Lake water quality abnormity monitoring method and system based on multi-source sensing data fusion

Abstract

The invention discloses a lake water quality abnormity monitoring method and system based on multi-source sensing data fusion, and the method comprises the steps: collecting water quality data through a multi-source sensor, and obtaining high-quality fusion data through weighted fusion and Kalman filtering processing; and an abnormal threshold is dynamically adjusted in combination with a seasonal baseline, and a polluted area is accurately identified. Incremental sensor layout is optimized based on hydrological characteristics, a fusion network is expanded through an adaptive algorithm, and monitoring coverage and data consistency are improved. A comprehensive abnormal index is constructed by utilizing principal component analysis and multi-parameter coupling, so that time-space accurate positioning and visual display of lake water quality abnormity are realized, and intelligence, real-time performance and early warning capability of water quality monitoring are remarkably enhanced.

Description

Technical Field

[0001] This invention belongs to the field of water quality monitoring, and in particular relates to a method and system for monitoring lake water quality anomalies based on multi-source sensor data fusion. Background Technology

[0002] Lake water quality monitoring is a crucial area of ​​environmental protection and water resource management, directly impacting ecological balance, drinking water safety, and regional sustainable development. Lake water quality is influenced by various factors, such as industrial emissions, agricultural runoff, and changes in the natural environment. Anomalies such as eutrophication or heavy metal pollution can severely damage ecosystems. Therefore, developing efficient and accurate monitoring methods is essential for the timely detection and response to water quality anomalies.

[0003] Current monitoring methods rely heavily on single or limited sensors, making it difficult to comprehensively capture the complex changes in lake water quality, especially when dealing with multi-source data integration and dynamic environmental adaptation. Existing water quality monitoring methods face the problem of insufficient data consistency when processing multi-source sensor data. Different types of sensors, such as optical and electrochemical sensors, collect data with differences in timestamps, formats, and accuracy, making direct comparison or integration difficult. For example, turbidity sensors may produce abnormal readings due to disturbance from sediment at the lake bottom, while pH sensors may drift due to equipment aging. These inconsistencies make it difficult to accurately distinguish between actual water quality anomalies and sensor errors during comprehensive analysis. Insufficient data consistency further poses challenges to setting dynamic thresholds. Lake water quality is affected by factors such as season, water flow, and rainfall, and anomaly detection thresholds need to be dynamically adjusted according to environmental changes.

[0004] However, existing methods struggle to effectively integrate multi-source data to construct highly adaptive threshold models. For instance, during the rainy season, a lake area might experience a temporary increase in turbidity due to runoff input; if the threshold isn't adjusted promptly, it could be misjudged as a pollution event. This lack of dynamic adaptation reduces the reliability of monitoring systems in complex environments. Furthermore, the difficulty in setting dynamic thresholds presents challenges for expanding monitoring networks.

[0005] As the monitoring scope of lakes expands, new sensors need to be quickly integrated into the existing system. However, the data differences between new and old equipment and the rationality of their deployment locations directly affect the fusion effect. For example, newly deployed sensors may collect redundant or biased data because they do not consider the direction of water flow, causing the overall monitoring network to fail to effectively reflect the spatial heterogeneity of lake water quality.

[0006] Therefore, how to construct a dynamically adaptable threshold model and achieve seamless integration of new sensors with existing networks when there is insufficient consistency in multi-source sensor data has become a key issue in monitoring lake water quality anomalies. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention provides a method and system for monitoring lake water quality anomalies based on multi-source sensor data fusion.

[0008] One method for monitoring lake water quality anomalies based on multi-source sensor data fusion includes: Multi-source sensor data is acquired based on multi-source sensors, and the weight coefficients of each data are calculated through a weighted fusion algorithm to obtain preliminary fused data; Based on the preliminary fused data, Kalman filtering is used to process outliers, perform initial screening, and align timestamps to obtain a high-quality fused dataset. If the high-quality fused dataset exceeds the preset seasonal baseline, the range of each water quality indicator is adjusted through a dynamic threshold model to obtain the anomaly judgment threshold. Based on the anomaly detection threshold, the distribution and flow direction of pollution sources in the lake's hydrological characteristics are obtained, and the optimal deployment location of the incremental sensors is determined. An extended fusion network is obtained by adjusting the association rules between the incremental sensor data and the original network using an adaptive weighting algorithm. Based on the extended fusion network, key factors are extracted using principal component analysis and coupled with multiple parameters to obtain a comprehensive anomaly index. Based on the comprehensive anomaly index, data segmentation is performed on lake sub-regions and time periods to obtain a spatiotemporal scale anomaly distribution map and generate a visual heat map.

[0009] Preferably, the process of obtaining preliminary fused data includes: Acquire multi-source sensor data, including dissolved oxygen data, turbidity data, pH data, and temperature data, and remove outliers through preprocessing to obtain a cleaning dataset; Based on the cleaning dataset, Kalman filtering is used to smooth the dissolved oxygen data, turbidity data, pH data, and temperature data to obtain a smoothed dataset; If the dissolved oxygen data and temperature data in the smoothed dataset deviate from a preset threshold, then outliers are marked, resulting in a marked dataset. Based on the labeled dataset, the correlation between the turbidity data and the pH data is determined, and the relationship is fitted using a linear regression model to obtain the relevant parameter set; The weighted fusion algorithm is used to calculate the weight coefficients for dissolved oxygen data, turbidity data, pH value data, and temperature data for the relevant parameter set, thus obtaining a set of weight coefficients. Based on the dissolved oxygen data, turbidity data, pH value data, and temperature data in the fusion smoothing dataset using the weighted coefficient set, preliminary fusion data is obtained; Principal component analysis was used to reduce the dimensionality of the preliminary fused data, retaining the principal components to obtain the dimensionality-reduced fused data.

[0010] Preferably, the process of obtaining a high-quality fused dataset includes: Based on the preliminary fused data, a Kalman filter is used to predict the state vector, and a preliminary state estimate is obtained. The preliminary state estimate is compared with the observed value. If it exceeds the preset threshold, it is determined to be an outlier and filtered out to obtain a clean state dataset. Based on the clean status dataset, timestamp sequences are extracted, and the timestamps are aligned using linear interpolation to obtain the aligned status dataset; Apply Kalman filtering to the homogeneous state dataset to update the covariance matrix and obtain the fused state vector; The residual sequence is calculated based on the fused state vector. If the variance of the residual sequence converges, the high-quality index is determined, and the filtered fused dataset is obtained. The selected fusion dataset is divided into a training subset and a validation subset. A random forest model is used to train the prediction error, and the judgment value is used to filter out the accuracy. The accuracy of value filtering is combined with the stamp alignment error, and a weighted average is used to fuse them to obtain the final high-quality fused dataset.

