AI intelligent underground water level dynamic prediction system based on multi-source data fusion

Abstract

The invention relates to the field of groundwater resource management and protection, and discloses an AI intelligent groundwater level dynamic prediction system based on multi-source data fusion, which comprises a data acquisition module used for acquiring multi-source data to break through the limitation of a single data source and cover various factors influencing the change of the groundwater level; and the data fusion module adopts a multi-criterion decision analysis method and a multi-agent decision mechanism, analyzes the influence weight of each factor on the underground water level under different scales through weighted linear combination, and determines a decisive factor of a specific area. Multi-source data of meteorology, hydrogeology, human activities, building construction and the like are integrated through the data acquisition module, the influence weights of different factors on the underground water level can be comprehensively quantified in combination with a multi-criterion decision analysis method and a multi-agent decision mechanism of the data fusion module, the limitation of a traditional single data source is changed, and the underground water level can be accurately measured. Driving factors of underground water level changes are comprehensively captured, and a foundation is laid for accurate prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of groundwater resource management and protection, and specifically to an AI intelligent groundwater level dynamic prediction system based on multi-source data fusion. Background Art

[0002] In the field of groundwater resource management and protection, accurately predicting groundwater level changes is crucial for urban water supply and drainage planning, building anti-floating design, and geological disaster prevention. Currently, existing groundwater level prediction systems often use a single data source or a fixed weight distribution approach. For example, they rely solely on historical data from groundwater level monitors or empirically set the weights of influencing factors such as meteorology and hydrology. These systems also lack the ability to dynamically identify dominant factors in different scenarios. However, groundwater level changes are influenced by a combination of multiple sources, including meteorology, hydrogeology, human activities, and construction. Furthermore, the weights of various influencing factors vary significantly across scales, such as cities, communities, and individual buildings. Meteorological factors have a greater impact at the urban scale, while human activities and construction are more prominent in community or individual building scenarios. Existing technologies are unable to effectively integrate multi-source data and dynamically quantify the weights of various factors, making it difficult for prediction models to adapt to complex scenario changes. These models suffer from low prediction accuracy and poor multi-scale adaptability, failing to meet the demands of practical engineering applications for high-precision, intelligent prediction. Summary of the Invention

[0003] In response to the shortcomings of the existing technology, the present invention provides an AI intelligent groundwater level dynamic prediction system based on multi-source data fusion, which solves the problems mentioned in the above background technology.

[0004] To achieve the above objectives, the present invention is implemented through the following technical solutions: an AI intelligent groundwater level dynamic prediction system based on multi-source data fusion, the system includes a data acquisition module, a data fusion module, an AI prediction model module, a data storage module and an output module, wherein:

[0005] Data acquisition module: used to collect multi-source data, including but not limited to meteorological data, hydrological data, regional groundwater dynamic monitoring data, and building lifecycle structural health monitoring data, in order to break through the limitations of a single data source and cover various factors that affect groundwater level changes;

[0006] Data fusion module: Using multi-criteria decision analysis and multi-agent decision-making mechanisms, the module analyzes the influence of various factors on groundwater levels at different scales through weighted linear combination to determine the decisive factors in specific areas.

[0007] AI prediction model module: Based on the main factors output by the data fusion module, it uses deep learning algorithms to achieve targeted prediction of groundwater levels by zoning and scale. The AI ​​model includes but is not limited to LSTM neural networks;

[0008] A deep collaboration mechanism is provided between the data fusion module and the AI ​​prediction model module to enhance the model's learning ability for high-weight factors.

[0009] Preferably, the data fusion module includes:

[0010] The data preprocessing unit is configured to perform median filtering, denoising and normalization processing on the multi-source data to generate a normalized fuzzy matrix F = [f ij ]n×m, where the normalization formula is:

[0011]

[0012] Among them, x norm is the normalized data, x is the original data, and x max is the maximum value in the original data, x min is the minimum value in the original data;

[0013] Multi-criteria decision-making unit, configured as:

[0014] Calculate the initial weight W to obtain the influence of each factor;

[0015] The data is converted into a matrix of factor indicators, wherein the factor indicators include:

[0016] Meteorological element index MP = a·RF day / Area, normalized range [0-1];

[0017] Human activity factor index PC = b·ΔS / Area, normalized range [0-1];

[0018] Groundwater level factor index GW = c·ΔH / E, normalized range [0-1];

[0019] Building health factor index BS = d·△H / h, normalized range [0-1];

