An intelligent water body ecological monitoring system based on multi-source data fusion

Abstract

The application discloses a water body ecological intelligent monitoring system based on multi-source data fusion, and relates to the technical field of water body ecological monitoring.Different from a traditional clustering anomaly detection scheme, a directed graph topology of an ecological chain is constructed, a cause-effect coupling relationship is quantitatively corrected through a transfer entropy, anomaly reverse tracing and positive cascade early warning are realized based on a graph network, and function upgrading from state alarm to cause-effect diagnosis is realized.The application has the following advantages: in view of the core defects of the existing water body ecological monitoring system, i.e., regarding multi-dimensional ecological elements as independent features for clustering analysis, losing cause-effect transmission information of the ecological chain, and being able to determine only an abnormal state of a monitoring point but unable to locate an abnormal cause, various ecological monitoring elements are constructed into a structured directed topology according to material and energy flow relationships, and coupling correlations among the ecological elements are established through quantitative determination of directional cause-effect strength, and when an ecological anomaly is detected, an upstream driving link is traced in a reverse direction along a cause-effect chain, and an ecological source triggering the anomaly is accurately located.

Description

Technical Field

[0001] This invention relates to the field of aquatic ecological monitoring technology, specifically to an intelligent aquatic ecological monitoring system based on multi-source data fusion. Background Technology

[0002] Water body ecological monitoring is the core support for water environment governance and ecological restoration. Traditional manual sampling and monitoring has shortcomings such as long cycle, poor continuity and limited coverage, which makes it difficult to meet the current needs of precise and intelligent ecological management.

[0003] Existing IoT-based water ecological monitoring systems, such as the patent solution with publication number CN109962982A entitled "An IoT-based River and Lake Water Ecological Environment Monitoring System", collect multi-source data such as hydrology, water quality, bottom sediment, and aquatic organisms through multiple types of sensors, and use an adaptive clustering method to identify abnormal monitoring points, thereby realizing automated monitoring and early warning of ecological status.

[0004] However, this solution still has significant limitations: First, it treats various ecological elements as independent feature dimensions for cluster analysis, failing to establish a causal coupling relationship within the ecological chain. It can only determine the overall abnormality of monitoring points, but cannot pinpoint the source of the abnormality, making it difficult to support targeted governance. Second, it can only trigger post-event alarms after the abnormality has fully manifested, unable to predict the spread trend and impact range of the abnormality along the ecological chain, lacking early warning capabilities and hindering proactive risk prevention and control. Third, it uses a general statistical analysis model, which cannot adaptively adjust according to the ecological characteristics of different water bodies. When applied across water bodies, it is prone to diagnostic biases due to differences in the ecological chain structure, resulting in insufficient scenario adaptability.

[0005] In summary, existing technologies have not yet broken through the functional limitations of "only alarming, not diagnosing," and cannot simultaneously address anomaly tracing, cascading early warning, and multi-scenario adaptive capabilities. There is an urgent need to develop a more diagnostically in-depth and adaptable intelligent water body ecological monitoring system. To this end, we propose an intelligent water body ecological monitoring system based on multi-source data fusion. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent water body ecological monitoring system based on multi-source data fusion.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an intelligent water body ecological monitoring system based on multi-source data fusion, comprising a multi-source ecological data acquisition subsystem, a wireless data transmission module and a monitoring and early warning terminal module, and further comprising a time series data preprocessing module, an ecological chain directed graph construction module, a cross-dimensional transfer entropy causal inference module, and an ecological chain anomaly tracing and cascading early warning module; The output of the multi-source ecological data acquisition subsystem is connected to the input of the wireless data transmission module, and the output of the wireless data transmission module is connected to the input of the time-series data preprocessing module; the output of the ecological chain directed graph construction module is connected to the first input of the cross-dimensional transfer entropy causal inference module, and the output of the time-series data preprocessing module is connected to the second input of the cross-dimensional transfer entropy causal inference module; the output of the cross-dimensional transfer entropy causal inference module is connected to the input of the ecological chain anomaly tracing and cascading early warning module, and the output of the ecological chain anomaly tracing and cascading early warning module is connected to the input of the monitoring and early warning terminal module. The multi-source ecological data acquisition subsystem is used to collect real-time time-series data of seven types of ecological monitoring elements in the water body, including hydrological and meteorological data, water quality, sediment quality, submerged plants, phytoplankton, zooplankton, and algae. The wireless data transmission module is used to wirelessly upload the raw time-series data collected by the multi-source ecological data acquisition subsystem to the cloud processing unit. The directed graph construction module of the ecological chain is used to map various ecological monitoring elements into directed graph nodes based on prior knowledge of aquatic ecology, define the directed ecological relationship edges between nodes according to the direction of material and energy flow in the ecosystem, construct the initial directed graph topology of the ecological chain, and output it to the cross-dimensional entropy causal inference module. The time series data preprocessing module is used to receive raw time series data from the cloud, sequentially perform outlier removal, missing value completion and standardization processing, and output standardized time series data to the cross-dimensional entropy causal inference module. The cross-dimensional transfer entropy causal inference module is used to receive the initial directed graph topology and standardized time series data of the ecosystem chain, quantify the directional causal coupling strength between the nodes corresponding to each directed edge, dynamically correct the topological structure and edge weight of the directed graph, generate a localized ecosystem chain causal network and output it to the ecosystem chain anomaly tracing and cascading early warning module. The ecosystem anomaly tracing and cascading early warning module is used to perform node state anomaly detection based on the localized ecosystem causal network. When an anomaly is detected in any node, it performs reverse tracing analysis and forward propagation prediction along the directed graph, and outputs an early warning result including the anomaly source, propagation path, affected nodes and the time window of impact. The monitoring and early warning terminal module is used to receive and visualize the results of anomaly tracing and cascading early warning information.

