A contaminated site remediation progress management method integrating multi-source data

Abstract

The present invention provides a contaminated site remediation progress management method that integrates multi-source data, including: collecting drone aerial survey data and real-time soil remediation data of the contaminated site, and performing semantic association to obtain a semantically fused multi-source data set; analyzing the association relationships between the data in the multi-source data set based on an association rule algorithm, combining calculations with the remediation progress correlation, and screening a key factor set based on the association relationships and correlations; inputting the data in the key factor set into a pre-trained remediation progress prediction model to perform real-time progress prediction and output a predicted value for the current remediation progress of the contaminated site; performing a deviation analysis between the predicted value and the actual remediation progress value uploaded by the construction party, calculating the error between the predicted value and the actual value, analyzing the changing trend of the error using a sliding window algorithm, and judging whether the remediation progress is abnormal based on the changing trend. In this invention, the drawback of inaccurate remediation progress prediction is overcome.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a contaminated site remediation progress management method that integrates multi-source data. Background Art

[0002] Traditional contaminated site remediation progress management often relies on a single data source, such as manual records from construction workers or a limited amount of soil testing data for progress assessment. This approach has numerous drawbacks: manual records are susceptible to subjective factors, making it difficult to ensure data accuracy and timeliness; and a single type of data cannot fully reflect the complexities of the remediation process, such as the migration patterns of soil contaminants and changes in site topography, leading to significant deviations in progress forecasts.

[0003] At the same time, when analyzing the relationship between data and repair progress, it is impossible to accurately screen out the key factors that truly affect the repair progress, resulting in poor accuracy of the prediction model.

[0004] Therefore, the current method has the defect of inaccurate repair progress prediction. Summary of the Invention

[0005] The main purpose of the present invention is to provide a contaminated site remediation progress management method that integrates multi-source data, aiming to overcome the defect of inaccurate prediction of remediation progress in current practice.

[0006] To achieve the above objectives, the present invention provides a contaminated site remediation progress management method integrating multi-source data, comprising the following steps:

[0007] Collect drone aerial survey data and real-time soil remediation data of contaminated sites, perform semantic association, and obtain a semantically fused multi-source data set;

[0008] Based on the association rule algorithm, the association relationship between each data in the multi-source data set is analyzed, and the correlation between the calculation and the repair progress is combined to screen out the key factor set according to the association relationship and the correlation;

[0009] Inputting the data in the key factor set into a pre-trained remediation progress prediction model to perform real-time progress prediction, and outputting a predicted value of the current remediation progress of the contaminated site;

[0010] A deviation analysis is performed on the predicted value and the actual value of the repair progress uploaded by the construction party, the error between the predicted value and the actual value is calculated, the error change trend is analyzed by the sliding window algorithm, and whether there is any abnormality in the repair progress is determined based on the change trend.

[0011] Furthermore, semantic association is performed to obtain a semantically fused multi-source data set, including:

[0012] Feature extraction is performed on UAV aerial survey data and real-time soil remediation data to obtain aerial survey data features and soil data features;

[0013] The aerial survey data features and soil data features are projected into a unified semantic space, and the semantic space is aligned by minimizing the distribution difference between the modalities to obtain a semantically fused multi-source data set.

[0014] Furthermore, based on the association rule algorithm, the association relationship between each data in the multi-source data set is analyzed, and the correlation between each data and the repair progress is calculated. According to the association relationship and the correlation, a set of key factors is screened out, including:

[0015] Discretizing the multi-source data set to obtain a formatted data set suitable for association rule mining;

[0016] Performing association mining on the formatted data set based on the Apriori algorithm to generate frequent item sets, generating association relationships based on the frequent item sets, calculating the support, confidence, and lift of each association relationship, and obtaining an association relationship set;

[0017] Based on the preset support threshold, confidence threshold and lift threshold, the set of association relationships is screened to filter out association relationships that do not meet the conditions and obtain valid association relationships;

[0018] For each valid association relationship, calculate the mutual information between the association relationship and the restoration progress data, and evaluate the correlation between the valid association relationship and the restoration progress based on the mutual information;

[0019] Based on the preset correlation threshold, data items that reach the correlation threshold are extracted from the effective association relationships to construct a set of key factors.

[0020] Furthermore, the error variation trend is analyzed by a sliding window algorithm, and based on the variation trend, whether the repair progress is abnormal is determined, including:

[0021] The error between the predicted value and the actual value is divided into sliding windows, and each sliding step generates a window data set containing continuous error values; the mean, standard deviation, range and coefficient of variation of the error in each window data set are calculated to construct a window statistical feature vector;

[0022] The ARIMA model is used to fit the changing trend of the error within the window to obtain the trend prediction model and its parameters; the trend prediction model parameters of adjacent windows are analyzed for differences, the parameter change rate and relative change amount are calculated, and the parameter change feature vector is constructed;

[0023] Compare the parameter change feature vector of the current window with the confidence interval. If it exceeds the confidence interval, it is determined that the window has an abnormal trend. Perform multi-window consistency verification on the window with abnormal trend to eliminate the single window fluctuation noise.

[0024] The direction and magnitude of the error change are determined based on the error change trend within the window, and the abnormal trend after consistency verification is typed to obtain the abnormal type.

[0025] Furthermore, the drone aerial survey data includes video data taken by the drone; the soil remediation real-time data includes pollutant concentration data, soil physical property data, and remediation equipment operation data.

[0026] Furthermore, drone aerial survey data and real-time soil remediation data of contaminated sites are collected, including:

[0027] The spatial distribution characteristics of the contaminated site remediation area were collected, and the flight window was predicted based on historical meteorological data. A genetic algorithm was used to optimize the UAV flight path to generate an aerial survey plan that included shooting missions at different time intervals and from multiple angles.

[0028] When executing the aerial survey plan, the exposure parameters of the camera on the drone are automatically adjusted according to the real-time light intensity, the laser radar scanning resolution is dynamically calibrated based on the flight altitude, and the acquisition timestamp and geographic coordinate information are synchronously recorded when the aerial survey data is collected;

[0029] Real-time soil remediation data is collected based on an Internet of Things system deployed at a contaminated site; wherein the Internet of Things system includes soil sensors connected to the network and remediation equipment monitoring terminals.

[0030] Furthermore, semantic association is performed to obtain a semantically fused multi-source data set, including:

[0031] Extracting the terrain semantic features of the contaminated site and the semantic features of the spatial layout of the remediation equipment from the aerial survey data to construct a three-dimensional spatial semantic feature vector;

[0032] Extracting pollutant concentration change trend semantics and remediation process execution status semantics from the real-time soil remediation data through a bidirectional long short-term memory network to generate a temporal semantic feature sequence;

[0033] The three-dimensional spatial semantic feature vector and the temporal semantic feature sequence are mapped to a unified semantic space, and the semantic alignment of the data is performed by minimizing the distribution differences between the modalities to form a semantically fused multi-source data set; wherein, when the repair progress data is updated in real time, the reinforcement learning algorithm is used to dynamically adjust the semantic association weights between different modal data.