[0020] Among them, RF day is the daily rainfall, A is the regional area, ΔS is the area of ​​the precipitation funnel caused by human activities, ΔH is the groundwater level, E is the area of ​​the monitoring point influence zone, and h is the water level of the building anti-floating protection device. The normalization principle of other factor indicators continues here, which is basically the ratio of daily monitoring data to regional area multiplied by the adjustment coefficient;

[0021] Construct a multi-criteria decision-making reachability matrix:

[0022]

[0023] Calculate the overlay weight WW1;

[0024] The multi-agent decision-making unit is configured to generate the multi-agent decision weight WW2 through multiple rounds of iterative scoring using the Delphi method;

[0025] The weight fusion unit is configured to fuse WW1 and WW2 using a weighted fusion algorithm to determine the final weight classification. The fusion process is as follows:

[0026]

[0027] Among them, F is the fused data, D i is the data of the i-th data source, w i is the weight of the i-th data source, and n is the number of data source types.

[0028] Preferably, the multi-criteria decision-making unit calculates the initial weight W by the hierarchical analysis method based on the normalized data output by the data preprocessing unit:

[0029] Construct the judgment matrix of factor importance A=[a ij ], where a ij reflects the relative importance of factors i and j, and

[0030] Solve the eigenvector of the judgment matrix and normalize it to obtain the initial weight W;

[0031] The rationality of the weights was tested by the consistency ratio CR, and the influence of each factor was determined when CR < 0.1.

[0032] Preferably, the process of the multi-agent decision unit generating the multi-agent decision weight WW2 by multiple rounds of iterative scoring using the Delphi method includes:

[0033] Establish a team of no fewer than five experts, including experts in hydrogeology, urban planning, and construction engineering, and introduce generative AI agents;

[0034] Provide the normalized fuzzy matrix output by the data preprocessing unit and the factor indicator matrix generated by the multi-criteria decision-making unit to the experts and generative AI agents;

[0035] Experts independently score the importance of each factor indicator based on their professional knowledge, with a score range of 1-9. The generative AI agent outputs the importance score of the corresponding factor indicator based on historical groundwater level data, industry literature and similar case analysis;

[0036] Perform mixed statistical analysis on the scoring results of experts and generative AI agents in each round and calculate the mean value x of the score i and standard deviation σ i , where i is the element index number;

[0037] Feedback the statistical results to the experts and generative AI agents for the next round of scoring. When the standard deviation of all indicators for two consecutive rounds is σ i When <0.5, stop iteration;

[0038] The final score is normalized to generate the multi-agent decision weight WW2.

[0039] Preferably, the scaled prediction logic of the AI ​​prediction model module is as follows: each scale is based on the final weight classification output by the data fusion module, and the prediction basic factors are determined through multiple rounds of dynamic evaluation. The specific process is as follows:

[0040] Urban scale: Dynamically score meteorological factors, human activity factors, and geological factors, select the factor with the highest weight, and make predictions based on the correlation between this factor and groundwater level;

[0041] Administrative district scale: Based on the dynamic scoring results, human activity factors and regional geological structure factors are prioritized as the core basis for prediction;

[0042] Community scale: Comprehensively evaluate the dynamic weights of greening irrigation factors, building construction factors, and local drainage factors, and make targeted predictions based on the factors with the highest weights and their correlation patterns with groundwater levels.

[0043] Preferably, the deep learning-based prediction model used by the AI ​​prediction model module is an LSTM neural network model, which includes an input layer, an LSTM layer, and an output layer. The model structure is as follows:

[0044] The input layer receives the fused data;

[0045] The LSTM layer processes the input data, and its operation process is as follows:

[0046] f t =σ(W f [h t-1 ,x t ]+b f )

[0047] i t =σ(W i [h t-1 ,x t ]+b i )

[0048] o t =σ(W o [h t-1 ,x t ]+b o )

[0049] ct =f t c t-1 +i t tanh(W c [h t-1 ,x t ]+b c )

[0050] h t =o t ·tanh(c t )

[0051] Where, f t is the output of the forget gate, i t is the input gate output, o t is the output gate output, c t is the cell state, is the hidden layer output, σ is the activation function, W f 、W i 、W o 、W c Are weight matrices, where W f Is the weight matrix of the forget gate, which is used to determine the input data x t and the hidden layer output h at the previous moment t-1 The weight in the forget gate operation, W i Is the weight matrix of the input gate, which determines x t With h t-1 The weight in the input gate operation, W o is the weight matrix of the output gate, determine x t With h t-1 The weight in the output gate operation, W c is the weight matrix used to update the cell state, controlling x t With h t-1 The degree of influence on cell state update, b f 、b i 、b o 、b c Represents the bias vector, corresponding to the bias terms in the forget gate, input gate, output gate and cell state update operation;

[0052] The output layer outputs the predicted results of groundwater level.

