, centered on the integration of quantum computing and deep learning, achieve in-depth mining of ecological asset information through quantum state transformation, multi-scale decomposition, semantic aggregation, and intelligent extraction based on topological graphs of ecological asset data. Essentially, they utilize the superposition and entanglement characteristics of quantum computing, combined with the adaptive learning capabilities of deep learning, and compared to traditional technologies, offer the following technical advantages:

[0203] 1. Traditional methods typically store and process ecological data in conventional forms such as numerical values ​​and text, for example, simply recording data such as the height and number of trees in a forest and storing it in tabular form. This approach ignores the potentially complex relationships between data, fails to effectively capture the dynamic changes and inherent laws of the ecosystem, and has a single data representation format, making it difficult to reflect the multidimensional characteristics of ecological data. This application utilizes a quantum state mapping function to transform the dynamic attribute feature sequence into a superposition state representation sequence, with a probability amplitude α. ijk Learned by deep convolutional neural networks, each data point exists in a quantum superposition state, reflecting the diversity of ecological states. This representation can not only describe multiple dimensions of ecological characteristics simultaneously, but also utilize the properties of quantum entanglement to reflect the potential connections between different ecological factors. For example, in forest ecological monitoring, it can simultaneously represent the complex relationship between tree growth and factors such as soil moisture and light intensity, providing a more comprehensive and accurate characterization of the ecosystem's state compared to traditional methods.

[0204] 2. Traditional feature extraction methods often employ fixed-window statistical approaches or simple filtering algorithms, such as calculating the average and standard deviation of forest temperature over a period of time. These methods can only extract surface features at a single scale and cannot adapt to changes in the ecosystem at different spatiotemporal scales, making them ineffective for analyzing specific ecological processes.

[0205] This application employs a multi-scale decomposition operator based on quantum wavelet transform. like Where the unitary transformation matrix U pr Through optimization using a quantum genetic algorithm, the weight coefficient w pr Adaptive adjustment. This enables the system to decompose ecological data at multiple scales, extracting detailed features at different spatiotemporal resolutions. For example, when monitoring river ecosystems, it can capture both the overall flow change trend of the river (large-scale features) and analyze the details of water quality fluctuations in local river sections (small-scale features). Compared with traditional methods, it can provide a deeper understanding of the evolution mechanism of the ecosystem and offer richer information for ecological protection and management.

[0206] 3. Traditional semantic aggregation and clustering methods rely on manually set rules or simple similarity measures, such as Euclidean distance-based clustering algorithms. In ecological applications, due to the ambiguity of ecological concepts and the complexity of data, these methods are prone to problems such as inaccurate clustering and semantic comprehension bias, failing to effectively identify ecological entities and the relationships between them.

[0207] This application defines a semantic similarity function based on quantum entanglement similarity and graph neural networks. A hybrid clustering algorithm is employed to generate entity candidate clusters. Quantum entanglement similarity can capture deep semantic connections between data, while graph neural networks can learn the contextual information of molecular units within a graph structure. For example, in marine ecological monitoring, for specific biological communities and environmental factors, this new technology more accurately aggregates molecular units with similar ecological functions or interrelationships into entity candidate clusters, thereby more precisely identifying different biological populations and ecological regions in the marine ecosystem. Compared to traditional clustering methods, the clustering results better reflect the actual ecological situation and improve the semantic understanding capability of ecological data.

[0208] 4. Traditional relationship mining and information extraction mainly rely on manually constructed knowledge bases and rule templates, such as extracting species relationships from ecological literature using preset keywords and grammatical rules. This approach is inefficient, difficult to adapt to massive dynamic data, and struggles to accurately identify newly emerging ecological relationships and complex semantics.

[0209] This application utilizes a Spatiotemporal Dynamic Graph Convolutional Network (STGCN) to identify the spatiotemporal evolution patterns of entity candidate clusters, combined with a reinforcement learning-based path search algorithm for targeted entity extraction, relation extraction, and attribute extraction. For example, in the assessment of urban green space ecological assets, it can automatically uncover the ecological connections between different green spaces over time (such as species migration, complementary ecological functions, etc.), as well as detailed attribute information for each green space (area, vegetation type, ecosystem service value, etc.). Compared to traditional methods, this new technology achieves automated and intelligent information extraction, not only improving the efficiency of information processing but also discovering complex ecological relationships that are difficult to detect using traditional methods, providing more comprehensive and accurate data support for the comprehensive assessment of ecological assets.

