Water resources allocation management method and system for water conservancy projects based on artificial intelligence

Abstract

The application discloses a water conservancy project water allocation management method and system based on artificial intelligence, relates to the technical field of artificial intelligence, and comprises the following steps: deploying movable ground monitoring equipment and constructing a wireless Mesh self-organizing network, acquiring ground monitoring data, obtaining inspection data through unmanned aerial vehicle inspection of water conservancy facilities, and importing the inspection data into the wireless Mesh self-organizing network; preprocessing the ground monitoring data and the inspection data to obtain a multi-source heterogeneous data set; analyzing the multi-source heterogeneous data set through a water resource scheduling model to generate a water resource scheduling strategy; generating a water resource scheduling instruction based on the water resource scheduling strategy and delivering the water resource scheduling instruction to water conservancy project equipment. The application realizes dynamic scheduling of multi-target optimization through an intelligent decision mechanism based on a genetic algorithm by constructing a mobile monitoring network and a wireless Mesh self-organizing network in air-ground cooperation, and through multi-source heterogeneous data fusion and artificial intelligence analysis, and realizes optimal allocation and efficient utilization of water resources.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and system for water resource allocation and management in water conservancy projects based on artificial intelligence. Background Technology

[0002] In recent years, with the development of Internet of Things (IoT) technology, some water conservancy projects have begun to adopt automated monitoring equipment to transmit data through wired or wireless communication for basic water condition monitoring. This model has improved the automation level of data collection to a certain extent. However, the monitoring equipment is mostly deployed in a fixed manner, which limits the coverage and flexibility. Moreover, the communication method is easily constrained by the geographical environment. In remote or disaster situations, it is difficult to guarantee the stability and real-time performance of data transmission, which cannot meet the needs of modern water conservancy projects for refined and dynamic management. In existing technologies, fixed monitoring networks have limited coverage, making it difficult to detect local anomalies in water conservancy facilities in a timely manner. Traditional data processing methods struggle to effectively integrate heterogeneous data from multiple sources, resulting in severe information silos. Existing technologies often employ simple rule-based logic or static models, lacking in-depth mining and dynamic prediction capabilities of hydrological time-series characteristics. They are unable to adapt to real-time changing scheduling needs, and their communication reliability is insufficient. They also lack self-healing capabilities in the event of node failures, affecting the timely issuance and execution of scheduling instructions. Existing solutions often consider monitoring, communication, and decision-making processes in isolation, lacking an integrated intelligent management architecture, leading to slow response times and making it difficult to achieve precise scheduling and efficient utilization of water resources. Summary of the Invention

[0003] The technical problem addressed by this invention is that fixed monitoring networks have limited coverage, making it difficult to detect local anomalies in water conservancy facilities in a timely manner. Traditional data processing methods struggle to effectively integrate heterogeneous data from multiple sources, leading to severe information silos. Existing technologies often employ simple rule-based logic or static models, lacking in-depth mining and dynamic prediction capabilities of hydrological time-series characteristics. They are unable to adapt to real-time changing scheduling needs, and their communication reliability is insufficient. They also lack self-healing capabilities in the event of node failures, affecting the timely issuance and execution of scheduling instructions. Existing solutions often consider monitoring, communication, and decision-making processes in isolation, lacking an integrated intelligent management architecture, resulting in slow response and difficulty in achieving precise scheduling and efficient utilization of water resources.

[0004] To address the aforementioned technical problems, this invention provides the following technical solution: a water resource allocation and management method for water conservancy projects based on artificial intelligence, comprising the following steps: Step S1: Deploy mobile ground monitoring equipment and build a wireless mesh self-organizing network to acquire ground monitoring data; Step S2: Inspection data is obtained by using drones to inspect water conservancy facilities and then incorporated into the wireless Mesh self-organizing network; Step S3: Preprocess the ground monitoring data and inspection data to obtain a multi-source heterogeneous dataset; Step S4: Analyze the multi-source heterogeneous dataset using the water resource scheduling model to generate a water resource scheduling strategy; Step S5: Generate water resource scheduling instructions based on the water resource scheduling strategy, and send the water resource scheduling instructions to water conservancy engineering equipment to complete water resource allocation and management.