[0053] Preferably, the AI ​​prediction model module uses a mean square error loss function to train the LSTM neural network model, and the mean square error loss function is:

[0054]

[0055] Among them, MSE is the mean square error, m is the number of samples, and y i is the actual groundwater level, To predict groundwater level.

[0056] Preferably, the deep collaboration mechanism between the data fusion module and the AI ​​prediction model module is specifically as follows:

[0057] The final weight classification obtained after the superposition weight WW1 output by the multi-criteria decision-making unit and the multi-agent decision weight WW2 generated by the multi-agent decision-making unit are processed by the weight fusion unit and applied to the input layer of the LSTM neural network in the form of a dynamic mask. Specifically, it includes:

[0058] The final weight hierarchy is converted into a mask matrix M that is consistent with the input data dimension, where each element m ij Corresponding input data x ij The weight coefficient of

[0059] Perform weighted operations on the input layer of the LSTM neural network: x weighted =M⊙x, where ⊙ represents the Hadamard product, which enables the model to prioritize learning features corresponding to high-weight factors;

[0060] During the training process of the LSTM layer, the elements of the mask matrix M are included in the back-propagation calculation, and the neural network parameters and weight matrix are synchronously updated through the chain rule to achieve the joint optimization of multi-criteria decision-making, multi-agent decision-making and deep learning.

[0061] Preferably, the multi-source data collected by the data collection module is obtained in the following manner:

[0062] Meteorological sensors collect rainfall, evaporation, and temperature data;

[0063] Geological exploration equipment collects data on soil permeability and aquifer thickness;

[0064] Geographic information system to obtain topographic data;

[0065] Building health monitoring sensors obtain data on foundation settlement and anti-floating protection water level changes.

[0066] Preferably, the connection relationship between the modules is:

[0067] The data acquisition module asynchronously transmits multi-source data to the data fusion module through the message queue. After processing, the latter synchronously pushes the weight results to the AI ​​prediction model module through the API interface. The model prediction results are sent and output in real time through WebSocket.

[0068] The present invention provides an AI intelligent groundwater level dynamic prediction system based on multi-source data fusion. It has the following beneficial effects:

[0069] 1. The present invention integrates multi-source data such as meteorology, hydrogeology, human activities, and construction through the data acquisition module, and combines the multi-criteria decision analysis method and multi-agent decision-making mechanism of the data fusion module to comprehensively quantify the impact weights of different factors on groundwater levels, changing the limitations of traditional single data sources, comprehensively capturing the driving factors of groundwater level changes, and laying the foundation for accurate prediction.

[0070] 2. For scenarios of different scales, such as cities, communities, and single buildings, the system can automatically adjust the weights of various influencing factors. For example, at the city scale, the weight calculation focuses on meteorological factors, while at the community scale, the weights of human activities and building construction factors are strengthened. Through the dynamic weight allocation mechanism, it effectively solves the problem that traditional fixed weights are difficult to adapt to scenario differences, and significantly improves the prediction fit under different scenarios.

[0071] 3. The AI ​​prediction model module deeply integrates the dynamic weights output by the data fusion module with the deep learning algorithm, and realizes accurate predictions at different scales and scenarios through the LSTM neural network. At the same time, it utilizes the collaborative optimization of the multi-agent decision-making mechanism and multi-criteria decision analysis to continuously adjust the weight distribution strategy, so that the model can still focus on key influencing factors in complex environments. Compared with traditional prediction methods, it greatly improves the prediction accuracy and reliability, and provides more valuable decision-making basis for engineering applications such as urban water supply and drainage planning and building anti-floating design. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 Schematic diagram of the overall system flow of the present invention;

[0073] Figure 2 This is a flow chart of the data fusion module of the present invention. DETAILED DESCRIPTION

[0074] The following will clearly and completely describe the technical solution of the present invention in conjunction with the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0075] Example:

[0076] Please see the attached Figure 1 -Attached Figure 2 The embodiment of the present invention provides an AI intelligent groundwater level dynamic prediction system based on multi-source data fusion. The system includes a data acquisition module, a data fusion module, an AI prediction model module, a data storage module, and an output module, wherein:

[0077] Data acquisition module: used to collect multi-source data, including but not limited to meteorological data, hydrological data, regional groundwater dynamic monitoring data, and building lifecycle structural health monitoring data, in order to break through the limitations of a single data source and cover various factors that affect groundwater level changes;

[0078] Data fusion module: Using multi-criteria decision analysis and multi-agent decision-making mechanisms, the module analyzes the influence of various factors on groundwater levels at different scales through weighted linear combination to determine the decisive factors in specific areas.