[0210] Optionally, entities, semantic relationships between entities, and entity attribute information are loaded into a graph data structure and then into the constructed initial-state knowledge graph to form a dynamic attribute knowledge graph of the target ecological asset, including:

[0211] Using graph attention networks, we can infer the semantic relationships between entities, the attribute information of entities, the ecological dependency semantic relationships between entities, and the material transport semantic relationships.

[0212] The relationship strength of the ecological dependency semantic relationship and the material transport semantic relationship is quantified and encapsulated into weighted semantic relationship triples;

[0213] Based on the semantic relation triples, construct an R-tree hybrid index;

[0214] The R-tree hybrid index is loaded into the constructed initial state knowledge graph to form a dynamic attribute knowledge graph of the target ecological assets.

[0215] Preferably, in a scenario, the above-mentioned technical implementation is described in an alternative or preferred manner.

[0216] 1. Utilize graph attention networks to infer semantic relationships between entities and their attributes, and mine ecological dependency semantic relationships and material transport semantic relationships between entities.

[0217] Let the graph data structure be G=(V,E), where V={v1,v2,…,v n} is the set of entity nodes, E = {(v i ,v j )} is a set of edges, representing existing semantic relationships between entities. Each node v i There is a corresponding attribute vector x i It contains the attribute information of the entity. For example, in a forest ecosystem, for the entity "tree", the attribute vector contains information such as tree height, diameter at breast height, and tree species.

[0218] The propagation process of a single layer in a graph attention network (GAT) is represented as follows: in: It is the feature representation of node i at layer l. W (l) It is the learnable weight matrix of the l-th layer, used to perform linear transformation on the input features. It is the set of neighboring nodes of node i. The attention coefficient represents the importance of node j to node i, and is calculated using the following formula:

[0219] Where a (l) It is a learnable attention vector, || represents the vector concatenation operation, and LeakyReLU is the activation function.

[0220] To uncover the semantic relationships of ecological dependence and material transport between entities, this application introduces an additional relational reasoning module. Let r ij The feature vector representing the relationship between node i and node j is calculated using the following formula: Where MLP is a multilayer perceptron, and L is the total number of layers in the graph attention network. By analyzing r... ij By classifying, this application determines whether there is an ecological dependency semantic relationship or a material transport semantic relationship between node i and node j.

[0221] 2. Quantify the relationship strength of the ecological dependency semantic relationship and the material transport semantic relationship, and encapsulate them into weighted semantic relationship triples.

[0222] Let R eco R represents the set of semantic relations of ecological dependence. mat This represents the set of semantic relationships for material transport. For each pair of nodes (v...) with ecological dependence or material transport relationships... i ,v j In this application, a relation strength function s is defined to quantify the strength of a relation.

[0223] For ecological dependency semantic relations, the relation strength function is defined as: Where f k It is the k-th feature function, used to measure the degree of interdependence between node i and node j in a specific ecological aspect. For example, in a forest ecosystem, f1 represents the degree of interdependence between "trees" and "birds" in terms of habitat provision, w k These are the corresponding weight coefficients, learned from the training data.

[0224] For semantic relations related to matter transport, the relation strength function is defined as: MLP′ is another multilayer perceptron, and w is a learnable weight vector.

[0225] Encapsulate entities, relations, and relation strengths into weighted semantic relation triples (v i ,r,v j ,s), where r∈R eco ∪R mat s represents the strength of the corresponding relationship.

[0226] 3. Based on the semantic relation triples, construct an R-tree hybrid index.

[0227] Let the set of semantic relation triples be T = {(v i ,r,v j This application maps each triple to a multidimensional space, where each dimension corresponds to an entity attribute or relation feature. For example, in a forest ecosystem, dimensions include the entity's geographical location, species type, ecological function, etc.

[0228] The process of building an R-tree hybrid index involves the following steps:

[0229] 3.1 Data Preprocessing

[0230] Each triplet (v i ,r,v j ,s) is transformed into a multidimensional vector t, which contains entity attributes and relational feature information.