[0005] As a preferred embodiment of the artificial intelligence-based water conservancy project water resource allocation and management method of the present invention, step S1 specifically includes: Deploy mobile ground monitoring equipment to form a mobile monitoring network, and use the mobile monitoring network to monitor the hydrological environment and safety status of water conservancy projects and obtain monitoring data; The monitoring data includes data collected by buoy monitoring stations, data collected by shore-based mobile monitoring robots, and data collected by vehicle-mounted monitoring equipment; The mobile ground monitoring equipment includes buoy-type monitoring stations, shore-based mobile monitoring robots, and vehicle-mounted monitoring equipment; The buoy-type monitoring station is used to monitor the cross-section of reservoirs and rivers to obtain water flow parameters, including water level, flow velocity and water quality; The shoreline mobile monitoring robot is used to inspect along the dike or canal and monitor abnormal conditions, including seepage, deformation and cracks. The vehicle-mounted monitoring equipment is used for temporary monitoring tasks; The construction of the wireless mesh self-organizing network includes: Mesh routing nodes with multi-hop communication capabilities are deployed on each ground monitoring device, each drone, and key locations to form a mesh topology. When any Mesh routing node in the wireless Mesh self-organizing network fails, the remaining Mesh routing nodes can automatically recalculate the route and generate a new path that bypasses the failed Mesh routing node. The key locations include water conservancy hubs and control facilities, key sections of rivers and canals, and communication and geographical high points.

[0006] As a preferred embodiment of the artificial intelligence-based water conservancy project water resource allocation and management method of the present invention, step S2 specifically includes: The drone is controlled to inspect water conservancy facilities according to a preset flight path, and the inspection data is collected and then incorporated into the wireless Mesh self-organizing network. The inspection data includes image data and point cloud data; The water conservancy facilities include dams, spillways, gates, and channels.

[0007] As a preferred embodiment of the artificial intelligence-based water conservancy project water resource allocation and management method described in this invention, step S3 specifically includes: The monitoring data and inspection data are subjected to spatiotemporal alignment, data cleaning and normalization to obtain a multi-source heterogeneous dataset.

[0008] As a preferred embodiment of the artificial intelligence-based water conservancy project water resource allocation and management method of the present invention, step S4 specifically includes: Based on the aforementioned multi-source heterogeneous dataset, a pre-trained water resource scheduling model is used for real-time analysis to generate water resource scheduling strategies. The real-time analysis using a pre-trained water resource scheduling model specifically includes: The water resource scheduling model includes an input layer, a feature extraction layer, a decision layer, and an output layer. The multi-source heterogeneous dataset is received through the input layer; Spatiotemporal features are extracted from the multi-source heterogeneous dataset through the feature extraction layer; The decision layer uses an LSTM network to analyze the extracted spatiotemporal features to obtain a prediction result sequence. Based on the prediction result sequence and the current state data, a water resource scheduling strategy is generated. The output layer is used to output water resource scheduling strategies.

[0009] As a preferred embodiment of the artificial intelligence-based water conservancy project water resource allocation and management method described in this invention, the extraction of spatiotemporal features from the multi-source heterogeneous dataset specifically includes: The feature extraction layer is used to map local ground monitoring equipment and water conservancy facilities into network nodes, and dynamically define the neighborhood relationship between each network node according to the node dynamic neighborhood determination algorithm to form a dynamic graph structure. For each network node in the dynamic graph structure, the original data features collected at the current moment of each network node are used as the initial spatial features of each network node. The raw data features include water level data, water flow data, water quality data, and seepage data; Obtain the spatial features of the neighboring nodes of the target node in the previous iteration from the dynamic graph structure; The spatial features of the neighboring nodes are aggregated, and the aggregation result is weighted and fused with the spatial features of the target node. The initial spatial features of each network node are dynamically updated by iterative aggregation and weighted fusion, and the updated initial spatial features are weighted along the time dimension to generate spatiotemporal features.