[0011] Preferably, the process of obtaining the anomaly determination threshold includes: The dataset quality score is calculated based on the high-quality fused dataset. If the dataset quality score is higher than the seasonal baseline, the dynamic threshold model is activated to process the seasonal baseline data to obtain preliminary adjustment parameters. A dynamic threshold model is used to integrate the preliminary adjustment parameters with historical water quality indicators to determine the boundary of the indicator range. By analyzing the current values ​​of water quality indicators within the aforementioned indicator range boundaries, an initial signal of anomaly can be determined. If the aforementioned abnormal initial signal exists, model adjustment is performed to obtain an optimized threshold model; The K-means clustering algorithm is used to group the index ranges in the optimized threshold model to obtain an anomaly detection threshold cluster. The final anomaly detection threshold is obtained by verifying changes in the fused dataset through the anomaly detection threshold cluster.

[0012] Preferably, the process of determining the optimal placement location of the incremental sensors includes: Acquire lake hydrological characteristic data and pollution source distribution data, and fuse them to generate the first dataset; The first dataset is judged by the anomaly determination threshold to obtain the range of the anomaly region; The pollution diffusion path within the abnormal region is calculated using the water flow direction vector. If the pollution diffusion path covers the source location, then the set of affected hydrological feature points is determined. Obtain the distribution data density gradient from the set of affected hydrological feature points; The candidate set of incremental sensor deployment locations is determined based on the density gradient of the distributed data. The optimal location cluster is obtained by grouping the candidate location set using the K-means clustering algorithm.

[0013] Preferably, the process of obtaining the extended converged network includes: Feature vectors are extracted from the incremental sensor data using an adaptive weighting algorithm; The similarity matrix with the original network association rules is calculated based on the feature vectors to obtain the matching score; If the matching score is higher than a preset threshold, a weighted average method is used to fuse incremental sensor data and existing network association rules to obtain a preliminary fusion structure. The K-means clustering algorithm is applied to the preliminary fusion structure to divide the data into clusters, thereby obtaining a set of cluster centroids; The optimized weight vector is obtained by adjusting the parameters of the adaptive weighting algorithm based on the cluster center point set. The incremental sensor data and cluster center point set are re-fused using the optimized weight vector to generate an extended fusion network.

[0014] Preferably, the process of obtaining the comprehensive anomaly index includes: Data sources are obtained based on extended fusion networks, and key factors are extracted through principal component analysis to obtain a preliminary factor set; The preliminary factor set is fused with relevant attributes using a multi-parameter coupling method to obtain a coupling parameter set; Calculate the abnormal deviation for the coupling parameter group. If the abnormal deviation exceeds a preset threshold, an abnormal event is identified and an abnormality identifier is obtained. Based on the anomaly identifier, related data is obtained from the extended fusion network to obtain an extended data stream; The extended data stream is processed using the random forest algorithm to determine data patterns and obtain pattern classification results; The pattern classification results are compared with the key factors. If the matching degree is higher than the threshold, the index component is fused to obtain the updated index value. The updated index value is verified using a support vector machine algorithm to obtain the final anomaly index.

[0015] Preferably, the process of obtaining a spatiotemporal scale anomaly distribution map and generating a visual heatmap includes: Raw monitoring data of lake sub-regions are obtained through a pre-established database, and time series analysis is used to determine the abnormal index sequence within a time period. The abnormal index sequence is judged, and if there is a continuous upward trend in the sequence, it is marked as a high abnormal segment, and a set of marked abnormal segments is obtained. Spatiotemporal scale features are extracted from the marked set of anomalous paragraphs, and the spatiotemporal scale features are grouped using the K-means clustering algorithm to obtain the grouped spatiotemporal anomalous clusters. The distribution density of the grouped spatiotemporal anomaly clusters is calculated, and the density distribution map is determined by a gridded mapping method. Obtain the peak points in the density distribution map. If the peak points exceed a preset threshold, delineate the core abnormal regions and obtain a list of core abnormal regions. Select the dominant region from the list of core abnormal regions, and use an interpolation algorithm to generate a continuous surface to obtain a continuous abnormal surface; A color gradient mapping is constructed based on the continuous anomalous surface to generate a visual heatmap.

[0016] This invention also provides a lake water quality anomaly monitoring system based on multi-source sensor data fusion, comprising: The multi-source sensor data acquisition and weighted fusion module is used to acquire multi-source sensor data based on multi-source sensors, calculate the weight coefficient of each data through a weighted fusion algorithm, and obtain preliminary fused data. The Kalman filtering module is used to process outliers, perform initial screening, and align timestamps using Kalman filtering based on the preliminary fused data to obtain a high-quality fused dataset. The dynamic threshold adjustment module is used to adjust the range of various water quality indicators through a dynamic threshold model to obtain an anomaly judgment threshold if the high-quality fused dataset exceeds the preset seasonal baseline. The incremental sensor deployment optimization module is used to obtain the distribution and flow direction of pollution sources in the lake based on the anomaly determination threshold, and to determine the optimal deployment location of the incremental sensors. An adaptive weight adjustment module is used to adjust the association rules between the incremental sensor data and the original network through an adaptive weight algorithm to obtain an extended fusion network. The principal component analysis and multi-parameter coupling module is used to extract key factors based on the extended fusion network and perform multi-parameter coupling to obtain a comprehensive anomaly index. The spatiotemporal anomaly distribution visualization module is used to perform data segmentation processing on lake sub-regions and time periods based on the comprehensive anomaly index, obtain spatiotemporal scale anomaly distribution maps, and generate visualized heat maps.

[0017] Compared with the prior art, the present invention has the following advantages and technical effects: This invention proposes a method for monitoring lake water quality anomalies and optimizing sensor deployment. It achieves the overall technical effect of intelligent expansion of the high-precision real-time monitoring sensor network for water quality anomalies and intuitive presentation of anomaly distribution, thereby improving the efficiency and response speed of lake ecological management.

[0018] This invention combines a weighted fusion algorithm with Kalman filtering to effectively suppress noise and outliers in multi-source sensor data, improve data fusion accuracy and robustness, and enhance the system's adaptability to environmental disturbances.

[0019] This invention introduces a dynamic threshold model, which combines seasonal baselines and historical water quality data to automatically adjust the anomaly judgment criteria, avoiding misjudgments caused by seasonal changes or short-term environmental fluctuations, and improving the timeliness and accuracy of anomaly detection.

[0020] Based on anomaly detection results and lake hydrological characteristics, this invention scientifically calculates pollution diffusion paths and combines density gradient and clustering algorithms to determine the optimal deployment location of incremental sensors, thereby improving monitoring spatial coverage and data representativeness.