[0231] 3.2 Node Partitioning

[0232] Using the R-tree partitioning algorithm, a multidimensional vector t is partitioned into different nodes. Let node N contain the set of vectors {t1, t2, ..., t...}. m The minimum bounding rectangle (MBR) of a node is calculated using the following formula:

[0233] MBR N =[min 1≤i≤m t i1 ,max 1≤i≤m t i1 ]×[min 1≤i≤m t i2 ,max 1≤i≤m t i2 ]×…×[min 1≤i≤m t id ,max 1≤i≤m t id ], where d is the dimension of the multidimensional space.

[0234] 3.3 Index Building

[0235] The nodes are recursively divided into child nodes until a set termination condition is met, such as the number of vectors in a node being less than a certain threshold. This ultimately constructs an R-tree hybrid index.

[0236] 4. Load the R-tree hybrid index into the constructed initial-state knowledge graph to form a dynamic attribute knowledge graph of the target ecological assets.

[0237] Let the initial knowledge graph be KG0 = (V0, E0, A0), where V0 is the initial entity set, E0 is the initial relation set, and A0 is the initial attribute set.

[0238] Each triplet (v) in the R-tree hybrid index i ,r,v j ,s) Loaded into the initial state knowledge graph:

[0239] if Then v i Add it to V0, and add its attribute information to A0.

[0240] if Then v j Add it to V0, and add its attribute information to A0.

[0241] if Then (v) i ,r,v j Add it to E0 and use the relation strength s as the relation weight.

[0242] The final result is a dynamic attribute knowledge graph KG = (V, E, A) of the target ecological asset, where V = V0, E = E0, and A = A0.

[0243] To this end, the aforementioned alternative or preferred technologies, with Graph Attention Network (GAT), relation strength quantification model, and R-tree hybrid index as the core, construct a dynamic attribute knowledge graph of ecological assets. GAT, based on an attention mechanism, adaptively learns the association weights between nodes to mine potential semantic relationships between ecological entities; the relation strength quantification model, using multilayer perceptron (MLP) and weighted computation, accurately measures the tightness of relationships such as ecological dependence and material transport; and the R-tree hybrid index maps semantic relation triples to a multidimensional space, achieving efficient knowledge storage and retrieval.

[0244] Compared with traditional technologies, it has the following technological advantages:

[0245] 1. Traditional ecological relationship analysis relies on manually coded rules or simple association rule mining. For example, in forest ecosystem research, a table of correspondences between "tree species and birds" is manually created, or the Apriori algorithm is used to mine species symbiotic relationships based on historical data. These methods require pre-defining relationship types, making it difficult to discover special ecological dependencies (such as allelopathy between plants) and dynamic material transport (such as seasonal nutrient cycling), and they cannot adapt to the dynamic changes in ecosystems.

[0246] This application utilizes GAT and Multilayer Perceptron (MLP) to construct a relational reasoning model, such as By performing multi-layer attention calculations on node features, non-linear relationships between entities can be automatically captured. For example, in wetland ecosystems, complex dependency chains such as "water purification by benthic organisms in reed communities" can be identified. Furthermore, by combining ecological knowledge with the classification of relationship feature vectors, novel ecological relationships that are difficult to detect using traditional methods can be discovered, providing support for systematic research on ecosystems.

[0247] 2. Traditional assessments of relationship strength often employ subjective scoring methods or calculations based on a single indicator. For example, when assessing species competition, the competition intensity coefficient is assigned solely based on the proportion of species, ignoring the influence of ecological factors (such as temperature and humidity) on the relationship. This approach lacks objectivity and comprehensiveness, and cannot accurately reflect the dynamic changes in ecological relationships.

[0248] This application employs a quantization model combining a weighted feature function and a softmax function, such as... The strength of relationships is calculated comprehensively from multiple dimensions (such as mass exchange, energy flow efficiency, and niche overlap). In marine ecosystems, the intensity of mass transport between phytoplankton and zooplankton can be precisely quantified as it fluctuates seasonally, providing accurate data support for ecological carrying capacity assessment and resource management.