[0079] AI prediction model module: Based on the main factors output by the data fusion module, it uses deep learning algorithms to achieve targeted prediction of groundwater levels by zoning and scale. The AI ​​model includes but is not limited to LSTM neural networks;

[0080] A deep collaboration mechanism is provided between the data fusion module and the AI ​​prediction model module to enhance the model's learning ability for high-weight factors.

[0081] Specifically, the data acquisition module integrates multi-source data such as meteorology, hydrology, groundwater monitoring, and building health to break the limitations of single data; the data fusion module uses multi-criteria decision analysis and multi-agent decision-making mechanism to analyze the impact weight of various factors on groundwater levels at different scales; the AI ​​prediction model module uses deep learning algorithms based on the main factors after fusion to achieve partition and scale predictions; the data storage module is used for data persistence; and the output module is responsible for result display.

[0082] The integrated collection of multi-source data comprehensively covers all factors that affect groundwater level changes, avoiding prediction bias caused by information missing; the combination of multi-criteria decision-making and multi-agent decision-making mechanisms can accurately quantify the weights of various factors at different scales, making the prediction more in line with actual scenarios; the deep collaboration mechanism enhances the model's ability to learn key factors, significantly improving prediction accuracy and reliability, and meeting the needs of high-precision predictions in projects such as urban water supply and drainage planning and building anti-floating design.

[0083] The data fusion module includes:

[0084] The data preprocessing unit is configured to perform median filtering, denoising and normalization processing on the multi-source data to generate a normalized fuzzy matrix F = [f ij ]n×m, where the normalization formula is:

[0085]

[0086] Among them, x norm is the normalized data, x is the original data, and x max is the maximum value in the original data, x min is the minimum value in the original data;

[0087] Multi-criteria decision-making unit, configured as:

[0088] Calculate the initial weight W to obtain the influence of each factor;

[0089] The data is converted into a matrix of factor indicators, wherein the factor indicators include:

[0090] Meteorological element index MP = a·RF day / Area, normalized range [0-1];

[0091] Human activity factor index PC = b·ΔS / Area, normalized range [0-1];

[0092] Groundwater level factor index GW = c·ΔH / E, normalized range [0-1];

[0093] Building health factor index BS = d·△H / h, normalized range [0-1];

[0094] Among them, RF day is the daily rainfall, A is the regional area, ΔS is the area of ​​the precipitation funnel caused by human activities, ΔH is the groundwater level, E is the area of ​​the monitoring point influence zone, and h is the water level of the building anti-floating protection device. The normalization principle of other factor indicators continues here, which is basically the ratio of daily monitoring data to regional area multiplied by the adjustment coefficient;

[0095] Construct a multi-criteria decision-making reachability matrix:

[0096]

[0097] Calculate the overlay weight WW1;

[0098] The multi-agent decision-making unit is configured to generate the multi-agent decision weight WW2 through multiple rounds of iterative scoring using the Delphi method;

[0099] The weight fusion unit is configured to fuse WW1 and WW2 using a weighted fusion algorithm to determine the final weight classification. The fusion process is as follows:

[0100]

[0101] Among them, F is the fused data, D i is the data of the i-th data source, w i is the weight of the i-th data source, and n is the number of data source types.

[0102] Specifically, data preprocessing improves data quality and consistency, ensuring the accuracy of subsequent analysis; the multi-criteria decision-making unit quantitatively analyzes the weights of each factor based on data, reducing the limitations of subjective experience; the multi-agent decision-making unit combines expert knowledge to make up for the possible shortcomings of the data-driven method; the fusion of the weights of the two realizes the complementary advantages of data and experience, ensuring that the final weights can accurately reflect the actual impact of each factor on the groundwater level, providing a reliable basis for the prediction model.

[0103] The multi-criteria decision-making unit calculates the initial weight W by the hierarchical analysis method based on the normalized data output by the data preprocessing unit:

[0104] Construct the judgment matrix of factor importance A=[a ij ], where a ij reflects the relative importance of factors i and j, and

[0105] Solve the eigenvector of the judgment matrix and normalize it to obtain the initial weight W;

[0106] The rationality of the weights was tested by the consistency ratio CR, and the influence of each factor was determined when CR < 0.1.