[0021] This invention achieves seamless integration of new sensor data with the existing network through an adaptive weighting algorithm, solving the problem of data differences between new and old devices, constructing an extended fusion network, and improving the system's adaptability when facing an expanded monitoring range.

[0022] This invention utilizes principal component analysis and multi-parameter coupling technology to extract key water quality factors, constructs a comprehensive anomaly index, and combines random forest and support vector machine models to achieve accurate identification and verification of complex water quality anomaly patterns, thereby improving the accuracy of early warning.

[0023] Based on a comprehensive anomaly index, this invention uses time series analysis, clustering, and interpolation algorithms to generate spatiotemporal scale anomaly distribution maps and visualized heat maps, which intuitively display the anomaly intensity and evolution trend of different areas and time periods of the lake, thus assisting in management decision-making.

[0024] The entire monitoring process of this invention, from data acquisition, fusion, analysis to visualization, can be completed automatically, reducing manual intervention, improving monitoring efficiency, and making it suitable for long-term, continuous, unattended lake water quality monitoring tasks.

[0025] In summary, this invention significantly improves the intelligence level, spatial coverage, anomaly identification accuracy, and response speed of lake water quality anomaly monitoring, providing strong technical support for lake ecological protection, water pollution prevention and control, and scientific management. Attached Figure Description

[0026] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the system structure according to an embodiment of the present invention. Detailed Implementation

[0027] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0028] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0029] Example 1 like Figure 1 As shown, this embodiment provides a method for monitoring lake water quality anomalies based on multi-source sensor data fusion, including: Multi-source sensor data is acquired based on multi-source sensors, and the weight coefficients of each data are calculated through a weighted fusion algorithm to obtain preliminary fused data; Based on the initial fused data, Kalman filtering was used to process outliers, perform initial screening, and align timestamps to obtain a high-quality fused dataset. If the high-quality fused dataset exceeds the preset seasonal baseline, the range of each water quality indicator is adjusted through a dynamic threshold model to obtain the anomaly judgment threshold. Based on the anomaly detection threshold, the distribution of pollution sources and the direction of water flow in the lake's hydrological characteristics are obtained, and the optimal deployment location of incremental sensors is determined. An extended fusion network is obtained by adjusting the association rules between incremental sensor data and the original network using an adaptive weighting algorithm. Based on the extended fusion network, principal component analysis is used to extract key factors and perform multi-parameter coupling to obtain a comprehensive anomaly index. Data segmentation of lake sub-regions and time periods is performed based on the comprehensive anomaly index to obtain a spatiotemporal scale anomaly distribution map and generate a visual heat map.

[0030] Furthermore, the process of obtaining preliminary fused data includes: Acquire multi-source sensor data, including dissolved oxygen data, turbidity data, pH data, and temperature data, and remove outliers through preprocessing to obtain a cleaning dataset; Based on the cleaning dataset, Kalman filtering was used to smooth the dissolved oxygen data, turbidity data, pH data, and temperature data to obtain a smoothed dataset. If the dissolved oxygen data and temperature data in the smoothed dataset deviate from a preset threshold, outliers are marked, resulting in a marked dataset. The correlation between turbidity data and pH data was determined based on the labeled dataset, and the relationship was fitted using a linear regression model to obtain the relevant parameter set; The weighted fusion algorithm is used to calculate the weight coefficients for dissolved oxygen data, turbidity data, pH value data, and temperature data for the relevant parameter set, thus obtaining a set of weight coefficients. Based on the weighted coefficient set, dissolved oxygen data, turbidity data, pH value data, and temperature data in the fusion smoothing dataset are used to obtain preliminary fused data; Principal component analysis was used to reduce the dimensionality of the data and initially fuse the data, retaining the main components to obtain the dimensionality-reduced fused data.

[0031] For example, in the field of water quality monitoring, the multi-source sensor data acquisition in this embodiment involves collecting dissolved oxygen, turbidity, pH value, and temperature data from river monitoring stations.

[0032] Specifically, dissolved oxygen data reflects oxygen content, turbidity data indicates water turbidity, pH data measures acidity and alkalinity, and temperature data affects the stability of other parameters.

[0033] Understandably, removing outliers through preprocessing can improve data quality. For example, if the dissolved oxygen in the original data is 8.5 mg / L but suddenly jumps to 50 mg / L, it is considered an outlier. Median filtering is used to replace it with the nearest mean of 7.2 mg / L, resulting in a clean dataset. This helps avoid bias in subsequent analysis and improves accuracy by more than 20%.

[0034] In one embodiment, this embodiment uses Kalman filtering to smooth and clean the dataset. Kalman filtering utilizes state prediction and measurement to update and merge historical and current values. For example, turbidity data sequences of 10, 12, and 11 NTU become 10.5, 11.2, and 11.0 NTU after filtering, reducing noise fluctuations, making the smoothed dataset more reliable, improving the stability of time series, and facilitating trend capture.

[0035] For example, if the deviation between the smoothed dissolved oxygen (8.0 mg / L) and the temperature (25℃) exceeds a threshold, such as a difference of more than 10% in the oxygen saturation calculation, an anomaly is marked, forming a labeled dataset. This can identify environmental anomalies early, such as pollution events causing oxygen-temperature mismatch, and prevent misjudgments of water quality health.

[0036] It should be noted that the correlation between turbidity and pH was determined by labeling the dataset, and a linear regression model was used for fitting. For example, the correlation coefficient between turbidity data and pH sequence reached 0.85, with a fitting slope of 0.2, resulting in a relevant parameter set such as an intercept of 7.0. This reveals the causal relationship between increased turbidity and decreased pH, supporting the tracing of pollution sources. Weights were calculated based on a weighted fusion algorithm using the relevant parameter set. For example, if the correlation is strong, turbidity would have a weight of 0.4, pH 0.3, dissolved oxygen 0.2, and temperature 0.1, forming a weighted coefficient set to optimize resource allocation and improve the robustness of the fusion.

[0037] In one possible implementation, this embodiment uses a set of weighted coefficients to fuse and smooth data, such as a weighted average to obtain a comprehensive index of 7.5, thus obtaining preliminary fused data. This can integrate multiple parameters and reduce the impact of single sensor failure on the analysis data preprocessing.

[0038] In one embodiment, after acquiring multi-source sensor data such as dissolved oxygen, turbidity, pH value and temperature data, outliers are removed through preprocessing to obtain a cleaning dataset.

[0039] For example, if a value such as 8.5 mg / L exists in the original dissolved oxygen data but significantly deviates from the normal range of 7-9 mg / L, it will be removed to ensure the reliability of the dataset. This method can reduce noise interference, improve the accuracy of subsequent analyses, and enhance the overall precision of water quality assessment.