[0008] As a further embodiment of the present invention: the multi-source ecological data acquisition subsystem includes a hydro-meteorological sensor unit, a water quality sensor unit, a sediment monitoring unit, and an aquatic organism sampling unit; the hydro-meteorological sensor unit is used to collect data on three types of hydro-meteorological elements: flow rate, water level, and water temperature; the water quality sensor unit is used to collect data on four types of water quality elements: dissolved oxygen, total phosphorus, ammonia nitrogen, and permanganate index; the sediment monitoring unit is used to collect data on five types of sediment elements: oxidation-reduction potential, pH value, total nitrogen, total phosphorus, and organic matter; the aquatic organism sampling unit is used to collect data on four types of aquatic organism elements: submerged plant biomass, phytoplankton community and residue, zooplankton classification and biomass, and algae classification and biomass.

[0009] As a further aspect of the present invention: the wireless data transmission module adopts a two-level transmission architecture of near-end networking and far-end uploading; the data acquisition devices at each monitoring point near the end complete data aggregation through the ZigBee wireless network, and the aggregated data is remotely uploaded to the cloud processing unit through the 4G / 5G cellular network.

[0010] As a further aspect of the present invention: in the initial directed graph of the ecological chain constructed by the ecological chain directed graph construction module, the nodes are divided into five categories according to ecological level: environment-driven layer nodes, sediment nutrient layer nodes, primary production layer nodes, primary consumption layer nodes, and water quality feedback layer nodes; the directed edges between nodes are divided into four categories according to ecological relationship type: nutrient release edge, feeding relationship edge, competition relationship edge, and decomposition feedback edge, and the direction of each directed edge is consistent with the direction of material and energy transfer in the ecosystem; each node carries the time-series monitoring data of the corresponding ecological element as node attribute, and each directed edge carries the corresponding ecological relationship type as edge attribute.

[0011] As a further aspect of the present invention: the time series data preprocessing module performs the following steps: First, outliers in the original time series data are removed using the 3σ criterion; Second, missing values ​​in the data are filled in using linear interpolation to form a continuous time series sequence; Third, the time series data of each element are standardized using the Z-score standardization method to unify the dimensions and output standardized time series data with a mean of 0 and a variance of 1.

[0012] As a further aspect of the present invention: the cross-dimensional transfer entropy causal inference module quantifies the intensity of directional information transmission between nodes through transfer entropy, and characterizes the final causal coupling strength through net transfer entropy; for any two upstream node time series X and downstream node time series Y with directed edges, the formula for calculating transfer entropy is: ; In the formula, Time series to time series The propagation entropy, Let Y be the value at time t+1; () represents the k-th order historical state vector of the Y sequence at time t, where k is the embedding dimension; Let be the l-th order historical state vector of the X sequence at time t. For the embedded dimension; This represents the probability distribution of the corresponding variable; The formula for calculating net transfer entropy is: ; In the formula, The reverse transfer entropy from time series Y to time series X. The net transfer entropy from X to Y is the value of X. The larger the value, the stronger the causal driving effect of X on Y. The cross-dimensional transfer entropy causal inference module uses the net transfer entropy as the weight of the directed edge, removes invalid edges whose net transfer entropy fails the significance test, completes the topology correction and weight update of the initial directed graph of the ecosystem chain, and generates a localized ecosystem chain causal network.