[0107] Specifically, the application of the hierarchical analysis method transforms the complex multi-factor weight determination problem into a hierarchical and structured analysis process, making the weight calculation more logical and systematic; the consistency ratio test ensures the reliability of the weight results, avoids weight deviation caused by inconsistent judgment matrices, and thus improves the accuracy of the output weights of the multi-criteria decision-making unit.

[0108] The process of the multi-agent decision-making unit generating the multi-agent decision weight WW2 through multiple rounds of iterative scoring using the Delphi method includes:

[0109] Establish a team of no fewer than five experts, including experts in hydrogeology, urban planning, and construction engineering, and introduce generative AI agents;

[0110] Provide the normalized fuzzy matrix output by the data preprocessing unit and the factor indicator matrix generated by the multi-criteria decision-making unit to the experts and generative AI agents;

[0111] Experts independently score the importance of each factor indicator based on their professional knowledge, with a score range of 1-9. The generative AI agent outputs the importance score of the corresponding factor indicator based on historical groundwater level data, industry literature and similar case analysis;

[0112] Perform mixed statistical analysis on the scoring results of experts and generative AI agents in each round and calculate the mean value x of the score i and standard deviation σ i, where i is the element index number;

[0113] Feedback the statistical results to the experts and generative AI agents for the next round of scoring. When the standard deviation of all indicators for two consecutive rounds is σ i When <0.5, stop iteration;

[0114] The final score is normalized to generate the multi-agent decision weight WW2.

[0115] Specifically, the human-machine collaborative Delphi method fully utilizes the professional experience and knowledge of experts, while using the big data analysis capabilities of AI to capture the impact of difficult-to-quantify factors in the data on groundwater levels; multiple rounds of iterative scoring and statistical analysis enable the weight results to be continuously optimized under the joint action of the expert group wisdom and AI data analysis, reducing individual subjective bias and improving the objectivity and accuracy of the weights; the generated multi-agent decision weights are integrated with the results of the multi-criteria decision-making unit, further improving the reliability and applicability of the overall weights.

[0116] The scaled prediction logic of the AI ​​prediction model module is as follows: each scale is based on the final weight classification output by the data fusion module, and the prediction basic factors are determined through multiple rounds of dynamic evaluation. The specific process is as follows:

[0117] Urban scale: Dynamically score meteorological factors, human activity factors, and geological factors, select the factor with the highest weight, and make predictions based on the correlation between this factor and groundwater level;

[0118] Administrative district scale: Based on the dynamic scoring results, human activity factors and regional geological structure factors are prioritized as the core basis for prediction;

[0119] Community scale: Comprehensively evaluate the dynamic weights of greening irrigation factors, building construction factors, and local drainage factors, and make targeted predictions based on the factors with the highest weights and their correlation patterns with groundwater levels.

[0120] Specifically, a dynamic weight-driven scale prediction mechanism is adopted. Its core is to break away from the limitations of traditional fixed factor prediction and achieve adaptive selection of basic prediction factors based on the final weight classification output by the data fusion module. Specifically, the system constructs an open factor evaluation set for three scales: city, administrative district, and community. At the city scale, the evaluation set covers factors such as meteorology (precipitation, evaporation, and temperature), human activities (groundwater extraction, municipal water use, and sewage discharge), and geology (soil permeability, aquifer thickness, and topography); at the administrative district scale, it expands to factors such as industrial water use, agricultural irrigation, stratum lithology, and groundwater flow fields; and at the community scale, it incorporates factors such as greening irrigation, construction drainage, local drainage system efficiency, and underground pipeline leakage.

[0121] The deep learning-based prediction model used by the AI ​​prediction model module is an LSTM neural network model, which includes an input layer, an LSTM layer, and an output layer. The model structure is as follows:

[0122] The input layer receives the fused data;

[0123] The LSTM layer processes the input data, and its operation process is as follows:

[0124] f t =σ(W f [h t-1 ,x t ]+b f )

[0125] i t =σ(W i [h t-1 ,x t ]+b i )

[0126] o t =σ(W o [h t-1 ,x t ]+b o )

[0127] c t =f t c t-1 +i t tanh(W c [h t-1 ,x t ]+b c )

[0128] h t =o t ·tanh(c t )

[0129] Where, f t is the output of the forget gate, i t is the input gate output, o t is the output gate output, c t is the cell state, is the hidden layer output, σ is the activation function, W f 、W i 、W o 、W c Are weight matrices, where W f Is the weight matrix of the forget gate, which is used to determine the input data x t and the hidden layer output h at the previous moment t-1 The weight in the forget gate operation, W i Is the weight matrix of the input gate, which determines x tWith h t-1 The weight in the input gate operation, W o is the weight matrix of the output gate, determine x t With h t-1 The weight in the output gate operation, W c is the weight matrix used to update the cell state, controlling x t With h t-1 The degree of influence on cell state update, b f 、b i 、b o 、b c Represents the bias vector, corresponding to the bias terms in the forget gate, input gate, output gate and cell state update operation;

[0130] The output layer outputs the predicted results of groundwater level.