[0040] For example, principal component analysis (PCA) is used to initially fuse data for dimensionality reduction. The first two principal components explain 90% of the variance, for example, reducing it to a 2-dimensional vector [0.8, 0.6]. This results in dimensionality-reduced fused data, which simplifies calculations while extracting essential features, facilitating input into subsequent machine learning models, reducing the risk of dimensionality disaster, and accelerating processing. This chain from preprocessing to dimensionality reduction supports real-time water quality early warning systems, improving overall monitoring efficiency and decision reliability.

[0041] Furthermore, the process of obtaining a high-quality fused dataset includes: Based on the preliminary fused data, Kalman filtering is used to predict the state vector and obtain preliminary state estimates. The preliminary state estimate is compared with the observed value. If it exceeds the preset threshold, it is identified as an outlier and filtered out to obtain a clean state dataset. The timestamp sequence is extracted from the clean state dataset, and the timestamps are aligned by linear interpolation to obtain the aligned state dataset. The covariance matrix of the aligned state dataset is updated using Kalman filtering to obtain the fused state vector; The residual sequence is calculated based on the fused state vector. If the variance of the residual sequence converges, the high-quality index is determined and the fused dataset is obtained. The selected fusion dataset is divided into a training subset and a validation subset. A random forest model is used to train the prediction error, and the judgment value is used to filter out the accuracy. The accuracy of value filtering and the stamp alignment error are combined and then fused using a weighted average to obtain the final high-quality fused dataset.

[0042] For example, in the field of water quality monitoring, this embodiment uses Kalman filtering to predict the state vector after obtaining preliminary fused data. This process predicts the state vector at the next moment based on preliminary fused values ​​of historical fused data such as dissolved oxygen and temperature.

[0043] Specifically, the state vector includes dissolved oxygen concentration of 8.5 mg / L, turbidity of 2 NTU, pH of 7.2, and temperature of 22 degrees Celsius. Future values ​​are inferred using a system model through a filter, such as a predicted dissolved oxygen level of 8.7 mg / L, thus obtaining a preliminary state estimate. This prediction helps to capture data trends in advance and avoid sudden fluctuations affecting subsequent analysis.

[0044] In one embodiment, the preliminary state estimate is compared with the actual observed value. If the deviation exceeds a preset threshold, such as a difference in dissolved oxygen exceeding 0.5 mg / L, it is judged as an outlier and screened out.

[0045] For example, if the observed dissolved oxygen is 9.5 mg / L but the estimated value is 8.7 mg / L, the deviation of 0.8 mg / L is marked as an anomaly. After screening, the remaining data forms a clean state dataset, ensuring the purity of the dataset and improving overall reliability.

[0046] In one embodiment, the clean status dataset extracts a timestamp sequence, such as recording once per minute, and aligns the timestamps using linear interpolation.

[0047] For example, if dissolved oxygen data is missing at 10:01, it can be interpolated from 8.5 mg / L at 10:00 and 8.6 mg / L at 10:02 to obtain 8.55 mg / L, thus aligning the state dataset. This alignment unifies the time reference of multi-source sensors, facilitating synchronous fusion.

[0048] For example, in this embodiment, the covariance matrix of the aligned state dataset is updated using Kalman filtering. Covariance reflects estimation uncertainty; for instance, if the initial covariance is 0.1, it is reduced to 0.05 after the update, resulting in a fused state vector that includes a more accurate dissolved oxygen level of 8.55 mg / L. This update process incorporates observation noise, enhancing the accuracy of the state vector.

[0049] In one possible implementation, the fused state vectors are used to calculate the residual sequence, i.e., the difference between the estimated value and the observed value, such as continuous residuals of 0.02, 0.01, and -0.01 mg / L. If the variance converges to below 0.0005, a high-quality index is determined, resulting in a filtered fused dataset. This convergence judgment filters out low-quality data and improves the stability of the dataset.

[0050] For example, the dataset is screened and divided into training and validation subsets, such as 70% training consisting of 1000 records, and the fusion process is analyzed.

[0051] In one embodiment, after acquiring preliminary fused data, a Kalman filter is used to predict the state vector to estimate the future state, which helps to capture dynamic changes in the data.

[0052] For example, in a water quality monitoring scenario, preliminary fusion data includes dissolved oxygen 0.5 mg / L, turbidity 10 NTU, pH 7.2, and temperature 20℃. Preliminary state estimates, such as dissolved oxygen 0.48 mg / L, are obtained through prediction. The random forest model is trained to predict errors, and the model uses decision tree ensemble to predict turbidity with an error of 0.1 NTU, achieving a 95% accuracy rate in filtering out incorrect values. This training helps quantify the error distribution, supporting precise filtering.

[0053] In one embodiment, the value filtering accuracy is combined with the timestamp alignment error, such as an accuracy of 95% and an alignment error of 0.05 minutes. A weighted average is then used for fusion, with weights set to 0.6 and 0.4, to obtain a final high-quality fused dataset that includes values ​​such as refined dissolved oxygen at 8.6 mg / L. This fusion integrates multi-dimensional indicators, ensuring high confidence in the final dataset and facilitating accurate decision-making in downstream water quality assessments.

[0054] Specifically, the logical progression from prediction to fusion involves first capturing dynamics through Kalman prediction, then filtering out and purifying data through anomaly screening, followed by alignment and updates to improve synchronization, residual convergence to ensure quality, model training to quantify errors, and finally weighted summation to obtain a reliable output. Multiple aspects of support, such as numerical deviation control and accuracy improvement, jointly enhance the robustness of multi-parameter fusion of water quality.

[0055] Furthermore, the process of obtaining the anomaly detection threshold includes: The dataset quality score is calculated based on the high-quality fused dataset. If the dataset quality score is higher than the seasonal baseline, the dynamic threshold model is activated to process the seasonal baseline data to obtain the initial adjustment parameters. A dynamic threshold model was used to integrate preliminary adjustment parameters with historical water quality indicators to determine the boundary of the indicator range. Analyze the current values ​​of water quality indicators by examining the boundaries of the indicator range to determine the initial signal of anomaly; If an abnormal initial signal exists, the model is adjusted to obtain an optimized threshold model; The K-means clustering algorithm was used to group the index ranges in the optimized threshold model to obtain an anomaly detection threshold cluster. The final anomaly detection threshold is obtained by verifying changes in the fused dataset through anomaly detection threshold cluster.

[0056] For example, in the processing of fused datasets for water quality monitoring, the fused dataset is first acquired and a quality score is calculated. The quality score can be obtained by weighting indicators such as completeness, accuracy and consistency, with a range of 0 to 100.