[0010] As a preferred embodiment of the artificial intelligence-based water conservancy project water resource allocation and management method described in this invention, the step of using an LSTM network to analyze extracted spatiotemporal features to obtain a prediction result sequence specifically includes: Based on the aforementioned spatiotemporal characteristics, key hydrological parameters within a preset future time period are predicted using an LSTM network to obtain the final prediction result sequence. The key hydrological parameters include water quantity, water demand, and water level change rate; The spatiotemporal feature vector of the current time step T0 and the LSTM output vector of the previous historical time step (T-1) are concatenated to obtain the concatenated input vector. The concatenated input vector is multiplied by the forget gate weights and then added to the forget gate bias to calculate the forgetting coefficient using the sigmoid function. The concatenated input vector is multiplied by the input gate weights and then added to the input gate bias. The input coefficients are then generated using the sigmoid function. The concatenated input vector is input into the candidate cell state weights and candidate cell state biases, and candidate coefficients are generated by the tanh function. The candidate coefficients are multiplied by the input coefficients to obtain the information to be fused; Multiply the cell state of the previous historical time step (T-1) by the forgetting coefficient, and sum the result of the multiplication with the information to be fused to obtain the updated cell state. The concatenated input vector is input into the output gate weights and biases, and the output coefficients are generated by the sigmoid function. The updated cell state is multiplied by the output coefficients to obtain the predicted value of the key hydrological parameters corresponding to the next time step T1. The rolling prediction iteration obtains the predicted values ​​of key hydrological parameters for all time steps within a preset future time period, and then summarizes them to obtain the final prediction result sequence.

[0011] As a preferred embodiment of the artificial intelligence-based water conservancy project water resource allocation and management method of the present invention, the generation of a water resource scheduling strategy based on the prediction result sequence and current state data specifically includes: An optimization method based on the genetic algorithm (GA) is used to generate an optimal water resource scheduling strategy according to the final prediction result sequence and the current state data. The processing logic includes: Obtain current status data of water conservancy facilities, including the opening degree of each gate, the start / stop status of each water pump, the power of each water pump, and the current water storage of the reservoir; Population initialization is performed based on genetic algorithms, and water resource scheduling strategies are encoded into gene chromosomes to randomly generate the initial population. Water resource allocation strategies include the operational objectives, timing, and scale of operations for each water conservancy facility within a pre-defined future time period. The genetic algorithm is used to perform simulation processing to deduce the scheduling strategy represented by each chromosome. The execution result parameters are obtained based on the scheduling strategy represented by each chromosome. The parameters after execution include safety score parameters, water supply guarantee score parameters, and operating cost parameters. The fitness score of each chromosome is calculated based on the post-execution parameters using the comprehensive fitness function. The calculation expression is as follows: F = a1*S1 + a2*S2 - a3*S3; Where: S1 represents the safety score item, S2 represents the water supply guarantee score item, and S3 represents the operating cost item; a1 represents the weight coefficient of the safety scoring item, a2 represents the weight coefficient of the water supply guarantee scoring item, and a3 represents the weight coefficient of the operating cost item. Genetic operations are performed on the current population using a genetic algorithm, including selection, crossover, and mutation, to obtain the next generation population. Based on the iterative optimization of the genetic algorithm, the deduction and genetic operations are repeated until the preset number of iterations or the highest fitness score of the population reaches the convergence threshold. The chromosome with the highest fitness score after the iteration is decoded to obtain the optimal water resource allocation strategy.

[0012] As a preferred embodiment of the artificial intelligence-based water conservancy project water resource allocation and management method of the present invention, step S5 specifically includes: The water resource scheduling strategy generates water resource scheduling instructions, which include target equipment, operation type, operation parameters, and execution time. The water resource scheduling instructions are then distributed to water conservancy engineering equipment bound to the target equipment via the wireless mesh self-organizing network, and executed according to the instructions to complete water resource allocation and management.