[0131] Specifically, the input layer receives the fused data output by the data fusion module; the LSTM layer selectively memorizes and updates the input data through the coordinated operation of the forget gate, input gate, and output gate. The forget gate determines the degree of retention of the cell state at the previous moment, the input gate controls the update of the cell state by the current input data, and the output gate generates the hidden layer output according to the cell state; the output layer outputs the predicted results of the groundwater level based on the output of the LSTM layer. Through this special structure, the model can effectively handle the long-term dependency problem in time series data.

[0132] Its gating mechanism enables the model to adaptively screen and memorize key information, avoiding prediction errors caused by data noise and redundant information. Compared with traditional neural networks, it significantly improves the accuracy and stability of groundwater level prediction, providing reliable prediction data for dynamic water resources management.

[0133] The AI ​​prediction model module uses the mean square error loss function to train the LSTM neural network model. The mean square error loss function is:

[0134]

[0135] Among them, MSE is the mean square error, m is the number of samples, and y i is the actual groundwater level, To predict groundwater level.

[0136] Specifically, the mean square error loss function can intuitively measure the degree of deviation between the model prediction results and the actual values, providing a clear optimization goal for model training; through iterative training and parameter adjustment, the model can continuously reduce the error and improve the prediction accuracy, making the groundwater level prediction results based on this model more in line with the actual situation.

[0137] The deep collaboration mechanism between the data fusion module and the AI ​​prediction model module is specifically as follows:

[0138] The final weight classification obtained after the superposition weight WW1 output by the multi-criteria decision-making unit and the multi-agent decision weight WW2 generated by the multi-agent decision-making unit are processed by the weight fusion unit and applied to the input layer of the LSTM neural network in the form of a dynamic mask. Specifically, it includes:

[0139] The final weight hierarchy is converted into a mask matrix M that is consistent with the input data dimension, where each element m ij Corresponding input data x ij The weight coefficient of

[0140] Perform weighted operations on the input layer of the LSTM neural network: x weighted =M⊙x, where ⊙ represents the Hadamard product, which enables the model to prioritize learning features corresponding to high-weight factors;

[0141] During the training process of the LSTM layer, the elements of the mask matrix M are included in the back-propagation calculation, and the neural network parameters and weight matrix are synchronously updated through the chain rule to achieve the joint optimization of multi-criteria decision-making, multi-agent decision-making and deep learning.

[0142] Specifically, the deep collaboration mechanism breaks the information barrier between data fusion and prediction models, enabling the model to fully utilize the weight fusion results to focus on key influencing factors; the dynamic masking technology enhances the model's learning ability for high-weight factors, avoiding interference from low-weight factors during training, and improving the model's learning efficiency and prediction accuracy; the joint optimization realizes the complementary advantages of the decision-making mechanism and deep learning, further improving the adaptability and prediction accuracy of the entire system to complex groundwater level change scenarios.

[0143] The multi-source data collected by the data acquisition module is obtained in the following ways:

[0144] Meteorological sensors collect rainfall, evaporation, and temperature data;

[0145] Geological exploration equipment collects data on soil permeability and aquifer thickness;

[0146] Geographic information system to obtain topographic data;

[0147] Building health monitoring sensors obtain data on foundation settlement and anti-floating protection water level changes.

[0148] Specifically, diversified data acquisition methods ensure the comprehensiveness and diversity of the data, covering all aspects that affect groundwater level changes; professional equipment and systems ensure the accuracy and real-time nature of the data, providing a high-quality data foundation for subsequent data processing and predictive analysis.

[0149] The connection relationship between the modules is:

[0150] The data acquisition module asynchronously transmits multi-source data to the data fusion module through the message queue. After processing, the latter synchronously pushes the weight results to the AI ​​prediction model module through the API interface. The model prediction results are sent and output in real time through WebSocket.

[0151] Specific, clear and reasonable module connection relationships ensure the smooth flow and efficient processing of data within the system.