[0057] Specifically, if the score is higher than the seasonal baseline, such as the summer average of 85 points, it indicates that the data is highly reliable and sufficient to support subsequent adjustments, thereby activating the dynamic threshold model. This model can adapt to environmental fluctuations and avoid misjudgments caused by fixed thresholds.

[0058] In one embodiment, the seasonal baseline data includes historical water temperature, pH value, etc., which are processed to obtain preliminary adjustment parameters such as an offset coefficient of 0.5, used to compensate for seasonal deviations. This helps to improve the specificity of the parameters and bring about a more accurate boundary setting effect.

[0059] It should be noted that the dynamic threshold model integrates preliminary adjustment parameters with historical water quality indicators. For example, it integrates the average pH record of the past 30 days (7.2) to determine the boundary of the indicator range as 6.5 to 8.0. This integration is achieved through weighted averaging to ensure that the boundary reflects the true distribution, thereby enhancing the robustness of anomaly detection.

[0060] For example, analyzing the current value of water quality indicators such as pH 8.5, if it exceeds the boundary, an anomaly is identified, an initial signal is detected, and potential problems are quickly isolated step by step, reducing the computational burden on downstream systems and improving real-time monitoring efficiency.

[0061] Preferably, the K-means clustering algorithm is used to group the index ranges in the optimized version. For example, the pH data is divided into a normal cluster with a mean of 7.0 and an edge cluster with a mean of 7.8 to obtain the anomaly detection threshold cluster. This clustering is based on the principle of minimizing Euclidean distance. After grouping, the cluster boundaries are clear, which makes it easier to identify pattern deviations and brings about a diversified expression effect of the anomaly threshold cluster.

[0062] Specifically, the anomaly detection threshold is used to verify changes in the fused dataset. For example, the variance of the new data (0.3) is compared with the standard deviation of the cluster (0.4). If the change exceeds the threshold, the threshold is adjusted, and the final anomaly detection threshold of 8.2 is obtained. This verification mechanism ensures that the threshold is dynamically updated and enhances the cleanliness of the dataset.

[0063] In one embodiment, the progression from core boundary analysis to cluster verification can support each other to form a closed loop. For example, the initial signal triggers optimization, clustering verifies changes, and the final threshold is used to screen out anomalies such as sudden pollution values ​​of 9.0. Through multi-faceted analysis, it is shown that the high-quality score activation mechanism avoids interference from low-quality data, while clustering grouping improves the granularity of the threshold. The overall anomaly detection accuracy is improved by 20%, and it supports seamless integration of water quality history and current data. It avoids the residual amplification analysis after timestamp alignment. The process of fusion dataset - obtaining fusion dataset and calculating quality score usually involves assessing data integrity and consistency.

[0064] For example, the quality score in this embodiment is calculated using metrics such as missing rate and noise level. If the score is higher than the seasonal baseline, such as 0.8, subsequent processing is triggered. This is closely linked to the previous Kalman filter cleaning dataset to ensure that anomaly handling is more reliable after the fused state vector residuals converge.

[0065] Furthermore, the process of determining the optimal placement location of the incremental sensors includes: Acquire lake hydrological characteristic data and pollution source distribution data, and fuse them to generate the first dataset; The first dataset is judged by an anomaly detection threshold to obtain the range of anomaly regions; The pollution diffusion path is calculated within the abnormal region using the water flow direction vector; If the pollution diffusion path covers the source location, then the set of affected hydrological feature points is determined. Obtain the distribution data density gradient from the set of affected hydrological feature points; Determine the candidate set of incremental sensor deployment locations based on the density gradient of the distributed data; The optimal location cluster is obtained by grouping the candidate location set using the K-means clustering algorithm.

[0066] For example, in this embodiment, when fusing lake hydrological characteristic data and pollution source distribution data into a first dataset, lake hydrological characteristic data, such as water flow velocity, flow direction, and water depth, are collected through a sensor network, and combined with pollution source distribution data, such as the geographical location and discharge volume of factory discharge points and agricultural runoff areas. Assuming a lake with a water area of ​​50 square kilometers, flow velocity data from 100 hydrological monitoring points, ranging from 0.1 to 0.5 m / s, are collected. Simultaneously, the coordinates and emission concentrations of five known pollution sources are recorded, such as a chemical plant emitting 20 mg / L of organic matter per day. This data is fused into a first dataset through spatiotemporal alignment to ensure data consistency. During fusion, timestamp matching must be ensured, for example, data should be recorded uniformly in hourly units.

[0067] In one possible implementation, this embodiment uses anomaly threshold determination rules to determine the extent of anomaly areas. A threshold is set based on historical water quality data; for example, a chemical oxygen demand (COD) exceeding 50 mg / L is considered abnormal. Assuming the first dataset shows a COD value of 60 mg / L in a certain area, exceeding the threshold, it is marked as an anomaly area. The extent of the anomaly area is determined using spatial interpolation, covering approximately 2 square kilometers of the southeastern corner of the lake. This method can accurately locate anomaly areas, facilitating subsequent analysis.

[0068] Specifically, in this embodiment, when calculating the pollution diffusion path using the water flow direction vector, the water flow direction is analyzed using a hydrological model.

[0069] For example, in the southeast corner of the lake, the water flow direction is northwest, with a velocity of 0.3 m / s. Using a pollutant concentration gradient, the path of pollution spreading northwest from the chemical plant's discharge outlet was simulated. The path covers an area of ​​approximately 1.5 kilometers and involves three hydrological monitoring points. This method helps to track the trajectory of pollutant spread.

[0070] For example, when obtaining the density gradient of distribution data from a set of affected hydrological feature points, the rate of change of pollutant concentration at monitoring points can be analyzed. Assuming the concentrations at three affected points are 60 mg / L, 55 mg / L, and 50 mg / L, the calculated gradient shows that the concentration gradually decreases along the water flow direction, indicating a pollution diffusion trend. The density gradient can be calculated using spatial analysis tools, yielding a trend of a 5 mg / L decrease in concentration per kilometer, providing a basis for sensor deployment.

[0071] In one possible implementation, when determining the candidate set of incremental sensor deployment locations based on the distribution data density gradient, this embodiment prioritizes regions with significant gradient changes.

[0072] For example, additional sensors are needed in areas where concentration decreases most rapidly to improve monitoring accuracy. The candidate set includes five locations, each situated at a concentration gradient change rate greater than 3 mg / L / km. This method optimizes sensor placement and improves monitoring efficiency.

[0073] Specifically, when using the K-means clustering algorithm to group the candidate locations to obtain the optimal location cluster, the five candidate locations are clustered into two groups based on their geographical coordinates and gradient values.

[0074] For example, three points close to the pollution source are grouped together, and two points further away are grouped together. The center point of each group is then selected as the optimal location, such as coordinates (100, 200) and (150, 250). This method ensures that sensor deployment covers key areas and improves the representativeness of data collection.