[0034] The present invention also provides a contaminated site remediation progress management device integrating multi-source data, comprising:

[0035] The acquisition module is used to collect drone aerial survey data and real-time soil remediation data of contaminated sites, and perform semantic association to obtain a semantically fused multi-source data set;

[0036] A screening module is used to analyze the association relationship between each data in the multi-source data set based on an association rule algorithm, and to screen out a set of key factors based on the association relationship and the correlation, in combination with the correlation between calculation and repair progress;

[0037] A prediction module is used to input the data in the key factor set into a pre-trained remediation progress prediction model to perform real-time progress prediction and output a predicted value of the current remediation progress of the contaminated site;

[0038] The management module is used to perform deviation analysis between the predicted value and the actual value of the repair progress uploaded by the construction party, calculate the error between the predicted value and the actual value, analyze the changing trend of the error through the sliding window algorithm, and judge whether there is any abnormality in the repair progress based on the changing trend.

[0039] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0040] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of any of the above methods are implemented.

[0041] The present invention provides a contaminated site remediation progress management method that integrates multi-source data, including: collecting drone aerial survey data and real-time soil remediation data of the contaminated site, and performing semantic association to obtain a semantically fused multi-source data set; analyzing the association relationship between each data in the multi-source data set based on an association rule algorithm, combining the calculation with the remediation progress correlation, and filtering out a key factor set based on the association relationship and the correlation; inputting the data in the key factor set into a pre-trained remediation progress prediction model to perform real-time progress prediction and output a predicted value of the current remediation progress of the contaminated site; performing a deviation analysis between the predicted value and the actual remediation progress value uploaded by the construction party, calculating the error between the predicted value and the actual value, analyzing the error change trend through a sliding window algorithm, and judging whether the remediation progress is abnormal based on the change trend. In the present invention, by collecting drone aerial survey data and real-time soil remediation data of the contaminated site and performing semantic association; and filtering out a key factor set based on an association rule algorithm, the defects of the current single data source and the failure to accurately filter out the key factors that truly affect the remediation progress, resulting in inaccurate remediation progress prediction, are overcome. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a schematic diagram of the steps of a contaminated site remediation progress management method integrating multi-source data in one embodiment of the present invention;

[0043] Figure 2 This is a structural block diagram of a contaminated site remediation progress management device integrating multi-source data in one embodiment of the present invention;

[0044] Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0045] The implementation, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0047] Reference Figure 1 In one embodiment of the present invention, a method for managing the remediation progress of a contaminated site by integrating multi-source data is provided, comprising the following steps:

[0048] Step S1: collecting drone aerial survey data and real-time soil remediation data of the contaminated site, and performing semantic association to obtain a semantically fused multi-source data set;

[0049] Step S2: Analyze the association relationship between each data in the multi-source data set based on the association rule algorithm, combine the calculation with the correlation of the repair progress, and select the key factor set according to the association relationship and the correlation;

[0050] Step S3: inputting the data in the key factor set into a pre-trained remediation progress prediction model to perform real-time progress prediction, and outputting a predicted value of the current remediation progress of the contaminated site;

[0051] Step S4: perform deviation analysis on the predicted value and the actual value of the repair progress uploaded by the construction party, calculate the error between the predicted value and the actual value, analyze the change trend of the error through a sliding window algorithm, and judge whether there is any abnormality in the repair progress based on the change trend.

[0052] In this embodiment, as described in step S1 above, during the contaminated site remediation work, specialized equipment such as high-resolution cameras and lidars, such as drones, are first used to conduct a comprehensive, multi-angle aerial survey of the site, acquiring spatial data such as the site's topography, remediation area boundaries, and equipment distribution. Simultaneously, various sensors are deployed within the site to collect real-time data on the soil remediation process, including information such as soil contaminant concentrations, pH, and the operating parameters of remediation equipment. Because this data comes from different sources and has varying formats and semantics, to enable the data to function synergistically, a contaminated site domain ontology model is constructed, utilizing natural language processing and knowledge graphs to identify key entities and relationships within the data. The spatial information in the drone aerial survey data is semantically associated and mapped with the process information in the real-time soil remediation data. This ultimately forms a semantically fused multi-source data set, laying a unified data foundation for subsequent analysis.

[0053] As described in step S2 above, after obtaining the semantically fused multi-source data set, use association rule algorithms, such as the Apriori algorithm and its improved algorithm, to conduct in-depth data mining to find frequently occurring combination patterns between data items, so as to determine the intrinsic correlation between the data, such as the correlation between the change in concentration of a certain pollutant and the operating time of a specific repair equipment. At the same time, by calculating indicators such as the Pearson correlation coefficient and mutual information, the correlation between each data and the repair progress is quantified, and the degree of influence of each data on the repair progress is measured. Finally, considering the strength of the association relationship and the size of the correlation, a reasonable screening threshold is set to screen out those data items that are closely related to the repair progress and have close connections with each other to form a set of key factors. The above key factors will play a key role in predicting the repair progress.

[0054] As described in step S3 above, a remediation progress prediction model is constructed using historical multi-source data and a deep learning model, such as a hybrid network structure combining a convolutional neural network (CNN) and a long short-term memory network (LSTM), or a Transformer architecture. During the training process, model parameters are continuously adjusted to enable the model to learn the complex mapping relationship between key factors and remediation progress. Once the current set of key factor data is obtained, it is input into the trained model. Based on the learned patterns and rules, the model makes a real-time prediction of the current remediation progress of the contaminated site and outputs the predicted value, providing data support for progress control of the remediation project.

[0055] As described in step S4 above, the predicted value of the repair progress output by the model is compared with the actual value of the repair progress actually recorded and uploaded by the construction party, and the error between the two is calculated. The error value can intuitively reflect the accuracy of the prediction. In order to more accurately judge whether the repair progress is abnormal, a sliding window algorithm is introduced to divide the error data into multiple windows of fixed length in chronological order, calculate the statistical characteristics of the error in each window, such as the mean, standard deviation, etc., and analyze the changing trend of the error in different time windows. If the changing trend of the error exceeds the pre-set normal range, for example, the error continues to increase and exceeds a certain threshold, or abnormal fluctuations occur, it can be determined that the repair progress is abnormal, so that corresponding measures can be taken in time to adjust and optimize, to ensure the smooth progress of the repair project.

[0056] In an embodiment of the present invention, drone aerial survey data and real-time soil remediation data of the contaminated site are collected and semantically associated; at the same time, a set of key factors is screened out based on an association rule algorithm, thereby overcoming the current defects of inaccurate prediction of remediation progress caused by a single data source and the failure to accurately screen out the key factors that truly affect the remediation progress.

[0057] In one embodiment, semantic association is performed to obtain a semantically fused multi-source data set, including:

[0058] Feature extraction is performed on UAV aerial survey data and real-time soil remediation data to obtain aerial survey data features and soil data features;

[0059] The aerial survey data features and soil data features are projected into a unified semantic space, and the semantic space is aligned by minimizing the distribution difference between the modalities to obtain a semantically fused multi-source data set.