[0013] As a further aspect of the present invention: the ecosystem anomaly tracing and cascading early warning module incorporates a graph attention network model. It uses the temporal characteristics of nodes in the localized ecosystem causal network as input and net propagation entropy as edge weights to construct a weighted adjacency matrix. Through a multi-head attention mechanism, it learns the propagation dynamics of anomalies within the ecosystem. When the monitoring data of any node deviates from the normal statistical range, the module performs bidirectional propagation analysis: Reverse tracing analysis: tracing upstream along the incoming edges of the directed graph, calculating the conditional attribution probability of each upstream node for the current anomaly, and outputting the anomaly source set in descending order of probability; Forward propagation prediction: propagating downstream along the outgoing edges of the directed graph, predicting the probability and time delay of the anomaly affecting each downstream node based on the propagation dynamics parameters learned by the graph attention network, and outputting the corresponding set of affected nodes and the time window of impact; Finally, the module outputs a cascading early warning result consisting of a quadruple containing the anomaly source, propagation path, expected affected nodes, and the time window of impact.

[0014] As a further embodiment of the present invention: the monitoring and early warning terminal module includes an ecological visualization unit, a hierarchical early warning push unit, and a disposal work order generation unit; the ecological visualization unit is used to display the localized ecological chain causal network topology and mark the location of abnormal nodes and the abnormal propagation path; the hierarchical early warning push unit is used to push corresponding level early warning information to management personnel according to the scope and severity of the abnormal impact; the disposal work order generation unit is used to automatically generate corresponding ecological disposal suggestion work orders according to the source and propagation path of the abnormality.

[0015] Compared with the prior art, the beneficial effects of the present invention by adopting the above technical solution are as follows: 1. This invention addresses the core shortcomings of existing aquatic ecological monitoring systems, which treat multidimensional ecological elements as independent features for cluster analysis, lose causal transmission information in the ecological chain, and can only determine the abnormal state of monitoring points without locating the cause of the abnormality. It constructs various ecological monitoring elements into a structured directed topology based on the material and energy flow relationship, and establishes the coupling relationship between ecological elements through directional causal strength quantification. When an ecological anomaly is detected, it traces back the upstream driving link along the causal chain to accurately locate the ecological source that triggered the anomaly. This solution overcomes the limitation of traditional cluster analysis, which can only achieve state identification, and solves the technical problem of "knowing what but not why" in aquatic ecological monitoring. It provides clear targeted decision-making basis for precise ecological governance and significantly improves the diagnostic value and practical guidance of the monitoring system.

[0016] 2. This invention employs a graph attention network to model the anomaly propagation dynamics of the ecological chain causal network. Addressing the shortcomings of existing monitoring systems that only trigger alarms after anomalies are fully manifested in the monitoring data and cannot predict the anomaly spread trend, this invention learns the anomaly propagation patterns along the ecological chain based on a localized ecological causal network. When an anomaly occurs at an upstream ecological node but has not yet spread to downstream sensitive links, it predicts the anomaly propagation path, affected downstream ecological elements, and corresponding impact time windows along the causal link in a forward direction, outputting complete cascading early warning information. This solution advances the trigger node for ecological early warning from "anomaly manifestation" to "anomaly initiation," reserving sufficient emergency response time windows for ecological management. It achieves a functional upgrade from passive alarm to proactive prevention and control, significantly enhancing the ability to prevent and control ecological risks in advance.

[0017] 3. This invention utilizes an adaptive learning mechanism that combines prior ecological chain directed graph constraints with data-driven causal inference. Addressing the shortcomings of existing monitoring systems—such as the use of general analytical models, inability to adapt to the differences in ecological characteristics of different water bodies, and susceptibility to diagnostic biases when applied across water bodies—this invention uses ecological prior topology as a reasonable constraint. Based on actual monitoring time-series data of the target water body, it dynamically corrects the topological structure and coupling strength of the ecological chain, automatically generating a localized ecological causal network adapted to the target water body. Furthermore, the model can be continuously iterated and optimized as monitoring data accumulates. This approach eliminates the dependence of fixed empirical models on general statistical laws, adaptively matching the ecological characteristics of different water bodies such as lakes, rivers, and reservoirs. It effectively avoids the adaptability problems when transferring general models, significantly improving the diagnostic accuracy and applicability of the system in different water body scenarios. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the system flow in an embodiment of the present invention; Figure 2This is a schematic diagram of the anomaly tracing and cascading early warning workflow in an embodiment of the present invention. Detailed Implementation