[0013] The AI-based water conservancy project water resource allocation management system is applied to the AI-based water conservancy project water resource allocation management method, and includes a monitoring module, an inspection module, a processing module, an analysis module, and a scheduling module. The monitoring module is used to deploy mobile ground monitoring equipment and build a wireless mesh self-organizing network to acquire ground monitoring data; The inspection module is used to obtain inspection data by using drones to inspect water conservancy facilities and to integrate it into the wireless Mesh self-organizing network; The processing module is used to preprocess ground monitoring data and inspection data to obtain multi-source heterogeneous datasets; The analysis module is used to analyze multi-source heterogeneous datasets through a water resource scheduling model and generate water resource scheduling strategies. The scheduling module is used to generate water resource scheduling instructions based on the water resource scheduling strategy, and to send the water resource scheduling instructions to water conservancy engineering equipment to complete water resource allocation and management.

[0014] The beneficial effects of this invention are as follows: By constructing a mobile monitoring network and a wireless mesh self-organizing network that are coordinated between air and ground, this invention achieves dynamic coverage of the entire area of ​​water conservancy projects, effectively solving the problem of blind spots in the coverage of fixed monitoring networks. Through the fusion of multi-source heterogeneous data and artificial intelligence analysis, it overcomes the shortcomings of traditional static models in mining hydrological time-series characteristics. The intelligent decision-making mechanism based on genetic algorithms realizes dynamic scheduling with multi-objective optimization, significantly improving the accuracy and adaptability of water resource scheduling. The self-healing network architecture ensures communication reliability. The integrated design realizes intelligent management of the entire link from data acquisition to command execution, thereby achieving optimized allocation and efficient utilization of water resources. Attached Figure Description

[0015] Figure 1 A schematic diagram of the basic process of a water conservancy project water resource allocation and management method based on artificial intelligence, provided as an embodiment of the present invention.

[0016] Figure 2 This is a basic flowchart of an artificial intelligence-based water resource allocation and management system for water conservancy projects, provided as an embodiment of the present invention. Detailed Implementation

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0018] Example 1, referring to Figure 1 As an embodiment of the present invention, a method and system for water resource allocation and management in water conservancy projects based on artificial intelligence are provided, comprising the following steps: Step S1: Deploy mobile ground monitoring equipment and build a wireless mesh self-organizing network to acquire ground monitoring data.

[0019] Step S2: Inspection data is obtained by using drones to inspect water conservancy facilities and then integrated into the wireless Mesh self-organizing network.

[0020] Step S3: Preprocess the ground monitoring data and inspection data to obtain a multi-source heterogeneous dataset.

[0021] Step S4: Analyze the multi-source heterogeneous dataset using the water resource scheduling model to generate a water resource scheduling strategy.

[0022] Step S5: Generate water resource scheduling instructions based on the water resource scheduling strategy, and send the water resource scheduling instructions to water conservancy engineering equipment to complete water resource allocation and management.

[0023] Deploying mobile ground monitoring equipment and a wireless mesh self-organizing network to acquire ground monitoring data, coupled with drone inspections and data aggregation into the network, solves the problems of fixed monitoring coverage and communication, fills in blind spots, and improves data consistency. After preprocessing the two types of data to obtain multi-source heterogeneous datasets, dynamic scheduling strategies are generated through water resource scheduling model analysis, and then converted into instructions and issued to water conservancy engineering equipment, forming a closed-loop chain. This significantly improves the accuracy and timeliness of water resource allocation, ensuring the safe operation of water conservancy projects and the optimal utilization of water resources.

[0024] Step S1 specifically includes: Deploy mobile ground monitoring equipment to form a mobile monitoring network, and use the mobile monitoring network to monitor the hydrological environment and safety status of water conservancy projects and obtain monitoring data.