[0060] In this example, drone aerial survey data contains rich spatial information, such as site topography, remediation area outlines, and equipment distribution, while real-time soil remediation data focuses on physical and chemical changes during the remediation process, such as fluctuations in pollutant concentrations, soil temperature and humidity indicators, and remediation equipment operating parameters. Due to the large amount of raw data and heterogeneous formats, direct processing is inefficient and difficult to discover internal correlations. Therefore, targeted technologies are needed to extract key features:

[0061] For drone aerial survey data, 3D point cloud processing algorithms (such as deep learning-based PointNet) are used to extract the 3D structural features of the site. Computer vision techniques are used to identify the shape and positional characteristics of objects such as remediation equipment and contaminated boundaries, thereby generating aerial survey data features. For real-time soil remediation data, time series analysis algorithms (such as LSTM and Transformer) are used to capture the temporal trends of pollutant concentrations. Signal processing techniques are used to extract abnormal fluctuations in equipment operating parameters to construct soil data features. This step achieves data dimensionality reduction through feature extraction, highlighting core information and laying the foundation for subsequent fusion.

[0062] Because aerial survey data and soil data belong to different modalities, namely spatial and temporal, their data distribution and semantic expressions differ significantly. Direct fusion can lead to information conflict. Therefore, a common semantic space is needed to enable the two types of data to interact within the same language. Specifically, a cross-modal mapping network (such as the Domain Adversarial Neural Network (DANN) based on adversarial learning) is constructed to project the features of the aerial survey data and soil data into the same high-dimensional space. During this projection process, a distribution alignment loss function (such as the Maximum Mean Difference (MMD)) is introduced to force the data distributions of the two modalities to be as similar as possible, thereby eliminating the semantic gap caused by the modal differences. For example, when the repair equipment movement features in the aerial survey data and the repair efficiency improvement features in the soil data are close in semantic space, an implicit semantic association is established. Ultimately, the aligned data form an organic whole in the unified semantic space, forming a semantically fused multi-source data set, providing structured and relevant data support for subsequent analysis.

[0063] In one embodiment, based on an association rule algorithm, the association relationships between the data in the multi-source data set are analyzed, and the correlation between the data and the repair progress is calculated. A set of key factors is screened out based on the association relationships and the correlation, including:

[0064] Discretizing the multi-source data set to obtain a formatted data set suitable for association rule mining;

[0065] Performing association mining on the formatted data set based on the Apriori algorithm to generate frequent item sets, generating association relationships based on the frequent item sets, calculating the support, confidence, and lift of each association relationship, and obtaining an association relationship set;

[0066] Based on the preset support threshold, confidence threshold and lift threshold, the set of association relationships is screened to filter out association relationships that do not meet the conditions and obtain valid association relationships;

[0067] For each valid association relationship, calculate the mutual information between the association relationship and the restoration progress data, and evaluate the correlation between the valid association relationship and the restoration progress based on the mutual information;

[0068] Based on the preset correlation threshold, data items that reach the correlation threshold are extracted from the effective association relationships to construct a set of key factors.

[0069] In this embodiment, screening the key factors that truly impact the progress of contaminated site remediation from massive multi-source data is a key step in achieving accurate prediction and effective management. This technical solution uses a series of rigorous and logically cohesive steps, based on association rule algorithms and correlation analysis, to scientifically identify key factors. The details are as follows:

[0070] Since the collected multi-source data sets contain a large amount of continuous data (such as the specific values ​​of soil pollutant concentrations, the operating time of remediation equipment, etc.) and categorical data (such as the type of pollutant, the type of remediation process, etc.), association rule algorithms (such as the Apriori algorithm) are generally more suitable for processing discrete data. Therefore, it is first necessary to discretize the multi-source data sets. For continuous data, discretization algorithms based on information entropy, such as equal-frequency binning, equal-width binning, or more complex clustering-based discretization methods, are used to divide the continuous numerical interval into several discrete intervals; for categorical data, appropriate encoding conversion (such as one-hot encoding) is performed. Through this processing, the original data is converted into a structured, discrete, formatted data set so that the data format meets the requirements of the association rule mining algorithm.

[0071] After obtaining a formatted data set, the classic Apriori algorithm is used to conduct in-depth data mining. The Apriori algorithm employs a layer-by-layer search strategy, starting with a single data item (a 1-itemset). By scanning the dataset, the frequency of occurrence of each item set is counted. Frequent 1-itemsets are then selected based on a pre-set minimum support threshold. Next, candidate 2-itemsets are generated from these frequent 1-itemsets through a concatenation operation. The dataset is scanned again to calculate their support, and these candidate 2-itemsets are selected. This process continues, generating and selecting higher-order frequent itemsets until no new frequent itemsets can be generated. These frequent itemsets reflect the patterns of combinations of data items that frequently co-occur in the dataset.

[0072] Based on the generated frequent item sets, we further generate associations (rules such as "If A occurs, then B occurs"). To assess the reliability and value of these associations, we calculate the support (indicates the frequency of transactions containing the antecedent and consequent of the association in the dataset, reflecting the prevalence of the rule), the confidence (indicates the proportion of transactions containing the antecedent that also contain the consequent, measuring the credibility of the rule), and the lift (indicates the ratio of the probability of the antecedent and consequent occurring simultaneously to the product of their independent probabilities, reflecting the actual effectiveness of the rule) for each association. These metrics comprehensively characterize each association, ultimately resulting in an association set containing all associations and their corresponding metrics.

[0073] Due to the large number of relationships generated during the relationship mining process, some of these relationships may appear by chance, lack practical significance, or have low reliability. Therefore, it is necessary to pre-set support thresholds, confidence thresholds, and lift thresholds based on actual application requirements and domain knowledge. Each relationship in the relationship set is compared with these thresholds. Only relationships that simultaneously meet the conditions of support greater than or equal to the support threshold, confidence greater than or equal to the confidence threshold, and lift greater than or equal to the lift threshold are considered to have practical value and reliability and are retained; relationships that do not meet these conditions are filtered out. Through this screening process, a large number of invalid or weakly associated relationships are removed, and the resulting valid relationship set focuses on those association patterns that appear frequently in the data, have high credibility, and have practical guiding significance.

[0074] Although the effective correlations identified through screening show strong correlations within the data, the relationship between these correlations and the remediation progress of the contaminated site is still unclear. To determine which correlations are truly relevant to remediation progress, further analysis of each effective correlation is required. Mutual information, an information-theoretic metric used to measure the degree of dependence between two random variables, is used here to assess the correlation between effective correlations and remediation progress data. Specifically, the data items in the effective correlation and the remediation progress data are treated as random variables, and the mutual information value between them is calculated to quantify the degree of correlation between the two. A higher mutual information value indicates a stronger correlation between the effective correlation and remediation progress; conversely, a lower mutual information value indicates a weaker correlation. In this way, a scientific and objective assessment of the correlation between effective correlations and remediation progress is conducted from an information-theoretic perspective.

[0075] After calculating the correlation between each valid correlation and the remediation progress, in order to further screen out data items that have a significant impact on the remediation progress, it is necessary to pre-set a correlation threshold based on the actual application scenario and accuracy requirements. The mutual information value of each valid correlation is compared with the threshold. Only when the mutual information value is greater than or equal to the correlation threshold, the data item in the corresponding valid correlation is considered to be closely related to the remediation progress and of great value to the remediation progress management. The data items that reach the correlation threshold are extracted and combined to form a set of key factors. The data items in the above-mentioned key factor set are the factors that have a real key impact on the remediation progress of the contaminated site, which are screened from massive multi-source data. They can be subsequently input into the remediation progress prediction model as core data to achieve more accurate progress prediction and more effective remediation progress management.