[0249] 3. Traditional ecological knowledge graphs often use relational databases or simple graph storage structures, such as using SQL tables to store entity relationship attribute data. This approach has low query efficiency when processing large-scale dynamic data (e.g., retrieving cross-regional ecological chains requires traversing a large number of tables) and is difficult to support complex spatial relationship queries (e.g., finding all ecological nodes with material transport relationships within a watershed).

[0250] This application utilizes a hybrid R-tree index to map semantic relation triples to a multi-dimensional space, constructing a minimum bounding rectangle (MBR), as shown in the example (Tex translation failed). In forest ecological monitoring scenarios, when querying "ecological dependency network at an altitude of 800-1200 meters, affected by a certain pollutant," the R-tree index can quickly locate nodes and relationships that meet the criteria, improving retrieval efficiency by tens of times. Furthermore, this index supports dynamic updates, adapting to real-time changes in ecological data and ensuring the timeliness of the knowledge graph.

[0251] 4. Traditional knowledge graphs have a fixed structure after construction, making it difficult to integrate newly discovered ecological entities or relationships. For example, when a new species invades an ecosystem, the data structure and association rules need to be manually modified, resulting in long update cycles and potential data inconsistencies.

[0252] This application's dynamic attribute knowledge graph continuously learns new data features through GAT, automatically discovering new relationships between entities; the R-tree hybrid index supports incremental updates, and new semantic relationship triples can be inserted into the index structure in real time. In urban green space ecosystems, as new plant species are introduced through greening projects, the system can automatically identify their ecological relationships with local species and quickly update the knowledge graph, providing immediate decision-making support for ecological planning.

[0253] Optionally, based on the ecological asset assessment model, a dynamic asset value assessment is performed on the target ecological asset according to the dynamic attribute knowledge graph to generate dynamic value sequence points and add them to the dynamic value tree, including:

[0254] Based on the graph embedding layer in the ecological asset assessment model, entities and relationships in the dynamic attribute knowledge graph are mapped to a low-dimensional vector space to obtain the semantic association relationships between ecological asset elements.

[0255] Based on the sequential decision layer in the ecological asset assessment model, temporal features are extracted to analyze the semantic relationships between ecological asset elements.

[0256] Based on the asset assessment layer in the ecological asset assessment model, dynamic value sequence points are generated according to the extracted time characteristics and added to the dynamic value tree.

[0257] Preferably, in a scenario, the above-mentioned technical implementation is described in an alternative or preferred manner.

[0258] 1. Based on the graph embedding layer in the ecological asset assessment model, entities and relationships in the dynamic attribute knowledge graph are mapped to a low-dimensional vector space to obtain the semantic relationships between ecological asset elements.

[0259] Let the dynamic attribute knowledge graph be G=(V,E), where V={v1,v2,…,v...} n} represents a set of entity nodes, where each entity node v i Carrying attribute feature vector E = {(v i ,r ij ,v j Let} be the set of relation edges, and r ij Represents entity v i With v j The relationship type between them, and the edges of the relationship carry weights w. ij Indicates the strength of the relationship. Taking a forest ecosystem as an example, entity v... i It is "red pine", attribute feature vector x i Includes information such as tree age, diameter at breast height (DBH), and volume; relation edge (v i ,r ij ,v j This indicates the habitat provision relationship between "red pine" and "squirrel", with a weight w. ij It reflects the degree of dependence.

[0260] The graph embedding layer employs a fusion model of Graph Convolutional Network (GCN) and MultiHeadAttention. For the l-th layer graph convolution operation, node v i The feature update formula is: in: For node v i The hidden feature vector at layer l, initially It is node v i The set of neighboring nodes; c ij It is a normalization constant, usually Used to balance the effects of different node degrees; W (l) b is the learnable weight matrix of the l-th layer. (l) σ is the bias vector; σ is the activation function, such as the ReLU function σ(x) = max(0,x). Based on this, a multi-head attention mechanism is introduced to enhance the ability to capture relational semantics. The k-th attention head computes node v. i With neighbor node v j Attention coefficient

[0261] in: It is the learnable weight matrix of the k-th attention head; a k It is the learnable attention vector of the k-th attention head; || represents the vector concatenation operation; LeakyReLU is a leaky ReLU activation function to avoid the gradient vanishing problem.