[0025] The monitoring data includes data collected by buoy monitoring stations, data collected by shore-based mobile monitoring robots, and data collected by vehicle-mounted monitoring equipment.

[0026] Mobile ground monitoring equipment includes buoy-type monitoring stations, shore-based mobile monitoring robots, and vehicle-mounted monitoring equipment.

[0027] Buoy-type monitoring stations are used to monitor water flow parameters at reservoir and river cross-sections, including water level, flow velocity, and water quality.

[0028] The shoreline mobile monitoring robot is used to inspect embankments or channels and monitor for abnormalities, including seepage, deformation, and cracks.

[0029] Vehicle-mounted monitoring equipment is used for temporary monitoring tasks.

[0030] Building a wireless mesh self-organizing network includes: Mesh routing nodes with multi-hop communication capabilities are deployed on each ground monitoring device, each drone, and key location to form a mesh topology. When any Mesh routing node in the wireless Mesh self-organizing network fails, the remaining Mesh routing nodes can automatically recalculate the route and generate a new path that bypasses the failed Mesh routing node.

[0031] Key locations include water conservancy hubs and control facilities, critical sections of rivers and canals, and communication and geographical high points.

[0032] For key locations, taking mountain reservoir projects as an example, nodes should be deployed within 100 meters of the gate control room at the water conservancy hub. Key sections of the river should be selected at bends with rapid water flow, with a spacing of ≤300 meters. The highest geographical point should be an unobstructed observation platform on a mountaintop to ensure signal coverage of the buoy monitoring area.

[0033] Buoys monitor the center of the reservoir (within a radius of 500 meters), robots patrol 2 kilometers along the dam every 2 hours, and vehicle-mounted equipment handles temporary monitoring after heavy rain. In terms of data priority, robot seepage data > buoy water level data > vehicle-mounted temporary data.

[0034] To overcome the limitations of fixed monitoring coverage, dynamic, full-area monitoring of hydrological environment (water level, flow velocity) and engineering safety (seepage, cracks) is achieved through three types of mobile devices: buoy-based, robotic, and vehicle-mounted. This ensures communication reliability. The mesh topology and automatic routing function of the Mesh self-organizing network can quickly generate new paths when nodes fail, avoiding data transmission interruptions in disasters or remote areas, thus meeting temporary monitoring needs. Vehicle-mounted equipment can flexibly respond to emergencies, supplementing the lack of flexibility in conventional monitoring.

[0035] Step S2 specifically includes: The drone is controlled to inspect water conservancy facilities according to a preset flight path, and the inspection data is collected and integrated into the wireless Mesh self-organizing network.

[0036] Inspection data includes image data and point cloud data.

[0037] Water conservancy facilities include dams, spillways, gates, and canals.

[0038] Preset flight paths by importing regional maps using professional software and setting parameters according to facility type—dam inspection height 20-50 meters, river channel 30-60 meters. Calculate flight path spacing based on camera parameters (1-2 meters in gate area). Set round-trip / circular flight paths along key parts such as the dam crest. Plan river channels along the water flow and cross-section. Add hovering and detour in areas with historical faults, and reduce speed to 2-5 km / h. Preset adjustment rules: pause when wind speed > 5 or visibility < 100 meters, switch to backup flight path when encountering obstacles, and mark at least 3 backup landing points.

[0039] In terms of sampling frequency, the dam is sampled every 10 minutes, and the canal is sampled every 30 minutes.

[0040] It fills the blind spots of ground monitoring, and conducts aerial inspections of facilities such as dams and spillways through preset flight paths, acquiring images and point cloud data to more intuitively discover facility defects that are difficult to detect on the ground, thus improving inspection efficiency. Compared with manual inspection, drones can quickly cover a large area, reducing labor costs and time consumption. They are especially suitable for water conservancy facilities in complex terrain, achieving seamless data integration. Inspection data is directly fed into the Mesh network, avoiding data delays or isolated data caused by separate transmissions, and ensuring the consistency of multi-source data.