[0152] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. AI intelligent groundwater level dynamic prediction system based on multi-source data fusion, characterized by: The system includes a data acquisition module, a data fusion module, an AI prediction model module, a data storage module and an output module, wherein: Data acquisition module: used to collect multi-source data, including but not limited to meteorological data, hydrological data, regional groundwater dynamic monitoring data, and building lifecycle structural health monitoring data, in order to break through the limitations of a single data source and cover various factors that affect groundwater level changes; Data fusion module: Using multi-criteria decision analysis and multi-agent decision-making mechanisms, the module analyzes the influence of various factors on groundwater levels at different scales through weighted linear combination to determine the decisive factors in specific areas. AI prediction model module: Based on the main factors output by the data fusion module, it uses deep learning algorithms to achieve targeted prediction of groundwater levels by zoning and scale. The AI ​​model includes but is not limited to LSTM neural networks; A deep collaboration mechanism is provided between the data fusion module and the AI ​​prediction model module to enhance the model's learning ability for high-weight factors.

2. The AI ​​intelligent groundwater level dynamic prediction system based on multi-source data fusion according to claim 1 is characterized in that: The data fusion module includes: The data preprocessing unit is configured to perform median filtering, denoising and normalization processing on the multi-source data to generate a normalized fuzzy matrix F = [f ij ]n×m, where the normalization formula is: Among them, x norm is the normalized data, x is the original data, and x max is the maximum value in the original data, x min is the minimum value in the original data; Multi-criteria decision-making unit, configured as: Calculate the initial weight W to obtain the influence of each factor; The data is converted into a matrix of factor indicators, wherein the factor indicators include: Meteorological element index MP = a·RF day / Area, normalized range [0-1]; Human activity factor index PC = b·ΔS / Area, normalized range [0-1]; Groundwater level factor index GW = c·ΔH / E, normalized range [0-1]; Building health factor index BS = d·△H / h, normalized range [0-1]; Among them, RF day is the daily rainfall, A is the regional area, ΔS is the area of ​​the precipitation funnel caused by human activities, ΔH is the groundwater level, E is the area of ​​the monitoring point influence zone, and h is the water level of the building anti-floating protection device. The normalization principle of other factor indicators continues here, which is basically the ratio of daily monitoring data to regional area multiplied by the adjustment coefficient; Construct a multi-criteria decision-making reachability matrix: Calculate the overlay weight WW1; The multi-agent decision-making unit is configured to generate the multi-agent decision weight WW2 through multiple rounds of iterative scoring using the Delphi method; The weight fusion unit is configured to fuse WW1 and WW2 using a weighted fusion algorithm to determine the final weight classification. The fusion process is as follows: Among them, F is the fused data, D i is the data of the i-th data source, w i is the weight of the i-th data source, and n is the number of data source types.

3. The AI ​​intelligent groundwater level dynamic prediction system based on multi-source data fusion according to claim 2 is characterized in that: The multi-criteria decision-making unit calculates the initial weight W by the hierarchical analysis method based on the normalized data output by the data preprocessing unit: Construct the judgment matrix of factor importance A=[a ij ], where a ij reflects the relative importance of factors i and j, and Solve the eigenvector of the judgment matrix and normalize it to obtain the initial weight W; The rationality of the weights was tested by the consistency ratio CR, and the influence of each factor was determined when CR < 0.

1.

4. The AI ​​intelligent groundwater level dynamic prediction system based on multi-source data fusion according to claim 2 is characterized in that: The process of the multi-agent decision-making unit generating the multi-agent decision weight WW2 through multiple rounds of iterative scoring using the Delphi method includes: Establish a team of no fewer than five experts, including experts in hydrogeology, urban planning, and construction engineering, and introduce generative AI agents; Provide the normalized fuzzy matrix output by the data preprocessing unit and the factor indicator matrix generated by the multi-criteria decision-making unit to the experts and generative AI agents; Experts independently score the importance of each factor indicator based on their professional knowledge, with a score range of 1-9. The generative AI agent outputs the importance score of the corresponding factor indicator based on historical groundwater level data, industry literature and similar case analysis; Perform mixed statistical analysis on the scoring results of experts and generative AI agents in each round and calculate the mean value x of the score i and standard deviation σ i , where i is the element index number; Feedback the statistical results to the experts and generative AI agents for the next round of scoring. When the standard deviation of all indicators for two consecutive rounds is σ i When <0.5, stop iteration; The final score is normalized to generate the multi-agent decision weight WW2.