[0262] Aggregate the results of the K attention heads to obtain node v. iThe final embedding vector z i :

[0263] Among them W out This is the output weight matrix used for dimensionality reduction. Through the above operations, entities and relations are mapped to a low-dimensional vector space. The geometric relationships such as distance and angle between the low-dimensional vectors of entities and relations can reflect the semantic associations between ecological asset elements. For example, the spatial distance between the embedding vectors of "red pine" and "lark" can reflect the degree of similarity between the two in terms of ecological function.

[0264] 2. Based on the sequential decision layer in the ecological asset assessment model, temporal features are extracted to analyze the semantic relationships between ecological asset elements.

[0265] Let the sequence of embedded vectors of ecological asset elements obtained after the graph embedding layer be Z = {z1, z2, ..., z}. T}, where T is the time step. In the wetland ecosystem monitoring scenario, Z records the semantic association states of various ecological elements (such as reed communities, fish populations, and water quality indicators) within the wetland at different time points. The sequence decision layer adopts an architecture combining gated recurrent units (GRUs) and an attention mechanism. For the t-th time step, the update formula for the GRU unit is:

[0266] r t =σ(W xr z t +W hr h t-1 +b r )

[0267] z′ t =tanh(W xz z t +W hz (r t ☉h t-1 )+b z )

[0268] u t =σ(W xu z t +W hu h t-1 +b u )

[0269] h t =u t ☉h t-1 +(1-u t )☉z′ t

[0270] 1. ResetGate calculation: r t =σ(Wxr z t +W hr h t-1 +b r )

[0271] Reset door r t Used to determine the hidden state h in the previous moment t-1 How much information is retained in the current candidate hidden state calculation? When r t When the value of r approaches 0, it means that most of the hidden state information from the previous time step has been discarded; when r... t When the value approaches 1, it indicates that a large amount of hidden state information from the previous moment has been retained.

[0272] During computation, firstly, the input vector z... t With weight matrix W xr Multiply the inputs to obtain their contribution to the reset gate; simultaneously, reset the previous hidden state h. t-1 With weight matrix W hr Multiply these two components to obtain the contribution of the hidden state from the previous time step to the reset gate. Then, add these two contributions together and add the bias vector b. r Finally, the result is mapped to the interval (0,1) using the Sigmoid activation function σ to obtain the reset gate vector r. t .

[0273] 2. Candidate Hidden State: z t′ =tanh(W xz z t +W hz (r t ⊙h t-1 )+b z )

[0274] Candidate hidden state z t′ It combines the current input z t The previous hidden state information, after being filtered by the reset gate, is used to generate the current hidden state.

[0275] During the calculation, first input vector z t With weight matrix W xz Multiply the inputs to obtain their contribution to the candidate hidden state; then, process the previous hidden state r after the reset gate. t ⊙h t-1 (⊙ represents element-wise multiplication) and the weight matrix W hz Multiply these two contributions to obtain the contribution to the candidate hidden state. Add these two contributions together and then add the bias vector b. zFinally, the result is mapped to the interval (-1, 1) using the hyperbolic tangent activation function tanh to obtain the candidate hidden state vector z. t′ .

[0276] 3. Update Gate: u t =σ(W xu z t +W hu h t-1 +b u )

[0277] Update Gate u t The hidden state h of the previous moment was determined. t-1 and the current candidate hidden state z t′ Hide state h at the current time respectively t The contribution ratio. When u t When the value approaches 1, the hidden state h from the previous time step... t-1 A large amount of information will be retained; when u t When z approaches 0, the current candidate hidden state is... t′ A large amount of information will be retained.

[0278] During computation, similar to the reset gate computation, the input vector z is... t With weight matrix W xu Multiply by the hidden state h from the previous time step. t-1 With weight matrix W hu Multiply by adding the two contributions together and then adding the bias vector b. u Finally, the updated gate vector u is obtained through the Sigmoid activation function σ. t .

[0279] 4. Current Hidden State: h t =u t ⊙h t-1 +(1-u t )⊙z t′

[0280] By updating gate u t The hidden state h from the previous moment t-1 and candidate hidden state z t′ We perform a weighted combination to obtain the hidden state h at the current time. t .