[0076] In one embodiment, analyzing the error variation trend using a sliding window algorithm and determining whether the repair progress is abnormal based on the variation trend includes:

[0077] The error between the predicted value and the actual value is divided into sliding windows, and each sliding step generates a window data set containing continuous error values; the mean, standard deviation, range and coefficient of variation of the error in each window data set are calculated to construct a window statistical feature vector;

[0078] The ARIMA model is used to fit the changing trend of the error within the window to obtain the trend prediction model and its parameters; the trend prediction model parameters of adjacent windows are analyzed for differences, the parameter change rate and relative change amount are calculated, and the parameter change feature vector is constructed;

[0079] Compare the parameter change feature vector of the current window with the confidence interval. If it exceeds the confidence interval, it is determined that the window has an abnormal trend. Perform multi-window consistency verification on the window with abnormal trend to eliminate the single window fluctuation noise.

[0080] The direction and magnitude of the error change are determined based on the error change trend within the window, and the abnormal trend after consistency verification is typed to obtain the abnormal type.

[0081] In this embodiment, in the management of the remediation progress of contaminated sites, accurately determining whether the remediation progress is abnormal is crucial to ensuring the smooth implementation of the project. In this embodiment, a sliding window algorithm is applied and trend analysis is performed to scientifically identify abnormalities in the remediation progress. The specific steps are as follows:

[0082] The error data between predicted and actual repair progress often exhibits dynamic fluctuations, making it difficult to capture patterns by directly analyzing the raw data. Therefore, a sliding window algorithm is first used to partition the error sequence. By setting a fixed window length and sliding step size, the continuous error values ​​are divided into multiple window data sets. Each window data set acts as a data "slice," containing information about error changes within a specific time period.

[0083] To better characterize the distribution and fluctuation characteristics of the error data within each window, the mean, standard deviation, range, and coefficient of variation of the error within the window are calculated. The mean reflects the average level of error, the standard deviation measures the dispersion of the error around the mean, the range reflects the maximum fluctuation range of the error, and the coefficient of variation eliminates the influence of dimension and comprehensively evaluates the relative degree of error fluctuation. These statistical indicators together constitute the window statistical feature vector, which transforms complex error data into a structured and analyzable feature description, providing a foundation for subsequent trend analysis.

[0084] Error data not only fluctuates but may also contain long-term trends and cyclical changes. To accurately capture these trends, the ARIMA (Autoregressive Integrated Moving Average) model is used to fit the error data within each window. By analyzing historical error data, the ARIMA model effectively extracts the autoregressive, differencing, and moving average characteristics of the data, constructing a forecasting model that reflects the error trend and determining model parameters (such as the autoregressive order p, the differencing order d, and the moving average order q).

[0085] Because changes in repair progress are a continuous process, the error trends between adjacent windows should exhibit a certain degree of consistency. Therefore, we perform a differential analysis of the ARIMA model parameters between adjacent windows, calculating the rate of parameter change and the relative amount of change. For example, if the autoregressive order p of a window changes dramatically between adjacent windows, this may indicate a sudden change in the error trend. This parameter change information forms a parameter change feature vector, which is used to quantify the dynamic evolution of the error trend over time.

[0086] To determine whether an error trend is abnormal, a reasonable evaluation standard must be established. Based on the statistical characteristics of historical normal data, a confidence interval for parameter changes is calculated and set. This interval represents the range of fluctuations in error trends under normal circumstances. The parameter change feature vector for the current window is compared with the confidence interval. If the parameter change exceeds this interval, an abnormal trend is preliminarily determined for the current window.

[0087] However, due to data fluctuations or accidental factors, single-window anomaly detection may lead to misjudgment. Therefore, a multi-window consistency verification mechanism is introduced. By analyzing the judgment results of multiple adjacent windows, a comprehensive judgment is made using strategies such as voting or weighting. For example, only if three consecutive windows are judged as anomalies can a true anomaly trend be confirmed for that period. This effectively eliminates the noise interference caused by single-window fluctuations and improves the accuracy and reliability of anomaly detection.

[0088] After confirming an abnormal trend, further analysis is conducted on the error trends within the window to determine the direction of the change (e.g., whether the error continues to increase, decrease, or fluctuate dramatically) and the magnitude (e.g., the specific value or percentage of the error change). Based on operational knowledge and historical experience in contaminated site remediation, rules are established to identify the type of anomaly. For example, if the error continues to increase and the magnitude exceeds a threshold, this is considered an abnormality of lagging remediation progress; if the error fluctuates dramatically without a clear pattern, this corresponds to an abnormality of unstable remediation process.

[0089] Through the above method, the abstract error change trend is converted into specific and explainable anomaly types, which not only achieves the accurate identification of abnormal situations in the remediation progress, but also provides clear guidance for subsequent targeted adjustment measures, thereby effectively ensuring the smooth progress of the contaminated site remediation project.

[0090] In one embodiment, the drone aerial survey data includes video data captured by the drone; the soil remediation real-time data includes pollutant concentration data, soil physical property data, and remediation equipment operation data.

[0091] In one embodiment, collecting drone aerial survey data and real-time soil remediation data of a contaminated site includes:

[0092] The spatial distribution characteristics of the contaminated site remediation area were collected, and the flight window was predicted based on historical meteorological data. A genetic algorithm was used to optimize the UAV flight path to generate an aerial survey plan that included shooting missions at different time intervals and from multiple angles.

[0093] When executing the aerial survey plan, the exposure parameters of the camera on the drone are automatically adjusted according to the real-time light intensity, the laser radar scanning resolution is dynamically calibrated based on the flight altitude, and the acquisition timestamp and geographic coordinate information are synchronously recorded when the aerial survey data is collected;

[0094] Real-time soil remediation data is collected based on an Internet of Things system deployed at a contaminated site; wherein the Internet of Things system includes soil sensors connected to the network and remediation equipment monitoring terminals.

[0095] In this embodiment, efficient collection of multi-source data is the foundation for subsequent analysis and decision-making in the management of contaminated site remediation progress. This technical solution uses a systematic and intelligent collection strategy to develop a scientific collection process for drone aerial survey data and real-time soil remediation data. The specific steps are as follows:

[0096] The spatial distribution of the contaminated site remediation area determines the coverage and key areas of drone aerial surveys. Historical meteorological data (such as wind speed, precipitation, and cloud thickness) directly impacts drone flight safety and data collection quality. To ensure the integrity and accuracy of the aerial survey data, Geographic Information System (GIS) technology is first used to comprehensively capture the spatial distribution characteristics of the remediation area, including key information such as topography, building distribution, and contamination boundaries. Furthermore, historical meteorological data is analyzed, and time series analysis models are used to predict future data collection windows suitable for drone flights, thereby minimizing the impact of inclement weather on aerial survey operations.