[0019] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0020] Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0021] Please see the appendix Figure 1 -Appendix Figure 2 The present invention discloses an intelligent water body ecological monitoring system based on multi-source data fusion, including a multi-source ecological data acquisition subsystem, a wireless data transmission module and a monitoring and early warning terminal module, as well as a time series data preprocessing module, an ecological chain directed graph construction module, a cross-dimensional transfer entropy causal inference module and an ecological chain anomaly tracing and cascading early warning module. The output of the multi-source ecological data acquisition subsystem is connected to the input of the wireless data transmission module, and the output of the wireless data transmission module is connected to the input of the time-series data preprocessing module; the output of the ecological chain directed graph construction module is connected to the first input of the cross-dimensional transfer entropy causal inference module, and the output of the time-series data preprocessing module is connected to the second input of the cross-dimensional transfer entropy causal inference module; the output of the cross-dimensional transfer entropy causal inference module is connected to the input of the ecological chain anomaly tracing and cascading early warning module, and the output of the ecological chain anomaly tracing and cascading early warning module is connected to the input of the monitoring and early warning terminal module. The multi-source ecological data acquisition subsystem is used to collect real-time time-series data of seven ecological monitoring elements in water bodies, including hydrological and meteorological data, water quality, sediment quality, submerged plants, phytoplankton, zooplankton, and algae. The wireless data transmission module is used to wirelessly upload the raw time-series data collected by the multi-source ecological data acquisition subsystem to the cloud processing unit; The directed graph construction module for the ecological chain is used to map various ecological monitoring elements into directed graph nodes based on prior knowledge of aquatic ecology, define directed ecological relationship edges between nodes according to the direction of material and energy flow in the ecosystem, construct the initial directed graph topology of the ecological chain, and output it to the cross-dimensional entropy causal inference module. The time series data preprocessing module receives raw time series data from the cloud, performs outlier removal, missing value completion, and standardization processing in sequence, and outputs standardized time series data to the cross-dimensional entropy causal inference module. The cross-dimensional entropy causal inference module is used to receive the initial directed graph topology and standardized time-series data of the ecosystem chain, quantify the directional causal coupling strength between the nodes corresponding to each directed edge, dynamically correct the topological structure and edge weight of the directed graph, generate a localized ecosystem chain causal network, and output it to the ecosystem chain anomaly tracing and cascading early warning module. The Ecosystem Anomaly Tracing and Cascade Early Warning Module is used to perform node state anomaly detection based on the localized ecosystem causal network. When an anomaly is detected in any node, it performs reverse tracing analysis and forward propagation prediction along the directed graph, and outputs early warning results including the anomaly source, propagation path, affected nodes, and impact time window. The monitoring and early warning terminal module is used to receive and visualize the results of anomaly tracing and cascading early warning information.

[0022] Example 1

[0023] Taking the monitoring of eutrophication in shallow lakes as an application scenario, this paper presents an intelligent water ecological monitoring system based on multi-source data fusion. The specific implementation process is as follows: (1) System deployment and data acquisition Five monitoring points were set up in the target lake, and a multi-source ecological data acquisition subsystem was deployed at each point, including a hydrological and meteorological sensor unit, a water quality sensor unit, a sediment monitoring unit, and an aquatic organism sampling unit.

[0024] The hydro-meteorological sensor unit uses ultrasonic level gauges, electromagnetic flow meters, and water temperature sensors to collect three types of data: flow rate, water level, and water temperature, with a sampling frequency of once every 15 minutes. The water quality sensor unit uses a multi-parameter water quality probe to collect four types of data: dissolved oxygen, total phosphorus, ammonia nitrogen, and permanganate index, with a sampling frequency of once every 30 minutes. The sediment monitoring unit uses in-situ sediment sensors to collect five types of data: oxidation-reduction potential, pH value, total nitrogen, total phosphorus, and organic matter, with a sampling frequency of once every 24 hours. The aquatic organism sampling unit uses underwater imaging and biological sampling equipment to collect four types of data: submerged plant biomass, phytoplankton community and residue, zooplankton classification and biomass, and algae classification and biomass, with a sampling frequency of once every 24 hours.

[0025] The raw time-series data collected from each monitoring point is uploaded via a wireless data transmission module: the data from the five nearby points is aggregated to the lakeside gateway via a ZigBee wireless network, and the gateway remotely uploads the data to the cloud processing unit via a 5G cellular network to complete the aggregation and storage of the raw data.

[0026] (2) Time series data preprocessing The cloud-based time series data preprocessing module receives the raw time series data and performs preprocessing in three steps: First, outliers are removed using the 3σ criterion: the mean and standard deviation of the single-factor time series data are calculated, and data with values ​​exceeding the range of [μ-3σ, μ+3σ] are identified as outliers and removed.

[0027] Taking water temperature data for a certain month in summer as an example, the mean μ of the water temperature time series data for this period is 22.5℃, the standard deviation σ is 1.8℃, and the corresponding 3σ confidence interval is [17.1℃, 27.9℃]. Water temperature data that exceed this interval are judged as outliers and removed.

[0028] The second step is to use linear interpolation to fill in the missing values: for missing data points caused by sensor failure or transmission interruption, the missing data points are filled in by the linear interpolation results of the two valid data points before and after them, forming a continuous time sequence.

[0029] For data segments that are missing for more than two sampling periods, they are marked and then filled with the historical average of the same location during the same period.