[0281] During computation, the gate vector u will be updated. t Compared to the hidden state h in the previous moment t-1 Perform element-wise multiplication to obtain the retained portion of the hidden state from the previous time step; then multiply 1-u... t With candidate hidden state z t′Element-wise multiplication yields the retained portion of the candidate hidden state. Finally, these two portions are added together to obtain the hidden state vector h at the current time step. t .

[0282] z t : The input vector at time t, in the scenario of dynamic monitoring and evaluation of ecological assets, may be a low-dimensional vector representing the semantic relationship of ecological asset elements after processing by the graph embedding layer, with a dimension of input_size. t-1 : The hidden state vector at time t-1, with dimension hidden_size, records information from previous time steps and is used for calculation at the current time step. h t : The hidden state vector at time t, with dimension hidden_size, is the output of the GRU unit at the current time step and will be used for calculation at the next time step and subsequent temporal feature extraction and analysis. W xr Input z t The weight matrix for the reset gate has dimensions of hidden_size × input_size. hr The hidden state h from the previous moment t-1 The weight matrix for the reset gate has dimensions of hidden_size × hidden_size. r : Reset the gate's bias vector, with dimension hidden_size. W xz Input z t The weight matrix to the candidate hidden state has dimensions of hidden_size × input_size. hz The hidden state r of the previous moment after the door was reset. t ⊙h t-1 The weight matrix to the candidate hidden state has dimensions hidden_size × hidden_size. z : The bias vector of the candidate hidden state, with dimension hidden_size. W xu Input z t The weight matrix of the updated gate has dimensions of hidden_size × input_size. W hu The hidden state h from the previous moment t-1 The weight matrix of the updated gate has dimensions of hidden_size × hidden_size. u : Update the gate's bias vector, with dimension hidden_size. σ: Sigmoid activation function. Maps the input to the (0,1) interval. tanh: hyperbolic tangent activation function. Map the input to the interval (-1, 1).

[0283] In the dynamic monitoring and assessment of ecological assets, GRU units are used to process time-series data on semantic relationships between ecological asset elements. For example, after obtaining a low-dimensional vector representation of the ecological asset elements through a graph embedding layer, this vector representation is used as the input z of the GRU. t As time goes on, the GRU unit continuously updates the hidden state h. t This allows us to capture the changing characteristics of the semantic relationships among ecological asset elements at different time steps. These characteristics can be used in subsequent asset valuation layers to conduct dynamic value assessments of ecological assets based on the extracted time features, providing strong support for the management and decision-making of ecological assets.

[0284] To further capture key temporal features, a temporal attention mechanism is introduced. The attention coefficient β of time step t on other time steps is calculated. tt′ :

[0285] Where Score(h) t ,h t′ The scoring function can be expressed as a dot product. Or additive form w s W hs W ht ,b s These are learnable parameters.

[0286] The final time feature vector f t We obtain the following by weighted summation:

[0287] 3. Based on the asset valuation layer in the ecological asset valuation model, dynamic value sequence points are generated according to the extracted time characteristics and added to the dynamic value tree.

[0288] The asset valuation layer employs a reinforcement learning-based valuation model. Let the state space be the time feature vector f. t The action space is for value assessment decisions (such as value growth forecasting, value risk assessment, etc.). Define the value assessment function V(f) t The model is performed using a Deep Q-Network (DQN) architecture: Q(f t ,a;θ)=w T MLP(f t ,a;θ), where: a is the action; θ is the network parameter; w is the output layer weight vector; MLP(f t ,a;θ) is a multilayer perceptron that takes time feature vectors and actions as inputs and outputs Q-values.

[0289] Update network parameters θ:y using Bellman equations t =r t +γmax a′Q(f t+1 ,a′;θ - ) in:

[0290] r t For immediate rewards, the value of ecological assets can be set based on changes in their actual value during ecological asset assessment; γ is a discount factor used to balance immediate rewards and future rewards; θ - The target network parameters are periodically copied and updated from the current network θ; L(θ) is the loss function, optimized using stochastic gradient descent. The value evaluation result V(f) at each time step is obtained. t Afterwards, it is added to the dynamic value tree as a dynamic value sequence point. The dynamic value tree adopts a quadtree structure, with each node storing value assessment information within a time interval. The node division is based on time granularity and value change magnitude. Let the root node of the dynamic value tree be R, and for a newly generated dynamic value sequence point V(f)... t The corresponding child node is recursively searched based on its timestamp t. If the node capacity is full, a node splitting operation is performed to ensure efficient storage and retrieval of value sequence points.