[0075] Furthermore, the process of obtaining the extended converged network includes: Feature vectors are extracted from incremental sensor data using an adaptive weighting algorithm; The similarity matrix with the original network association rules is calculated based on the feature vectors to obtain the matching score; If the matching score is higher than the preset threshold, the weighted average method is used to fuse the incremental sensor data and the original network association rules to obtain a preliminary fusion structure. The K-means clustering algorithm was applied to the preliminary fusion structure to divide the data into clusters, and the set of cluster centroids was obtained. The optimized weight vector is obtained by adjusting the parameters of the adaptive weighting algorithm based on the cluster center point set; The incremental sensor data and cluster center point set are re-fused using optimized weight vectors to generate an extended fusion network.

[0076] For example, in the lake hydrological monitoring scenario of this embodiment, an adaptive weighting algorithm is used to extract feature vectors from incremental sensor data. The data collected by the sensors includes hydrological parameters such as water temperature, turbidity, and dissolved oxygen, which exhibit complex distributions due to temporal and spatial differences. The adaptive weighting algorithm dynamically adjusts the weights of each parameter by analyzing the correlation between data points, thereby extracting key feature vectors.

[0077] For example, when the water temperature in a certain area of ​​a lake rises abnormally, the algorithm will assign higher weight to water temperature, generating a feature vector that includes trends in water temperature and turbidity. Such feature vectors can reflect the dynamic changes in the hydrological environment, facilitating subsequent analysis.

[0078] Specifically, this embodiment uses the cosine similarity method to calculate the similarity matrix between the feature vector and the original network association rules. Assuming the original network association rules are based on historical data and describe the association pattern between water temperature and dissolved oxygen, feature vectors are extracted and compared with these rules to generate a matching score.

[0079] For example, the matching score of a feature vector from incremental sensor data is 0.85, which is higher than the preset threshold of 0.8, indicating that the new data is highly consistent with historical patterns. This step ensures the reliability of the new data and helps with accurate fusion.

[0080] In one possible implementation, when fusing incremental sensor data and existing network association rules, this embodiment uses a weighted average method to construct a preliminary fusion structure.

[0081] For example, new sensor data from a monitoring point in a lake shows abnormal turbidity. By using a weighted average, the new data is fused with existing rules to generate a preliminary fusion structure that includes parameters such as turbidity and flow velocity. This structure preserves the stability of historical data while incorporating the real-time nature of new data, providing a foundation for subsequent clustering.

[0082] For example, in this embodiment, when applying the K-means clustering algorithm to divide the data into clusters, the set of cluster centroids is determined based on the characteristic distribution of the preliminary fusion structure. Assume the lake monitoring data is divided into three clusters, representing normal, slightly anomaly, and severely anomaly areas, respectively. The centroid set of each cluster reflects the hydrological characteristics of the corresponding area, such as mean turbidity and flow direction. These centroid sets provide a basis for optimizing weights, ensuring that subsequent analysis is closer to the actual environment.

[0083] Specifically, in this embodiment, when adjusting the parameters of the adaptive weight algorithm, the weight vector is optimized based on the distribution characteristics of the cluster center point set.

[0084] For example, if the center point of a cluster shows significant changes in turbidity, the algorithm will increase the weight of the turbidity parameter to generate an optimized weight vector. This vector can more accurately reflect the trend of pollution diffusion and improve the targeting of data analysis.

[0085] In one possible implementation, incremental sensor data and cluster centroid sets are re-fused using optimized weight vectors to generate an extended fusion network.

[0086] For example, when newly added sensor data for a lake is fused with the cluster centroid set, a network structure is generated that includes pollution diffusion paths and water flow directions. This structure can more comprehensively describe the lake's hydrological state and facilitate the monitoring of pollution source dynamics.

[0087] For example, in this embodiment, when verifying the consistency of association rules in the extended fusion network, an integrity metric is used for evaluation.

[0088] For example, integrity metrics include rule coverage and data consistency. If a network structure has a rule coverage of 90% and a data consistency rate higher than 0.9, it indicates that the fused network is reliable. This verification ensures the accuracy of the monitoring system and provides strong support for lake water quality management.

[0089] Furthermore, the process of obtaining the comprehensive anomaly index includes: Data sources are obtained based on extended fusion networks, and key factors are extracted through principal component analysis to obtain a preliminary factor set; A multi-parameter coupling method is used to fuse relevant attributes in the initial factor set to obtain a coupling parameter set; Calculate the abnormal deviation for the coupling parameter group. If the abnormal deviation exceeds the preset threshold, an abnormal event is identified and an abnormality is identified. Based on the anomaly identifier, retrieve the associated data from the extended converged network to obtain the extended data stream; The random forest algorithm is used to process the extended data stream, determine the data patterns, and obtain the pattern classification results; The pattern classification results are compared with the key factors. If the matching degree is higher than the threshold, the index component is fused to obtain the updated index value. The updated index values ​​were verified using a support vector machine algorithm to obtain the final anomaly index.

[0090] Furthermore, the process of obtaining a spatiotemporal scale anomaly distribution map and generating a visual heatmap includes: Raw monitoring data of lake sub-regions are obtained through a pre-established database, and time series analysis is used to determine the abnormal index sequence within a time period. The abnormal index sequence is judged. If there is a continuous upward trend in the sequence, it is marked as a high abnormal segment, and the set of marked abnormal segments is obtained. Spatiotemporal scale features are extracted from the marked set of anomalous paragraphs, and K-means clustering algorithm is used to group the spatiotemporal scale features to obtain the grouped spatiotemporal anomalous clusters. The distribution density of the grouped spatiotemporal anomaly clusters is calculated, and the density distribution map is determined by a gridded mapping method. Obtain the peak points in the density distribution map. If the peak points exceed the preset threshold, delineate the core abnormal regions and obtain a list of core abnormal regions. Select the dominant region from the list of core anomaly regions, and use an interpolation algorithm to generate a continuous surface to obtain a continuous anomaly surface; A color gradient mapping is constructed based on continuous anomalous surfaces to generate a visual heatmap.

[0091] In one embodiment, the original monitoring data of a lake sub-region is obtained through a pre-established database. This database is similar to the structure of an extended fusion network and is used to store historical records of multi-source water quality indicators such as pH, dissolved oxygen, and temperature.

[0092] For example, when monitoring the Taihu Lake sub-region, extracting daily water temperature data sequences from the past year from the database helps to quickly integrate key factors, avoid data loss, and thus improve the accuracy of anomaly detection.

[0093] Specifically, time series analysis is used to determine the abnormal index sequence within a time period. This method calculates the deviation based on historical trends. For example, the water quality index within a month is smoothed to generate an upward sequence of values ​​such as 0.2, 0.5, and 0.8. This can reveal potential pollution trends and is beneficial for early warning of water body changes.