[0030] The third step is to unify the dimensions using Z-score standardization: the time series data of each element are standardized according to the formula... The transformation is performed, where x is the original value, μ is the mean of the time series data of the element, and σ is the standard deviation. The output is standardized time series data with a mean of 0 and a variance of 1.

[0031] (3) Construction of the initial directed graph of the ecosystem chain The directed graph construction module for the ecological chain is based on prior knowledge of lake ecology to construct an initial directed graph for the ecological chain: the nodes are divided into 5 categories with a total of 19 nodes according to the ecological level: the environmental driving layer nodes include 3 nodes of flow, water level and water temperature; the sediment nutrient layer nodes include 5 nodes of redox potential, pH value, total nitrogen, total phosphorus and organic matter; the primary production layer nodes include 3 nodes of submerged plant biomass, phytoplankton biomass and algae biomass; the primary consumption layer nodes include 1 node of zooplankton biomass; the water quality feedback layer nodes include 4 nodes of dissolved oxygen, total phosphorus, ammonia nitrogen and permanganate index, as well as 3 nodes of nitrogen and phosphorus nutrient cycling.

[0032] Four types of directed edges are defined between nodes: nutrient release edges from the sediment nutrient layer to the primary production layer, feeding relationship edges from the primary production layer to the primary consumption layer, competition relationship edges between submerged plants and phytoplankton, and decomposition feedback edges from the decomposition of each biological community to the water quality feedback layer. A total of 27 initial directed edges are defined, and each edge is marked with the corresponding ecological relationship type as an edge attribute.

[0033] (4) Generation of localized ecological chain causal network The cross-dimensional transfer entropy causal inference module receives the initial directed graph of the ecosystem chain and standardized time series data, and performs causal inference and topology correction: First, the Cao method is used to determine the embedding dimension of each time series sequence. In this embodiment, the embedding dimension k of the downstream sequence is 3, and the embedding dimension l of the upstream sequence is 2. The first minimum value method of mutual information is used to determine the optimal time delay parameter. In this embodiment, the time delay τ is 1 sampling step, corresponding to a time interval of 30 minutes.

[0034] For each directed edge in the initial directed graph Using the upstream node's time-series data as X and the downstream node's time-series data as Y, calculate the propagation entropy TE. Simultaneously calculate the reverse transit entropy TE This leads to the net transfer entropy.

[0035] The formula for calculating the transfer entropy is: ; This embodiment uses the Kozachenko-Leonenko nearest neighbor estimator to calculate the probability distribution, with the number of nearest neighbors set to 4 to reduce the computational complexity under high-dimensional data.

[0036] Taking the directed edge “total phosphorus in sediment → algal biomass” as an example, by substituting the standardized time series data for 30 consecutive days, we can calculate the forward propagation entropy TE(total phosphorus in sediment → algal biomass) = 0.48 bits and the reverse propagation entropy TE(algal biomass → total phosphorus in sediment) = 0.13 bits.

[0037] The formula for calculating net transfer entropy is: ; Substituting the numerical values, we obtain the net transit entropy of this edge: t.

[0038] Significance test on net transfer entropy: 100 sets of surrogate data were generated by randomly shuffling the time order of the upstream X sequence, and the transfer entropy value corresponding to each set of surrogate data was calculated. A zero distribution was constructed, and the 95th percentile of the zero distribution was taken as the significance threshold.

[0039] In this embodiment, the saliency threshold corresponding to this edge is 0.20 bits. Since the calculated net transit entropy of 0.35 bits is greater than the threshold, this edge is determined to be a significant and valid edge and is retained.

[0040] Similarly, all 27 initial edges are calculated and tested one by one. Finally, 3 invalid edges with net transfer entropy lower than the threshold are removed. For example, the net transfer entropy of the edge "water temperature → submerged plant biomass" is 0.12 bits, which is lower than the corresponding threshold of 0.18 bits, so it is removed. 24 valid edges are retained. The net transfer entropy value is used as the weight of the corresponding edge to finally generate a localized ecological chain causal network adapted to the lake.

[0041] (5) Anomaly tracing and cascading early warning The ecosystem anomaly tracing and cascading early warning module incorporates a graph attention network model. It uses a localized ecosystem causal network as its topological foundation, standardized temporal characteristics of each node as input, and net transit entropy as edge weights to construct a weighted adjacency matrix.

[0042] The model employs a two-head attention mechanism, with a hidden layer dimension of 64, a learning rate of 0.001, and 200 training epochs. The loss function uses cross-entropy loss combined with a causal consistency regularization term, with weights of 0.8 and 0.2, respectively. Training is completed using labeled data from historical normal and abnormal periods to learn the propagation dynamics of anomalies in the ecosystem.