5. The AI ​​intelligent groundwater level dynamic prediction system based on multi-source data fusion according to claim 1 is characterized in that: The scaled prediction logic of the AI ​​prediction model module is as follows: each scale is based on the final weight classification output by the data fusion module, and the prediction basic factors are determined through multiple rounds of dynamic evaluation. The specific process is as follows: Urban scale: Dynamically score meteorological factors, human activity factors, and geological factors, select the factor with the highest weight, and make predictions based on the correlation between this factor and groundwater level; Administrative district scale: Based on the dynamic scoring results, human activity factors and regional geological structure factors are prioritized as the core basis for prediction; Community Scale: Comprehensively evaluate the dynamic weights of greening irrigation factors, building construction factors, and local drainage factors, and make targeted predictions based on the factors with the highest weights and their correlation patterns with groundwater levels.

6. The AI ​​intelligent groundwater level dynamic prediction system based on multi-source data fusion according to claim 1 is characterized in that: The deep learning-based prediction model used by the AI ​​prediction model module is an LSTM neural network model, which includes an input layer, an LSTM layer, and an output layer. The model structure is as follows: The input layer receives the fused data; The LSTM layer processes the input data, and its operation process is as follows: f t =σ(W f [h t-1 ,x t ]+b f ) i t =σ(W i [h t-1 ,x t ]+b i ) the t =σ(W o [h t-1 ,x t ]+b o ) c t =f t ·c t-1 +i t ·tanh(W c [h t-1 ,x t ]+b c ) h t =o t ·tanh(c t ) Where, f t is the output of the forget gate, i t is the input gate output, o t is the output gate output, c t is the cell state, is the hidden layer output, σ is the activation function, W f 、W i 、W o 、W c Are weight matrices, where W f Is the weight matrix of the forget gate, which is used to determine the input data x t and the hidden layer output h at the previous moment t-1 The weight in the forget gate operation, W i Is the weight matrix of the input gate, which determines x t With h t-1 The weight in the input gate operation, W o is the weight matrix of the output gate, determine x t With h t-1 The weight in the output gate operation, W c is the weight matrix used to update the cell state, controlling x t With h t-1 The degree of influence on cell state update, b f 、b i 、b o 、b c Represents the bias vector, corresponding to the bias terms in the forget gate, input gate, output gate and cell state update operation; The output layer outputs the predicted results of groundwater level.

7. The AI ​​intelligent groundwater level dynamic prediction system based on multi-source data fusion according to claim 1 is characterized in that: The AI ​​prediction model module uses the mean square error loss function to train the LSTM neural network model. The mean square error loss function is: Among them, MSE is the mean square error, m is the number of samples, and y i is the actual groundwater level, To predict groundwater level.

8. The AI ​​intelligent groundwater level dynamic prediction system based on multi-source data fusion according to claim 1 is characterized in that: The deep collaboration mechanism between the data fusion module and the AI ​​prediction model module is specifically as follows: The final weight classification obtained after the superposition weight WW1 output by the multi-criteria decision-making unit and the multi-agent decision weight WW2 generated by the multi-agent decision-making unit are processed by the weight fusion unit and applied to the input layer of the LSTM neural network in the form of a dynamic mask. Specifically, it includes: The final weight hierarchy is converted into a mask matrix M that is consistent with the input data dimension, where each element m ij Corresponding input data x ij The weight coefficient of Perform weighted operations on the input layer of the LSTM neural network: x weighted =M⊙x, where ⊙ represents the Hadamard product, which enables the model to prioritize learning features corresponding to high-weight factors; During the training process of the LSTM layer, the elements of the mask matrix M are included in the back-propagation calculation, and the neural network parameters and weight matrix are synchronously updated through the chain rule to achieve the joint optimization of multi-criteria decision-making, multi-agent decision-making and deep learning.

9. The AI ​​intelligent groundwater level dynamic prediction system based on multi-source data fusion according to claim 1 is characterized in that: The multi-source data collected by the data acquisition module is obtained in the following ways: Meteorological sensors collect rainfall, evaporation, and temperature data; Geological exploration equipment collects data on soil permeability and aquifer thickness; Geographic information system to obtain topographic data; Building health monitoring sensors obtain data on foundation settlement and anti-floating protection water level changes.

10. The AI ​​intelligent groundwater level dynamic prediction system based on multi-source data fusion according to claim 1 is characterized in that: The connection relationship between the modules is: The data acquisition module asynchronously transmits multi-source data to the data fusion module through the message queue. After processing, the latter synchronously pushes the weight results to the AI ​​prediction model module through the API interface. The model prediction results are sent and output in real time through WebSocket.

Images