[0097] On this basis, a genetic algorithm was introduced to optimize the drone's flight path. This algorithm simulates the process of biological evolution, searching for the optimal solution within a complex path search space through operations such as selection, crossover, and mutation. Using comprehensive aerial survey coverage, minimal flight distance, and optimal shooting angles as objective functions, and using key monitoring points and no-fly zones within the restoration area as constraints, multiple rounds of iterative calculations generated an aerial survey plan encompassing shooting missions at varying time intervals and from multiple angles. For example, shorter shooting intervals and multi-angle shooting instructions were set for areas with complex terrain to ensure refined data collection and provide a rich spatial data foundation for subsequent restoration progress analysis.

[0098] When executing aerial surveys, real-time changes in environmental conditions can affect data collection quality. To ensure the reliability and consistency of aerial survey data, the drone's onboard acquisition equipment features adaptive adjustment capabilities. The camera automatically adjusts exposure parameters based on real-time light intensity. Using a built-in light sensor to monitor ambient lighting conditions in real time, an automatic exposure algorithm dynamically adjusts the aperture, shutter speed, and sensitivity to prevent overexposure or underexposure due to excessive or insufficient light, ensuring the clarity and color reproduction of aerial survey images.

[0099] As a key device for acquiring three-dimensional spatial information, lidar dynamically calibrates scanning resolution based on flight altitude. When the drone is flying at low altitude, the scanning resolution is automatically increased to capture highly accurate terrain details. At higher altitudes, the resolution is appropriately reduced to expand the scanning range, balancing data accuracy and acquisition efficiency. Furthermore, while collecting aerial survey data, the acquisition timestamp and geographic coordinate information are simultaneously recorded. Using the high-precision Global Positioning System (GPS) and Inertial Navigation System (INS), each data set is accurately labeled in time and space, facilitating subsequent spatiotemporal alignment and fusion analysis with other data.

[0100] The collection of real-time soil remediation data relies on an IoT system deployed at the contaminated site. This system establishes an efficient data collection network to comprehensively monitor the soil remediation process. The IoT system consists of soil sensors, remediation equipment monitoring terminals, and a communication network.

[0101] Soil sensors, serving as data collection nerve endings, are located at key locations across contaminated sites, enabling real-time sensing of soil physical and chemical properties. For example, pH sensors monitor changes in soil acidity and alkalinity, heavy metal sensors detect concentrations of pollutants such as mercury, cadmium, and lead, and temperature and humidity sensors record soil temperature and humidity. These sensors connect to the network via wireless or wired communication, transmitting the collected real-time data to a data center.

[0102] The remediation equipment monitoring terminal is responsible for collecting operational parameters of the remediation equipment, such as the processing flow rate of the soil washing equipment, the stirring speed of the bioremediation reactor, and the temperature control data of the heating equipment. By installing various sensors and intelligent controllers on the equipment, the equipment's operating status, energy consumption data, and fault information are collected in real time and uploaded to the IoT system. Ultimately, the data collected by the soil sensors and remediation equipment monitoring terminal is efficiently transmitted and stored under the unified management of the IoT system, providing rich process data support for real-time analysis and decision-making on the remediation progress of the contaminated site.

[0103] In one embodiment, semantic association is performed to obtain a semantically fused multi-source data set, including:

[0104] Extracting the terrain semantic features of the contaminated site and the semantic features of the spatial layout of the remediation equipment from the aerial survey data to construct a three-dimensional spatial semantic feature vector;

[0105] Extracting pollutant concentration change trend semantics and remediation process execution status semantics from the real-time soil remediation data through a bidirectional long short-term memory network to generate a temporal semantic feature sequence;

[0106] The three-dimensional spatial semantic feature vector and the temporal semantic feature sequence are mapped to a unified semantic space, and the semantic alignment of the data is performed by minimizing the distribution differences between the modalities to form a semantically fused multi-source data set; wherein, when the repair progress data is updated in real time, the reinforcement learning algorithm is used to dynamically adjust the semantic association weights between different modal data.

[0107] In this embodiment, drone aerial survey data contains rich site spatial information, but the raw data lacks structured semantic representation. To obtain effective semantic features, computer vision and 3D modeling technologies are used to deeply analyze the aerial survey data. Point cloud processing algorithms are used to identify topographic semantic features of the contaminated site, such as topographic undulations and slope direction, from the aerial survey point cloud data. Object detection algorithms (such as YOLO and Mask R-CNN) are used to locate remediation equipment and analyze layout semantic features, such as their spatial position and arrangement. The extracted semantic information about the terrain and equipment layout is quantized and encoded to construct a 3D spatial semantic feature vector. This vector not only preserves the site's spatial structure but also imparts clear semantic meaning to the data, providing a structured spatial data foundation for subsequent fusion.

[0108] Real-time soil remediation data exhibits typical time series characteristics, and the semantic information it contains changes dynamically over time. To capture this dynamic semantics, a bidirectional long short-term memory (BiLSTM) network is introduced for feature extraction. Using recurrent neural networks in both forward and reverse directions, BiLSTM can simultaneously learn past and future time series information, effectively addressing the inability of traditional unidirectional networks to utilize future data. When processing soil data, BiLSTM analyzes continuous monitoring data of pollutant concentrations to extract semantic information about changing trends, such as rising / falling concentration trends and fluctuation cycles. Simultaneously, it parses remediation process-related data (such as equipment operating parameters and chemical dosages) to identify semantic information about process execution status (e.g., normal operation, abnormal interruption). Ultimately, this semantic information is converted into a time series semantic feature sequence, enabling a semantic representation of the dynamic changes in the soil remediation process.

[0109] Because the three-dimensional spatial features of aerial survey data and the temporal features of soil data belong to different modalities and have a semantic gap, they need to be mapped into a unified semantic space for fusion. By constructing a cross-modal mapping network (such as a generative adversarial network (GAN) based on adversarial learning), the three-dimensional spatial semantic feature vectors and temporal semantic feature sequences are projected into the same high-dimensional semantic space. During this process, metrics such as the maximum mean difference (MMD) are introduced to minimize the distribution differences between the two modal data in the semantic space, ensuring semantic consistency. For example, when the "equipment movement trajectory" in the spatial features and the "repair efficiency change" in the temporal features are associated in the semantic space, the data are semantically aligned. Ultimately, the data from different modalities are organically integrated in the unified semantic space, forming a semantically fused multi-source data set.

[0110] Contaminated site remediation is a dynamic process, and the strength of semantic associations between data will also change with the progress of remediation. To accommodate this dynamic nature, a reinforcement learning algorithm is used to dynamically adjust the semantic association weights between different modal data as remediation progress data is updated in real time. This reinforcement learning algorithm constructs a "state-action-reward" model, using semantic association weights as the "action" and the accuracy of remediation progress predictions after data fusion as the "reward." During training, the algorithm continuously adjusts the weight combination, optimizing the strategy based on reward feedback to automatically identify the semantic associations that play a dominant role at different remediation stages. For example, in the early stages of remediation, the semantic weight of changes in soil contaminant concentrations may be higher; however, during the equipment commissioning phase, the semantic weight of the spatial layout of remediation equipment will be increased. This dynamic adjustment mechanism ensures that the semantically fused data always meets the needs of the actual remediation scenario, improving the accuracy of data-driven decision-making.