[0043] When the system detects in real time that the algal biomass node data deviates from the normal statistical range (exceeding the 95% confidence interval of the historical normal range), it triggers bidirectional propagation analysis: Reverse source tracing analysis: Tracing upstream from the anomalous algal biomass node along the incoming edge direction, calculating the conditional attribution probability of upstream nodes such as total phosphorus in sediment, water temperature, and zooplankton biomass for the current anomaly.

[0044] In this embodiment, the attribution probability of total phosphorus in the sediment was calculated to be 0.72, the attribution probability of water temperature was 0.18, and the attribution probability of zooplankton feeding pressure was 0.10. The sorting output "sediment nutrient release" was the most likely source of the anomaly.

[0045] Forward propagation prediction: The anomaly propagates downstream from the node with abnormal algal biomass along the outgoing edge direction. Based on the propagation parameters learned by the graph attention network, it is predicted that the anomaly will successively affect downstream water quality nodes such as dissolved oxygen and ammonia nitrogen.

[0046] The probability of an abnormality in the dissolved oxygen node is 0.89, with an expected impact time window of 12-24 hours after the abnormality occurs; the probability of an abnormality in the ammonia nitrogen node is 0.76, with an expected impact time window of 24-48 hours after the abnormality occurs.

[0047] The final module outputs a cascaded early warning result for the quadruple: the source of the anomaly is the release of total phosphorus from the sediment, and the propagation path is total phosphorus in the sediment → algal biomass → dissolved oxygen → ammonia nitrogen. The expected affected nodes are dissolved oxygen and ammonia nitrogen, and the impact time window is 12-48 hours.

[0048] (6) Output of early warning results The monitoring and early warning terminal module receives the early warning results: the ecological visualization unit marks the location and propagation path of abnormal nodes on the lake's ecological topology map; the graded early warning push unit determines the level II early warning based on the scope of impact and pushes the early warning information to the lake management department; the disposal work order generation unit automatically generates a disposal suggestion work order of "control of endogenous pollution in bottom sediment + emergency disinfection of algae" based on the source of the abnormality.

[0049] This embodiment uses historical monitoring data of the lake throughout 2023 for verification, comparing the performance of existing adaptive clustering schemes with that of this system: the anomaly identification accuracy of the existing clustering scheme is 82.3%, and it cannot output source tracing results, with the anomaly already fully manifested when the warning is triggered; the anomaly identification accuracy of this system is 91.5%, the top 1 accuracy of anomaly source tracing is 78.2%, the top 3 accuracy is 92.6%, and the average warning lead time is 18.7 hours, with significant improvements in all indicators.

[0050] Example 2

[0051] Taking the monitoring of external pollution sources in rivers as an application scenario, this paper provides an intelligent water ecological monitoring system based on multi-source data fusion. The difference between this system and Example 1 is that the ecological chain topology is adapted to the ecological characteristics of the river. The specific implementation process is as follows: (1) System deployment and data acquisition Monitoring points were set up at three sections of the target river, and a multi-source ecological data acquisition subsystem was deployed to monitor hydrological flow, water quality indicators, sediment quality, and plankton communities.

[0052] The hydro-meteorological sensor unit collects flow velocity, water level, and water temperature data at a sampling frequency of once every 10 minutes; the water quality sensor unit collects dissolved oxygen, total phosphorus, ammonia nitrogen, and permanganate index data at a sampling frequency of once every 20 minutes; the sediment monitoring unit collects sediment total nitrogen, total phosphorus, and organic matter data at a sampling frequency of once every 48 hours; and the aquatic organism sampling unit collects biomass and community data of phytoplankton, zooplankton, and algae at a sampling frequency of once every 48 hours.

[0053] Data from each location is aggregated via a ZigBee network and then uploaded to the cloud processing unit via a 4G network.

[0054] (2) Time series data preprocessing The time series data preprocessing module adopts the same three-step preprocessing process as in Example 1. In view of the characteristics of large fluctuations in river flow and stronger data noise, the outlier removal step combines the flow change rate for joint judgment to exclude pseudo-anomaly data caused by natural hydrological fluctuations and output standardized time series data.

[0055] (3) Construction of the initial directed graph of the ecosystem chain In response to the characteristics of river ecosystems, which are highly mobile and significantly affected by external inputs, the directed graph construction module of the ecological chain adds external pollution input nodes as environmental driving layer nodes in the initial topology, and adjusts the hierarchical relationship of the ecological chain: the environmental driving layer includes three nodes: flow, water temperature, and external pollution input; the sediment nutrient layer includes three nodes: total nitrogen, total phosphorus, and organic matter; the primary production layer includes two nodes: phytoplankton and algae; the primary consumption layer includes one node: zooplankton; and the water quality feedback layer includes four nodes: dissolved oxygen, total phosphorus, ammonia nitrogen, and permanganate index.

[0056] The initial definition includes 21 directed edges, with a focus on strengthening the edges that directly affect water quality and plankton from external input nodes.