[0291] To this end, the aforementioned alternative or preferred technologies integrate cutting-edge technologies such as graph neural networks, sequence analysis, and reinforcement learning to construct a dynamic value assessment model for ecological assets. The graph embedding layer utilizes graph convolutional networks and multi-head attention mechanisms to map entities and relationships in the knowledge graph to a low-dimensional vector space. Through node feature updates and attention coefficient calculations, it mines semantic relationships between ecological asset elements. The sequence decision layer employs gated recurrent units and a temporal attention mechanism to extract features from the embedding vectors of time series, capturing the dynamic characteristics of ecological assets changing over time. The asset assessment layer, based on a deep Q-network using reinforcement learning, treats temporal features as states and achieves dynamic value assessment of ecological assets through iterative optimization using a value assessment function and the Bellman equation. The value is then stored in a dynamic value tree structured as a quadtree. Compared with traditional technologies, it has the following technical advantages in data processing, feature extraction, and value assessment:

[0292] 1. Traditional methods typically identify ecological asset relationships based on manually defined rules or simple statistical analysis. For example, in forest ecological assessments, expert experience is used to develop "species symbiotic relationship tables," or association rule mining algorithms are used to analyze species frequency to determine relationships. This approach relies on prior knowledge, making it difficult to discover complex and implicit ecological relationships, and it cannot adapt to the dynamic changes in ecosystems.

[0293] This application employs graph convolutional networks and multi-head attention mechanisms to automatically learn complex relationships between nodes. In forest ecosystems, it can uncover indirect connections such as the "Korean pine soil microbial nutrient cycle" and the potential similarities in ecological functions among different tree species (e.g., the embedding vector distance between "Korean pine" and "larch" reflects functional similarity). Compared to traditional methods, this new technology can more comprehensively and deeply analyze the semantic relationships between ecological asset elements, providing rich data for systematic research on ecosystems.

[0294] 2. Traditional time series analysis often employs methods such as moving averages and autoregressive models (ARIMA) to perform simple trend fitting and prediction of ecological data. For example, in wetland ecological monitoring, predicting future water quality changes solely based on the average value of historical water quality data fails to account for the interactive effects of multiple factors and complex dynamic processes within the ecosystem.

[0295] This application utilizes gated recurrent units and a temporal attention mechanism to adaptively learn the long-term and short-term dependencies of ecological asset time series. In wetland ecosystems, it can simultaneously capture the changing characteristics of multiple factors at different time scales, such as reed community growth, fish population migration, and water quality index fluctuations, and focus on key time points (such as the impact of flooding on wetland ecology) through attention coefficients. Compared with traditional methods, this new technology can more accurately characterize the dynamic evolution process of ecosystems and improve the ability to predict and respond to ecological changes.

[0296] 3. Traditional ecological asset assessment relies on static assessment indicator systems and fixed weight allocations. For example, it uses expert scoring to determine the carbon sink value and biodiversity value of forest ecosystems, and then calculates the total value through weighted summation. This method lacks consideration for the dynamic changes in ecological assets and cannot reflect the fluctuations in ecosystem value in real time.

[0297] This application utilizes a deep Q-network based on reinforcement learning, treating the temporal characteristics of ecological assets as states and optimizing the value assessment strategy through continuous trial and error and reward feedback. In practical applications, the value assessment results can be dynamically adjusted according to the real-time state of ecological assets (such as ecological damage after forest fires or ecological improvement after wetland restoration projects). Furthermore, a dynamic value tree with a quadtree structure is used for storage, supporting efficient value sequence point querying and updating, enabling rapid response to ecosystem changes and providing timely and accurate value references for dynamic management and decision-making of ecological resources.