[0111] In one embodiment, after determining whether the repair progress is abnormal based on the change trend, the method includes:

[0112] If an exception occurs, obtain the exception type; determine the corresponding graph structure template based on the exception type; the graph structure template is an undirected graph, including multiple nodes, and the nodes are connected by undirected edges;

[0113] Rearranging the coding table based on the graph structure attributes of the graph structure template to obtain a new coding table;

[0114] Obtaining parameter information of the new coding table in which the variation occurs, and assigning serial numbers to each node of the graph structure template based on the parameter information;

[0115] Encoding the UAV aerial survey data and soil remediation real-time data based on the new coding table to obtain coded data;

[0116] Separating the characters in the encoded data in sequence, and inserting them into the nodes of the graph structure template in sequence according to the sequence numbers assigned to the nodes of the graph structure template, to obtain a data graph structure;

[0117] The data graph structure is sent to a management terminal of a contaminated site restoration entrusting party.

[0118] In this embodiment, after determining that the repair progress is abnormal, the technical solution constructs a data graph structure to express the complex abnormal information in a structured manner and efficiently transmit it to the client.

[0119] First, when an anomaly is detected in the repair progress based on the error change trend, the anomaly type must be identified, such as "delayed repair progress," "decreased efficiency due to equipment failure," and "soil pollutant treatment not meeting expectations." Different anomaly types correspond to different data association characteristics and influencing factors. In order to systematically sort out and present anomaly information, a mapping relationship library between anomaly types and graph structure templates is pre-established. Each graph structure template is an undirected graph consisting of multiple nodes and undirected edges. The nodes are used to carry data elements related to the anomaly (such as equipment parameters and pollutant indicators). For example, the "equipment failure" anomaly type corresponds to a graph structure template containing nodes such as "equipment model," "failure time," and "affected range." Through template matching, the visualization framework suitable for the current anomaly type is quickly determined.

[0120] The number of nodes, connection relationships and other properties of the graph structure template determine how the data is organized. In order to enhance the security of the data, the coding table needs to be rearranged. The coding table is a character representation rule for various types of information in drone aerial survey data and soil remediation real-time data. Based on the properties of the graph structure template, adjust the arrangement order, coding length and other parameters of the data items in the coding table. For example, it is possible to obtain the number of nodes in each row of the graph structure template, obtain the coding data of the corresponding serial number in the coding table based on the number of nodes in each row, and then align and rearrange them (including reordering, or adding suffixes, prefixes, etc.). There are many ways to rearrange the above, which will not be described in detail here. The purpose is to make the new coding table obtained by rearrangement unique and traceable, and enhance the security of data coding.

[0121] In the process of re-arranging the coding table, some parameter information may change (such as data type conversion, data range adjustment). After obtaining these mutated parameter information, a unique serial number is assigned to each node based on its degree of association with the graph structure template node. For example, the target serial number corresponding to the changed coding parameter in the new coding table can be obtained, and then the node with the same original serial number as the above target serial number can be found in the graph structure template as the target node, and the serial number is reallocated to the target node. When reallocating, the smallest serial number is allocated, for example, multiple target nodes are sequentially allocated serial numbers such as 1, 2, 3, etc. Then, other nodes other than the target node are assigned new serial numbers in sequence.

[0122] The new coding table was then used to encode drone aerial survey data and real-time soil remediation data, converting the raw data into character sequences that conform to the graph structure template. The encoding process strictly adhered to the rules of the new coding table, uniformly formatting different types of data (such as numerical and textual data).

[0123] The encoded data characters are separated in sequence and inserted into the corresponding nodes in sequence according to the sequence numbers assigned to each node in the graph structure template to form a complete data graph structure.

[0124] The constructed data graph structure is sent to the management terminal of the contaminated site remediation client. The client simply parses the data graph structure according to the same process to obtain authentic drone aerial survey data and real-time soil remediation data. However, if the parsing process is incorrect, authentic drone aerial survey data and real-time soil remediation data will not be obtained. In this embodiment, the above technical solution ensures the security of data during transmission, and even if a leak occurs, the actual data will not be disclosed.

[0125] In one embodiment, obtaining parameter information of the new coding table in which the new coding table is mutated, and assigning serial numbers to the nodes of the graph structure template based on the parameter information, includes:

[0126] Obtain parameter information of the new coding table where the variation occurs, the parameter information including the number of variations and the sequence number corresponding to the specific position;

[0127] generating a simulation curve based on the parameter information, and superimposing the simulation curve on the graph structure template, with both ends of the simulation curve located on two designated nodes;

[0128] Based on the simulation curve, the graph structure template is divided into a plurality of sub-regions; in each sub-region, the nodes in the sub-region are reordered according to the trend direction of the simulation curve;

[0129] The sub-regions are sorted according to their sizes; based on the order of the sub-regions and the sorting of the nodes in each sub-region, the sorting of the nodes in the graph structure template is obtained, and corresponding serial numbers are assigned.

[0130] In this embodiment, by converting the coding table variation parameter information into a visual simulation curve, the graph structure template nodes are innovatively sorted and serial numbers are assigned.

[0131] In the process of re-arranging the new coding table according to the abnormal type, mutations such as data item position adjustment and coding rule change will occur. This step accurately identifies the mutated parameters by comparing the new coding table with the original coding table byte by byte. Specifically, the number of mutated parameters is recorded to clarify the number of parameters that have changed in the coding table; at the same time, the specific position number corresponding to each mutated parameter in the new coding table is obtained. These numbers reflect the new position of the parameter in the coding system. For example, if a data item is adjusted from the 5th position in the original coding table to the 8th position in the new coding table, then "8" is the position number corresponding to the mutated parameter. This information constitutes the basic data for subsequent operations and provides a quantitative basis for node sorting.

[0132] Based on the number and position numbers of the acquired mutation parameters, a simulation curve is generated. The curve generation rules can be based on the distribution characteristics of the parameter position numbers. For example, a spline interpolation algorithm is used, with the position numbers of the mutation parameters as key points, to construct a smooth curve that intuitively reflects the overall trend of the code table variation. This simulation curve is then superimposed on the graph structure template, with the two ends of the curve designated to be located at two specific nodes. These two designated nodes can be determined based on key nodes of the graph structure template (such as the starting node or core function node) or nodes with a strong correlation with the code table variation. By superimposing the simulation curve, the abstract code table variation information is visually associated with the graph structure template, providing visual clues for the subsequent division of node regions.

[0133] The simulation curve divides the graph structure template into multiple sub-regions, each containing several nodes. Within each sub-region, the nodes are reordered according to the direction of the simulation curve. If the simulation curve extends from left to right, the nodes are arranged in order from left to right; if the curve shows a circular trend, the nodes are arranged in a clockwise or counterclockwise direction. This sorting method makes the node order correspond to the variation trend of the coding table. For example, in the graph structure sub-region corresponding to the area where the variation in the coding table is more concentrated, the node order also more closely reflects this variation trend. In this way, the variation characteristics of the coding table are mapped to the spatial arrangement of the graph structure nodes, enhancing the logic of data expression.