[0057] (4) Generation of localized ecological chain causal network The cross-dimensional transfer entropy causal inference module adopts the same transfer entropy calculation method as in Example 1. In view of the non-stationary characteristics of river time series data, a sliding window method is used to dynamically calculate the net transfer entropy. The window length is set to 30 days and the step size is set to 7 days to realize the dynamic update of the ecological chain coupling strength.

[0058] After significance testing, 18 valid edges were retained from the initial 21 edges. Among them, the net transfer entropy value of exogenous pollution input → ammonia nitrogen was the highest, which verified the direct driving effect of exogenous input on river water quality. Finally, a localized ecological chain causal network adapted to the river was generated.

[0059] (5) Anomaly tracing and cascading early warning When a sudden anomaly is detected at the ammonia nitrogen level, the ecological chain anomaly tracing and cascading early warning module triggers two-way propagation analysis: Reverse source tracing analysis: Tracing upwards along the ingress edge, the attribution probability of external pollution input was calculated to be 0.85, the attribution probability of sediment nitrogen release was 0.12, and the attribution probability of plankton decomposition was 0.03. The abnormal source was determined to be upstream external pollution input.

[0060] Forward propagation prediction: Predicting downstream along the corresponding ecological chain outflow direction of the river, it is expected that the anomaly will affect the permanganate index node of the downstream section in 6-12 hours and the algal biomass node in 12-24 hours, and output the corresponding cascade warning information.

[0061] The verification results of this embodiment show that, for the scenario of sudden external pollution in rivers, the anomaly tracing accuracy of this system is 83.5%, and the average early warning lead time is 10.2 hours, which can provide clear tracing basis and response time window for emergency response to river pollution.

[0062] In summary, the two embodiments cover two typical application scenarios: eutrophication of shallow lakes and external pollution of rivers, respectively, verifying the adaptability of the ecological chain causal network architecture of this system under different water body ecological characteristics.

[0063] Compared to traditional clustering monitoring schemes, this system achieves accurate anomaly tracing and early cascading warnings, significantly improving diagnostic depth, warning timeliness, and scenario adaptability. It can provide reliable technical support for precise water body ecological governance and proactive risk prevention and control.

[0064] While the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the invention. Any variations and modifications can be made by those skilled in the art without departing from the spirit and scope of the invention. Therefore, any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention, without departing from the scope of the invention, fall within the protection scope defined by the claims of the present invention.

Claims

1. A smart water body ecological monitoring system based on multi-source data fusion, comprising a multi-source ecological data acquisition subsystem, a wireless data transmission module, and a monitoring and early warning terminal module, characterized in that, It also includes a time-series data preprocessing module, an ecosystem directed graph construction module, a cross-dimensional entropy causal inference module, and an ecosystem anomaly tracing and cascading early warning module; The wireless data transmission module is used to wirelessly upload the raw time-series data collected by the multi-source ecological data acquisition subsystem to the cloud processing unit. The directed graph construction module of the ecological chain is used to map various ecological monitoring elements into directed graph nodes based on prior knowledge of aquatic ecology, define the directed ecological relationship edges between nodes according to the direction of material and energy flow in the ecosystem, construct the initial directed graph topology of the ecological chain, and output it to the cross-dimensional entropy causal inference module. The time series data preprocessing module is used to receive raw time series data from the cloud, sequentially perform outlier removal, missing value completion and standardization processing, and output standardized time series data to the cross-dimensional entropy causal inference module. The cross-dimensional transfer entropy causal inference module is used to receive the initial directed graph topology and standardized time series data of the ecosystem chain, quantify the directional causal coupling strength between the nodes corresponding to each directed edge, dynamically correct the topological structure and edge weight of the directed graph, generate a localized ecosystem chain causal network and output it to the ecosystem chain anomaly tracing and cascading early warning module. The ecosystem anomaly tracing and cascading early warning module is used to perform node state anomaly detection based on the localized ecosystem causal network. When an anomaly is detected in any node, it performs reverse tracing analysis and forward propagation prediction along the directed graph, and outputs an early warning result including the anomaly source, propagation path, affected nodes and the time window of impact. The monitoring and early warning terminal module is used to receive and visualize the results of anomaly tracing and cascading early warning information.

2. The intelligent water body ecological monitoring system based on multi-source data fusion according to claim 1, characterized in that, The multi-source ecological data acquisition subsystem includes a hydro-meteorological sensor unit, a water quality sensor unit, a sediment monitoring unit, and an aquatic organism sampling unit. The hydro-meteorological sensor unit is used to collect data on three types of hydro-meteorological elements: flow rate, water level, and water temperature. The water quality sensor unit is used to collect data on four types of water quality elements: dissolved oxygen, total phosphorus, ammonia nitrogen, and permanganate index. The sediment monitoring unit is used to collect data on five types of sediment elements: oxidation-reduction potential, pH value, total nitrogen, total phosphorus, and organic matter. The aquatic organism sampling unit is used to collect data on four types of aquatic organism elements: submerged plant biomass, phytoplankton community and residue, zooplankton classification and biomass, and algae classification and biomass.