[0298] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A method for dynamic monitoring and assessment of ecological assets, characterized in that, include: The acquired attribute feature carriers are subjected to quantum state fusion to generate a quantum state hybrid dataset; The quantum state hybrid dataset is mapped as dynamic spatiotemporal graph structure data; The dynamic spatiotemporal graph structure data is cross-domain mapped, and cross-domain semantic information is given to the dynamic spatiotemporal graph structure data based on graph attention to generate a core feature subset with evaluation value. Input a subset of core features into the encoder to learn the probability distribution representation of the features and generate a latent feature vector; The generator generates enhanced feature samples based on the latent feature vectors, while the discriminator distinguishes between the real core features and the generated feature samples. The variational autoencoder reconstructs and optimizes the real core features and generated feature samples to generate the target ecological asset attribute features; The model for dynamic monitoring of ecological assets is used to dynamically collect the attribute characteristics of target ecological assets and form a dynamic attribute characteristic sequence accordingly. Construct a dynamic attribute knowledge graph based on dynamic attribute feature sequences; Based on the ecological asset assessment model, the target ecological asset is dynamically valued according to the dynamic attribute knowledge graph, so as to generate dynamic value sequence points and add them to the dynamic value tree.

2. The method for dynamic monitoring and assessment of ecological assets according to claim 1, characterized in that, Based on dynamic attribute feature sequences, a dynamic attribute knowledge graph is constructed, including: Entity extraction, relation extraction, and attribute extraction are performed on dynamic attribute feature sequences to obtain the entities of ecological assets, the semantic relationships between entities, and the attribute information of entities, respectively. The extracted entities are preprocessed, and the semantic relationships and attribute information are updated based on the preprocessed entities. The preprocessing includes alignment and disambiguation. The entities, their semantic relationships, and their attribute information are loaded into a graph data structure and then into the constructed initial-state knowledge graph to form a dynamic attribute knowledge graph of the target ecological assets.

3. The method for dynamic monitoring and assessment of ecological assets according to claim 2, characterized in that, Entity extraction, relation extraction, and attribute extraction are performed on the dynamic attribute feature sequence to obtain the entities of ecological assets, the semantic relationships between entities, and the attribute information of entities, including: The dynamic attribute feature sequence is converted into a superposition state representation sequence; The superposition state representation sequence is decomposed at multiple scales to obtain a set of quantum features containing multi-scale detailed features; Each feature element in the quantum feature set is abstracted into a molecular unit; Semantic feature aggregation is performed between molecular units to generate entity candidate clusters; Spatiotemporal evolution pattern recognition is performed on the candidate entity clusters to generate relationship pre-mining results; Based on the connectivity and path constraints of the constructed topology graph, targeted entity extraction, relation extraction, and attribute extraction are performed on the entity candidate clusters and the relation pre-mining results to obtain the entities of the ecological assets, the semantic relationships between entities, and the attribute information of the entities, respectively.

4. The method for dynamic monitoring and assessment of ecological assets according to claim 2, characterized in that, The entities, their semantic relationships, and attribute information are loaded into a graph data structure and then into the constructed initial-state knowledge graph to form a dynamic attribute knowledge graph of the target ecological assets, including: Using graph attention networks, we can infer the semantic relationships between entities, the attribute information of entities, the ecological dependency semantic relationships between entities, and the material transport semantic relationships. The relationship strength of the ecological dependency semantic relationship and the material transport semantic relationship is quantified and encapsulated into weighted semantic relationship triples; Based on the semantic relation triples, construct an R-tree hybrid index; The R-tree hybrid index is loaded into the constructed initial state knowledge graph to form a dynamic attribute knowledge graph of the target ecological assets.

5. The method for dynamic monitoring and assessment of ecological assets according to claim 1, characterized in that, Based on the ecological asset assessment model, dynamic asset value assessment of target ecological assets is performed according to a dynamic attribute knowledge graph to generate dynamic value sequence points and add them to the dynamic value tree, including: Based on the graph embedding layer in the ecological asset assessment model, entities and relationships in the dynamic attribute knowledge graph are mapped to a low-dimensional vector space to obtain the semantic association relationships between ecological asset elements. Based on the sequential decision layer in the ecological asset assessment model, temporal features are extracted to analyze the semantic relationships between ecological asset elements. Based on the asset assessment layer in the ecological asset assessment model, dynamic value sequence points are generated according to the extracted time characteristics and added to the dynamic value tree.

Images