[0134] All sub-regions are sorted according to the size of each sub-region (which can be measured by indicators such as the number of nodes in the sub-region and the area occupied by the graph structure template). For example, the sub-regions are arranged in order from most to least or from small to large according to the number of nodes. Then, based on the sorting results of the sub-regions and the sorting of the nodes in each sub-region, all nodes in the graph structure template are connected in series to form a complete node sorting sequence. Finally, starting from 1, each node is assigned a sequence number according to this sorting sequence. In this way, the sequence number not only reflects the relative position of the node in its respective sub-region, but also reflects the overall relationship between the sub-regions. This allows the node sequence number assignment of the graph structure template to be closely integrated with the variation of the new coding table, providing an orderly and logically clear foundation for subsequent data display, analysis, and decision-making based on the graph structure.

[0135] In one embodiment, obtaining parameter information of the new coding table in which the new coding table is mutated, and assigning serial numbers to the nodes of the graph structure template based on the parameter information, includes:

[0136] Compare the new code table with the original code table field by field to identify the changes such as addition or deletion of data items, change of code length, and conversion of data types, and generate a parameter change list in the order of the field positions where the changes occurred;

[0137] The nodes in the graph structure template are divided into three categories: basic information, process monitoring, and result feedback. The priority order of nodes in different categories is determined according to the contaminated site remediation business process.

[0138] Match each mutation parameter with the graph structure template node in the order of the parameter mutation list;

[0139] Assign a serial number to each node associated with the mutated parameter. The serial numbers are incremented in sequence. Nodes in the same category are assigned serial numbers in the order of their association in the parameter mutation list.

[0140] For nodes that are not associated with mutation parameters, serial numbers are assigned to them in sequence after the nodes that have been assigned serial numbers according to their category priority and topological level in the graph structure template;

[0141] Perform an integrity check on the assigned node serial numbers to check whether there are any missing or duplicate serial numbers. If any problems are found, readjust the serial numbers based on the node category priority and topological order.

[0142] In this embodiment, the new coding table is reorganized based on the abnormal progress of the contaminated site remediation, and there are differences with the original coding table. This step accurately identifies three major types of variations by comparing the new coding table with the original coding table field by field: addition and deletion of data items (such as adding soil moisture monitoring data items and deleting redundant equipment number fields), changes in code length (such as adjusting the code length of pollutant concentration from 6 bits to 8 bits to improve accuracy), and data type conversion (such as converting the numerical code of the equipment operating status to an enumeration code). According to the order of the field position where the variation occurs in the coding table, the detailed information of each variation (including variation type, original content, changed content, field position, etc.) is recorded to generate a parameter variation list. This list clearly presents the details of the changes in the coding table and provides a data basis for the subsequent node number allocation.

[0143] The nodes in the graph structure template carry various types of information related to contaminated site remediation. To ensure orderly number assignment, these nodes are divided into three categories: basic information (static information nodes such as the site's geographic location and initial soil composition), process monitoring (dynamic monitoring nodes such as remediation equipment operating parameters and real-time pollutant concentration data), and result feedback (summary nodes such as remediation effectiveness evaluation and compliance status determination). The order of priority for nodes in these different categories is determined based on the business process logic of contaminated site remediation: basic information, as a prerequisite for remediation work, has the highest priority; process monitoring, reflecting the dynamics of the remediation process, has the next highest priority; and result feedback, used for final evaluation, has a relatively lower priority. This categorization and prioritization ensures that number assignment aligns with business logic and data importance.

[0144] Using the parameter variation list generated in step 1 as a clue, each mutated parameter is matched to a node in the graph structure template in the order of the list. This matching process is based on the semantic relevance of the data. For example, if a parameter variation involves a coding change for soil heavy metal concentration, it is associated with the soil pollutant concentration monitoring node in the graph structure template; if the variation involves a coding adjustment for the remediation equipment model, it is matched to the remediation equipment information node. This matching establishes a correspondence between coding table variations and graph structure nodes, providing a direct basis for node number assignment.

[0145] Nodes successfully matched to a variation parameter are assigned serial numbers according to the order in the parameter variation list. Serial numbers are incremented starting from 1 to ensure that each node is uniquely numbered. Within nodes of the same category, serial numbers are assigned strictly according to the order in which they are associated in the parameter variation list. For example, if the soil initial pH node in the basic information category is associated with a variation parameter before the site area node, the former will receive a lower serial number. This assignment method allows serial numbers to intuitively reflect the order in which the coding table changes, as well as the importance or processing order of nodes within the same data category.

[0146] For nodes that are not associated with mutation parameters, serial numbers are assigned according to the priority of their category and the topological level in the graph structure template. First, according to the category priority determined in step 2, serial numbers are assigned to nodes in high-priority categories first; within the same category, serial numbers are assigned in descending order based on the topological level of the nodes in the graph structure (such as higher levels closer to the root node). For example, the remaining nodes in the basic information class obtain serial numbers before the remaining nodes in the process monitoring class; within the process monitoring class, nodes that are more closely associated with the core repair process (higher topological levels) are given serial numbers first. In this way, it is ensured that all nodes can obtain reasonable serial numbers and fully cover the graph structure template.

[0147] After assigning serial numbers to all nodes, a comprehensive integrity check is performed on the serial numbers. All nodes are checked for missing serial numbers (e.g., skipped numbers) or duplicate serial numbers (e.g., multiple nodes using the same number). If any issues are found, the serial numbers are readjusted based on the node's category priority and topological order: the serial numbers of nodes in high-priority categories are retained first, while those of lower-priority nodes are adjusted. Within the same category, serial numbers are adjusted based on the topological hierarchy and business logic until all nodes have continuous and unique serial numbers. This verification and adjustment process ensures the accuracy and reliability of node serial number assignment, providing a stable foundation for subsequent graph-based data display, analysis, and decision-making.

[0148] Reference Figure 2 In another embodiment of the present invention, a contaminated site remediation progress management device integrating multi-source data is provided, comprising:

[0149] The acquisition module is used to collect drone aerial survey data and real-time soil remediation data of contaminated sites, and perform semantic association to obtain a semantically fused multi-source data set;

[0150] A screening module is used to analyze the association relationship between each data in the multi-source data set based on an association rule algorithm, and to screen out a set of key factors based on the association relationship and the correlation, in combination with the correlation between calculation and repair progress;

[0151] A prediction module is used to input the data in the key factor set into a pre-trained remediation progress prediction model to perform real-time progress prediction and output a predicted value of the current remediation progress of the contaminated site;

[0152] The management module is used to perform deviation analysis between the predicted value and the actual value of the repair progress uploaded by the construction party, calculate the error between the predicted value and the actual value, analyze the changing trend of the error through the sliding window algorithm, and judge whether there is any abnormality in the repair progress based on the changing trend.

[0153] In this embodiment, for the specific implementation of each module in the above device embodiment, please refer to the above method embodiment, which will not be repeated here.

[0154] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3 As shown. The computer device includes a processor, memory, display screen, input device, network interface and database connected via a system bus. The processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the above method is implemented.