[0094] In one embodiment, if there is a continuous upward trend in the abnormal index sequence, it is marked as a high-abnormal segment, and a set of marked abnormal segments is obtained. For example, if the index rises from 0.3 to 0.7 for three consecutive days in the sequence, it is marked as a high-abnormal segment. This supports the continuity of anomaly identification from the time dimension and improves the reliability of monitoring.

[0095] Specifically, spatiotemporal scale features, such as the location coordinates and duration of the anomaly, are extracted from the set of marked abnormal segments. For example, the location features of points with latitude and longitude of 31.2°N and 120.1°E are extracted. This is similar to multi-parameter coupling, which integrates spatiotemporal attributes to enhance the depth of analysis and is beneficial for accurately locating pollution sources.

[0096] In one embodiment, the K-means clustering algorithm is used to group the features to obtain spatiotemporal anomaly clusters. For example, features similar to pH anomalies are divided into three clusters. This supports the classification of anomaly patterns from multiple perspectives and optimizes resource allocation.

[0097] Specifically, this embodiment calculates the distribution density of spatiotemporal anomaly clusters after grouping. For example, the average density of points within each cluster is calculated to be 5 anomaly points per square kilometer. This helps to quantify the concentration of anomalies and improve decision-making efficiency.

[0098] In one embodiment, the density distribution map is determined by a gridded mapping method. For example, the lake area is divided into 10x10 grids, and the mapped density values ​​are from low to high. This corresponds to the extraction of key factors by principal component analysis, which simplifies the visualization of complex data and is beneficial for intuitively identifying hotspots.

[0099] Specifically, this embodiment obtains peak points in the density distribution map. If a peak point exceeds a preset threshold, such as 1.5, a core abnormal region is defined, and a list of core abnormal regions is obtained. For example, points with a peak value of 2.0 are included in the list. This supports the scientific nature of the region division from a density perspective and reduces misjudgments.

[0100] In one embodiment, the dominant region is selected from the core abnormal region list, and a continuous surface is generated using an interpolation algorithm to obtain a continuous abnormal surface. For example, Kriging interpolation is used to extend discrete points into a smooth surface. This integrates correlated data, improves the continuity of the abnormal index, and is beneficial for a comprehensive assessment of the impact range.

[0101] Specifically, this embodiment constructs a color gradient mapping based on continuous anomalous surfaces to generate a heatmap representation, for example, from blue (low anomaly) to red (high anomaly). This is similar to the pattern classification of the random forest algorithm, providing a visualization tool, supporting multi-parameter decision-making, and significantly improving the effectiveness of anomaly management.

[0102] Example 2 like Figure 2 As shown, based on the same inventive concept, this embodiment also provides a lake water quality anomaly monitoring system based on multi-source sensor data fusion, including: The multi-source sensor data acquisition and weighted fusion module is used to acquire multi-source sensor data based on multi-source sensors, calculate the weight coefficient of each data through a weighted fusion algorithm, and obtain preliminary fused data. The Kalman filtering module is used to process outliers, perform initial screening, and align timestamps based on the preliminary fused data to obtain a high-quality fused dataset. The dynamic threshold adjustment module is used to adjust the range of various water quality indicators through a dynamic threshold model to obtain the anomaly judgment threshold if the high-quality fused dataset exceeds the preset seasonal baseline. The incremental sensor deployment optimization module is used to obtain the distribution and flow direction of pollution sources in lake hydrological characteristics based on the anomaly detection threshold, and to determine the optimal deployment location of incremental sensors. The adaptive weight adjustment module is used to adjust the association rules between incremental sensor data and the original network through an adaptive weight algorithm to obtain an extended fusion network. The principal component analysis and multi-parameter coupling module is used to extract key factors based on the extended fusion network and perform multi-parameter coupling to obtain a comprehensive anomaly index. The spatiotemporal anomaly distribution visualization module is used to segment and process data of lake sub-regions and time periods based on the comprehensive anomaly index, obtain spatiotemporal scale anomaly distribution maps, and generate visualized heat maps.

[0103] The lake water quality anomaly monitoring system based on multi-source sensor data fusion provided in this embodiment has all the advantages of the lake water quality anomaly monitoring method based on multi-source sensor data fusion provided in Embodiment 1.

[0104] Example 3 This embodiment also discloses a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method described in Embodiment 1.

[0105] Example 4 This embodiment also discloses a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the method described in Embodiment 1.

[0106] Example 5 This embodiment also discloses a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described in Embodiment 1.

[0107] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A lake water quality anomaly monitoring method based on multi-source sensor data fusion, characterized in that, The method comprises the following steps: Based on multi-source sensor to obtain multi-source sensor data, through weighted fusion algorithm to calculate the weight coefficient of each data, to obtain the preliminary fusion data; According to the preliminary fusion data, using Kalman filter to process abnormal value, preliminary screening and timestamp alignment, to obtain high quality fusion data set; If the high quality fusion data set exceeds the preset seasonal baseline, adjust the range of each water quality index through the dynamic threshold model, to obtain the abnormal judgment threshold; According to the abnormal judgment threshold, the distribution of pollution sources of lake hydrological characteristics is obtained, and the optimal layout position of incremental sensor is determined; Through the self-adaptive weight algorithm, the data of the incremental sensor and the original network association rule are adjusted, and the extended fusion network is obtained; Based on the extended fusion network, the key factors are extracted by principal component analysis and multi-parameter coupling, and the comprehensive abnormal index is obtained; According to the comprehensive abnormal index, the data of the sub-regions and time periods of the lake are segmented and processed, and the abnormal distribution graph of time and space scale is obtained, and the visual heat map is generated.

2. The method of claim 1, wherein the process of obtaining preliminary fusion data comprises: Obtaining multi-source sensor data, including dissolved oxygen data, turbidity data, pH value data, temperature data, removing abnormal values through preprocessing to obtain cleaned data set; According to the cleaned data set, Kalman filter is used to smooth the dissolved oxygen data, turbidity data, pH value data and temperature data to obtain smoothed data set; If the deviation of dissolved oxygen data and temperature data in the smoothed data set exceeds the preset threshold, mark the abnormal point to obtain the marked data set; According to the marked data set, the correlation between the turbidity data and the pH value data is judged, and the linear regression model is used to fit the relationship to obtain the correlation parameter set; The correlation parameter set is calculated by weighted fusion algorithm to obtain the weight coefficient set of dissolved oxygen data, turbidity data, pH value data and temperature data; Based on the weight coefficient set, the dissolved oxygen data, turbidity data, pH value data and temperature data in the smoothed data set are fused to obtain the preliminary fusion data; The preliminary fusion data is reduced by principal component analysis, and the main components are retained to obtain the reduced fusion data.