3. The intelligent water body ecological monitoring system based on multi-source data fusion according to claim 1, characterized in that, The wireless data transmission module adopts a two-level transmission architecture of near-end networking and remote uploading; the data acquisition devices at each monitoring point near the end complete data aggregation through the ZigBee wireless network, and the aggregated data is remotely uploaded to the cloud processing unit through the 4G / 5G cellular network.

4. The intelligent water body ecological monitoring system based on multi-source data fusion according to claim 1, characterized in that, In the initial directed graph of the ecological chain constructed by the ecological chain directed graph construction module, the nodes are divided into five categories according to ecological level: environment-driven layer nodes, sediment nutrient layer nodes, primary production layer nodes, primary consumption layer nodes, and water quality feedback layer nodes. The directed edges between nodes are divided into four categories according to ecological relationship type: nutrient release edge, feeding relationship edge, competition relationship edge, and decomposition feedback edge. The direction of each directed edge is consistent with the direction of material and energy transfer in the ecosystem. Each node carries the time-series monitoring data of the corresponding ecological element as the node attribute, and each directed edge carries the corresponding ecological relationship type as the edge attribute.

5. The intelligent water body ecological monitoring system based on multi-source data fusion according to claim 1, characterized in that, The time series data preprocessing module performs the following steps: First, outliers in the original time series data are removed using the 3σ criterion; second, missing values ​​in the data are filled in using linear interpolation to form a continuous time series; third, the Z-score standardization method is used to unify the dimensions of the time series data of each element, and the output is standardized time series data with a mean of 0 and a variance of 1.

6. The intelligent water body ecological monitoring system based on multi-source data fusion according to claim 1, characterized in that, The cross-dimensional transfer entropy causal inference module quantifies the intensity of directional information transmission between nodes through transfer entropy and characterizes the final causal coupling strength through net transfer entropy. For any two upstream nodes with directed edges corresponding to time series X and downstream nodes with directed edges corresponding to time series Y, the formula for calculating transfer entropy is: ； In the formula, Time series to time series The propagation entropy, Let Y be the value at time t+1; () represents the k-th order historical state vector of the Y sequence at time t, where k is the embedding dimension; Let be the l-th order historical state vector of the X sequence at time t. For the embedded dimension; This represents the probability distribution of the corresponding variable; The formula for calculating net transfer entropy is: ； In the formula, The reverse transfer entropy from time series Y to time series X. The net transfer entropy from X to Y is the value of X. The larger the value, the stronger the causal driving effect of X on Y. The cross-dimensional transfer entropy causal inference module uses the net transfer entropy as the weight of the directed edge, removes invalid edges whose net transfer entropy fails the significance test, completes the topology correction and weight update of the initial directed graph of the ecosystem chain, and generates a localized ecosystem chain causal network.

7. The intelligent water body ecological monitoring system based on multi-source data fusion according to claim 1, characterized in that, The ecosystem anomaly tracing and cascading early warning module incorporates a graph attention network model. It uses the temporal characteristics of nodes in the localized ecosystem causal network as input and net propagation entropy as edge weights to construct a weighted adjacency matrix. Through a multi-head attention mechanism, it learns the propagation dynamics of anomalies within the ecosystem. When any node's monitoring data deviates from the normal statistical range, the module performs bidirectional propagation analysis: Reverse tracing analysis: tracing upstream along the incoming edges of the directed graph, calculating the conditional attribution probability of each upstream node for the current anomaly, and outputting the anomaly source set in descending order of probability; Forward propagation prediction: propagating downstream along the outgoing edges of the directed graph, predicting the probability and time delay of the anomaly affecting each downstream node based on the propagation dynamics parameters learned by the graph attention network, and outputting the corresponding set of affected nodes and the time window of impact. Finally, the module outputs a cascading early warning result consisting of a quadruple containing the anomaly source, propagation path, expected affected nodes, and the time window of impact.

8. The intelligent water body ecological monitoring system based on multi-source data fusion according to claim 1, characterized in that, The monitoring and early warning terminal module includes an ecological visualization unit, a tiered early warning push unit, and a disposal work order generation unit. The ecological visualization unit is used to display the localized ecological chain causal network topology and mark the location of abnormal nodes and the abnormal propagation path. The tiered early warning push unit is used to push corresponding level early warning information to management personnel according to the scope and severity of the abnormal impact. The disposal work order generation unit is used to automatically generate corresponding ecological disposal suggestion work orders according to the source and propagation path of the abnormality.

Images