[0155] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0156] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which implements the above-described method when executed by a processor. It is understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0157] In summary, the contaminated site remediation progress management method for integrating multi-source data provided in an embodiment of the present invention includes: collecting drone aerial survey data and real-time soil remediation data of the contaminated site, and performing semantic association to obtain a semantically fused multi-source data set; based on an association rule algorithm, analyzing the association relationship between each data in the multi-source data set, combining the calculation with the remediation progress correlation, and filtering out a key factor set based on the association relationship and the correlation; inputting the data in the key factor set into a pre-trained remediation progress prediction model to perform real-time progress prediction and output a predicted value of the current remediation progress of the contaminated site; performing a deviation analysis between the predicted value and the actual remediation progress value uploaded by the construction party, calculating the error between the predicted value and the actual value, analyzing the change trend of the error through a sliding window algorithm, and judging whether the remediation progress is abnormal based on the change trend. In the present invention, by collecting drone aerial survey data and real-time soil remediation data of the contaminated site and performing semantic association; and at the same time, filtering out a key factor set based on an association rule algorithm, the defects of the current single data source and the failure to accurately filter out the key factors that truly affect the remediation progress, resulting in inaccurate remediation progress prediction, are overcome.

[0158] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing the relevant hardware using a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the above-described method embodiments. Any reference to memory, storage, database, or other media provided herein and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct RAMbus dynamic RAM (DRDRAM), and RAMbus dynamic RAM.

[0159] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, apparatus, article, or method comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, apparatus, article, or method. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, apparatus, article, or method comprising the element.

[0160] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A contaminated site remediation progress management method integrating multi-source data, characterized in that: The following steps are involved: Collect drone aerial survey data and real-time soil remediation data of contaminated sites, perform semantic association, and obtain a semantically fused multi-source data set; Based on the association rule algorithm, the association relationship between each data in the multi-source data set is analyzed, and the correlation between the calculation and the repair progress is combined to screen out the key factor set according to the association relationship and the correlation; Inputting the data in the key factor set into a pre-trained remediation progress prediction model to perform real-time progress prediction, and outputting a predicted value of the current remediation progress of the contaminated site; Perform a deviation analysis on the predicted value and the actual value of the repair progress uploaded by the construction party, calculate the error between the predicted value and the actual value, analyze the change trend of the error through a sliding window algorithm, and determine whether there is any abnormality in the repair progress based on the change trend; If an exception occurs, get the exception type; Determine a corresponding graph structure template based on the exception type; The graph structure template is an undirected graph, including multiple nodes, and the nodes are connected by undirected edges; Rearranging the coding table based on the graph structure attributes of the graph structure template to obtain a new coding table; Obtaining parameter information of the new coding table in which the variation occurs, and assigning serial numbers to each node of the graph structure template based on the parameter information; Encoding the UAV aerial survey data and soil remediation real-time data based on the new coding table to obtain coded data; Separating the characters in the encoded data in sequence, and inserting them into the nodes of the graph structure template in sequence according to the sequence numbers assigned to the nodes of the graph structure template, to obtain a data graph structure; The data graph structure is sent to a management terminal of a contaminated site restoration entrusting party.

2. The contaminated site remediation progress management method integrating multi-source data according to claim 1 is characterized in that: Perform semantic association to obtain a semantically fused multi-source data set, including: Feature extraction is performed on UAV aerial survey data and real-time soil remediation data to obtain aerial survey data features and soil data features; The aerial survey data features and soil data features are projected into a unified semantic space, and the semantic space is aligned by minimizing the distribution difference between the modalities to obtain a semantically fused multi-source data set.

3. The contaminated site remediation progress management method integrating multi-source data according to claim 1 is characterized in that: Based on the association rule algorithm, the association relationship between each data in the multi-source data set is analyzed, and the correlation between each data and the repair progress is calculated. According to the association relationship and the correlation, a set of key factors is screened out, including: Discretizing the multi-source data set to obtain a formatted data set suitable for association rule mining; Performing association mining on the formatted data set based on the Apriori algorithm to generate frequent item sets, generating association relationships based on the frequent item sets, calculating the support, confidence, and lift of each association relationship, and obtaining an association relationship set; Based on the preset support threshold, confidence threshold and lift threshold, the set of association relationships is screened to filter out association relationships that do not meet the conditions and obtain valid association relationships; For each valid association relationship, calculate the mutual information between the association relationship and the restoration progress data, and evaluate the correlation between the valid association relationship and the restoration progress based on the mutual information; Based on the preset correlation threshold, data items that reach the correlation threshold are extracted from the effective association relationships to construct a set of key factors.

4. The contaminated site remediation progress management method integrating multi-source data according to claim 1 is characterized in that: Analyze the error trend using a sliding window algorithm and determine whether the repair progress is abnormal based on the trend, including: The error between the predicted value and the actual value is divided into sliding windows, and each sliding step generates a window data set containing continuous error values; the mean, standard deviation, range and coefficient of variation of the error in each window data set are calculated to construct a window statistical feature vector; The ARIMA model is used to fit the changing trend of the error within the window to obtain the trend prediction model and its parameters; the trend prediction model parameters of adjacent windows are analyzed for differences, the parameter change rate and relative change amount are calculated, and the parameter change feature vector is constructed; Compare the parameter change feature vector of the current window with the confidence interval. If it exceeds the confidence interval, it is determined that the window has an abnormal trend. Perform multi-window consistency verification on the window with abnormal trend to eliminate the single window fluctuation noise. The direction and magnitude of the error change are determined based on the error change trend within the window, and the abnormal trend after consistency verification is typed to obtain the abnormal type.

5. The contaminated site remediation progress management method integrating multi-source data according to claim 1 is characterized in that: The drone aerial survey data includes video data taken by the drone; the soil remediation real-time data includes pollutant concentration data, soil physical property data, and remediation equipment operation data.

6. The contaminated site remediation progress management method integrating multi-source data according to claim 1 is characterized in that: Collect drone aerial survey data and real-time soil remediation data of contaminated sites, including: The spatial distribution characteristics of the contaminated site remediation area were collected, and the flight window was predicted based on historical meteorological data. A genetic algorithm was used to optimize the UAV flight path to generate an aerial survey plan that included shooting missions at different time intervals and from multiple angles. When executing the aerial survey plan, the exposure parameters of the camera on the drone are automatically adjusted according to the real-time light intensity, the laser radar scanning resolution is dynamically calibrated based on the flight altitude, and the acquisition timestamp and geographic coordinate information are synchronously recorded when the aerial survey data is collected; Real-time soil remediation data is collected based on an Internet of Things system deployed at a contaminated site; wherein the Internet of Things system includes soil sensors connected to the network and remediation equipment monitoring terminals.

7. The contaminated site remediation progress management method integrating multi-source data according to claim 1 is characterized in that: Perform semantic association to obtain a semantically fused multi-source data set, including: Extracting the terrain semantic features of the contaminated site and the semantic features of the spatial layout of the remediation equipment from the aerial survey data to construct a three-dimensional spatial semantic feature vector; Extracting pollutant concentration change trend semantics and remediation process execution status semantics from the real-time soil remediation data through a bidirectional long short-term memory network to generate a temporal semantic feature sequence; The three-dimensional spatial semantic feature vector and the temporal semantic feature sequence are mapped to a unified semantic space, and the semantic alignment of the data is performed by minimizing the distribution differences between the modalities to form a semantically fused multi-source data set; wherein, when the repair progress data is updated in real time, the reinforcement learning algorithm is used to dynamically adjust the semantic association weights between different modal data.

Images