3. The method of claim 1, wherein the process of obtaining high quality fusion data set comprises: Based on the preliminary fusion data, Kalman filter is used to predict state vector to obtain preliminary state estimation value; Compare the preliminary state estimation value with the observation value, if it exceeds the preset threshold, it is judged as abnormal value and excluded, to obtain clean state data set; Based on the clean state data set, the timestamp sequence is extracted, and the timestamp is aligned by linear interpolation to obtain aligned state data set; The covariance matrix is updated by applying Kalman filter to the aligned state data set to obtain fusion state vector; Based on the fusion state vector, the residual sequence is calculated, if the residual sequence variance converges, the high quality index is determined, and the screening fusion data set is obtained; The screening fusion data set is divided into training subset and verification subset, and the random forest model is used to train the prediction error to judge the value exclusion accuracy. ​ ​ The value screening accuracy rate is combined with the punch alignment error to obtain a final high-quality fusion data set through weighted average fusion.

4. The method of claim 1, wherein, The process of obtaining the anomaly determination threshold value comprises: Based on the high-quality fusion data set, a data set quality score is calculated, and if the data set quality score is higher than the seasonal baseline, a dynamic threshold model is activated to process the seasonal baseline data to obtain preliminary adjustment parameters; The preliminary adjustment parameters and water quality index historical records are fused using the dynamic threshold model to determine the index range boundary; The index range boundary is used to analyze the current value of the water quality index to determine the abnormal initial signal; If the abnormal initial signal exists, the model is adjusted to obtain an optimized threshold model; The K-means clustering algorithm is used to group the index range in the optimized threshold model to obtain an anomaly determination threshold value cluster; The final anomaly determination threshold value is obtained by verifying the change of the fusion data set through the anomaly determination threshold value cluster.

5. The method of claim 1, wherein, The process of determining the optimal layout position of the incremental sensor comprises: Obtain lake hydrological feature data and pollution source distribution data to generate a first data set; Determine the abnormal area range by using the anomaly determination threshold value on the first data set; Calculate the pollution diffusion path in the abnormal area range using the water flow direction vector; If the pollution diffusion path covers the source point position, determine the affected hydrological feature point set; Obtain the distribution data density gradient from the affected hydrological feature point set; Determine the incremental sensor layout position candidate set according to the distribution data density gradient; Use the K-means clustering algorithm to group the layout position candidate set to obtain an optimal point cluster.

6. The method of claim 1, wherein, The process of obtaining the extended fusion network comprises: Extract a feature vector from the incremental sensor data through an adaptive weight algorithm; Calculate a similarity matrix of the associated rules of the original network according to the feature vector to obtain a matching degree score; If the matching degree score is higher than a preset threshold, fuse the incremental sensor data and the associated rules of the original network using a weighted average method to obtain a preliminary fusion structure; Apply the K-means clustering algorithm to the preliminary fusion structure to divide data clusters and obtain a cluster center point set; Adjust the parameters of the adaptive weight algorithm according to the cluster center point set to obtain an optimized weight vector; Re-fuse the incremental sensor data and the cluster center point set using the optimized weight vector to generate an extended fusion network.

7. The method of claim 1, wherein, The process of obtaining the comprehensive anomaly index comprises: Based on the extended fusion network, obtain data sources, extract key factors through principal component analysis to obtain a preliminary factor set; Fuse related attributes of the preliminary factor set using a multi-parameter coupling method to obtain a coupled parameter group; Calculate the anomaly deviation of the coupled parameter group, and if the anomaly deviation exceeds a preset threshold, determine an abnormal event to obtain an anomaly identifier; According to the anomaly identifier, obtain associated data from the extended fusion network to obtain an extended data stream; Adopting a random forest algorithm to process the extended data stream, judging a data mode, and obtaining a mode classification result; Comparing the mode classification result with a key factor, if a matching degree is higher than a threshold value, fusing a pointing component, and obtaining an updated index value; Verifying the updated index value by adopting a support vector machine algorithm, and obtaining a final abnormal index.

8. The method of claim 1, wherein The process of obtaining a spatiotemporal scale abnormal distribution diagram and generating a visualized heat map comprises: Obtaining original monitoring data of a sub-region of a lake through a pre-established database, and determining an abnormal index sequence in a time period by adopting a time series analysis method; Judging the abnormal index sequence, if there is a continuous rising trend in the sequence, marking it as a high abnormal paragraph, and obtaining a marked abnormal paragraph set; Extracting spatiotemporal scale features from the marked abnormal paragraph set, grouping the spatiotemporal scale features by adopting a K-means clustering algorithm, and obtaining a grouped spatiotemporal abnormal cluster; Calculating a distribution density of the grouped spatiotemporal abnormal cluster, and determining a density distribution diagram by a gridding mapping method; Obtaining peak points in the density distribution diagram, if the peak points exceed a preset threshold value, dividing out a core abnormal region, and obtaining a core abnormal region list; Selecting a dominant region from the core abnormal region list, generating a continuous surface by adopting an interpolation algorithm, and obtaining a continuous abnormal surface; Constructing a color gradient mapping according to the continuous abnormal surface, and generating a visualized heat map.

9. A lake water quality anomaly monitoring system based on multi-source sensor data fusion, characterized in that, Comprises: A multi-source sensor data acquisition and weighted fusion module, configured to acquire multi-source sensor data based on multi-source sensors, calculate data weight coefficients by a weighted fusion algorithm, and obtain preliminary fusion data; A Kalman filter processing module, configured to process abnormal values, preliminary screening, and timestamp alignment by adopting a Kalman filter according to the preliminary fusion data, and obtain a high-quality fusion data set; A dynamic threshold adjustment module, configured to adjust the range of each water quality index by a dynamic threshold model if the high-quality fusion data set exceeds a preset seasonal baseline, and obtain an abnormal judgment threshold value; An incremental sensor deployment optimization module, configured to obtain a lake hydrological feature pollution source distribution water flow direction according to the abnormal judgment threshold value, and determine an optimal deployment position of an incremental sensor; An adaptive weight adjustment module, configured to adjust the incremental sensor data and original network association rules by an adaptive weight algorithm, and obtain an extended fusion network; A principal component analysis and multi-parameter coupling module, configured to extract key factors by a principal component analysis and perform multi-parameter coupling based on the extended fusion network, and obtain a comprehensive abnormal index; A spatiotemporal abnormal distribution visualization module, configured to perform data segmentation processing on a lake sub-region and a time period according to the comprehensive abnormal index, obtain a spatiotemporal scale abnormal distribution diagram, and generate a visualized heat map.

Images