[0041] Step S3 specifically includes: The monitoring and inspection data were spatiotemporally aligned, cleaned, and normalized to obtain a multi-source heterogeneous dataset.

[0042] The spatiotemporal alignment algorithm uses the NTP protocol for timestamp synchronization. All devices are calibrated once per hour. The geographic location is uniformly set using the WGS84 coordinate system. UAV data is matched with the coordinates of ground equipment via GPS positioning.

[0043] Data cleaning specifically includes: removing outliers using the 3σ principle and filling in missing data using interpolation.

[0044] Normalization specifically includes mapping data of different dimensions to the [0,1] interval. For water level data (monitoring range 0-50 meters), the Min-Max normalization method is used, and the calculation formula is as follows: ; in, This is the original water level data. This represents the maximum value of the water level data. This is the minimum value of the water level. The water level data is normalized. In this formula, when the original water level data When the value is 0, the calculated result is A value of 0 indicates the lower limit of the normalized data range, meaning the current water level is at the monitoring baseline. Similarly, the same logic applies to water quality data (monitoring range 0-10 mg / L). Set to 0 mg / L, The concentration was set to 10 mg / L, thereby enabling feature fusion of multi-source heterogeneous data at a unified scale.

[0045] Data interference was eliminated by cleaning and removing outliers from monitoring and inspection, ensuring the accuracy of subsequent analysis. The problem of heterogeneous multi-source data was solved by spatiotemporal alignment (unifying data timestamps and geographic location labels) and normalization (unifying data units), which allows data from different devices and of different types to be integrated and analyzed, laying the foundation for model analysis. The pre-processed multi-source heterogeneous dataset can be directly input into the water resource scheduling model, reducing the time cost of model data adaptation.

[0046] Step S4 specifically includes: Based on multi-source heterogeneous datasets, a pre-trained water resource scheduling model is used for real-time analysis to generate water resource scheduling strategies.

[0047] Real-time analysis using a pre-trained water resource scheduling model includes: The water resource scheduling model consists of an input layer, a feature extraction layer, a decision layer, and an output layer.

[0048] The input layer receives heterogeneous datasets from multiple sources.

[0049] Spatiotemporal features are extracted from multi-source heterogeneous datasets through a feature extraction layer.

[0050] The decision layer uses an LSTM network to analyze the extracted spatiotemporal features to obtain a prediction result sequence. Based on the prediction result sequence and the current state data, a water resource scheduling strategy is generated.

[0051] The output layer is used to output water resource scheduling strategies.

[0052] The dataset used for pre-training is selected from the past 5 years of engineering data, including data from the high-water season (June-September), the low-water season (December-February), and the normal-water season, covering two types of parameters: water level and water demand.

[0053] The pre-trained water resource scheduling model adopts a 3-layer hidden layer design with 64 neurons in each layer to balance prediction accuracy and computational efficiency. The Adam optimizer is selected with a learning rate of 0.001 to ensure the stability of parameter updates while avoiding slow convergence or oscillations. The mean squared error (MSE) loss function is used to accurately measure the prediction bias of hydrological parameters (water quantity, water demand, and water level change rate) by calculating the mean of the squared differences between the predicted and actual values. The model validation method adopts a dual mechanism of data partitioning and cross-validation. First, the dataset of nearly 5 years, including the wet season, dry season, and normal season, is divided into a training set (for model parameter learning) and a test set (for generalization ability validation) in an 8:2 ratio. Then, 5-fold cross-validation is performed on the training set—the training set is divided into 5 equal parts, and 4 parts are used for training and 1 part for validation each time. After 5 iterations, the average error is used as the evaluation index of the model training effect to ensure that the model has stable prediction ability under different hydrological scenarios.

[0054] The feature extraction layer outputs a 128-dimensional spatiotemporal feature vector, which is received by the decision layer through TensorFlow tensor format to ensure data dimension matching.

[0055] This approach achieves a structured transformation from data to strategy. The model's input layer, feature extraction layer, decision layer, and output layer have clearly defined functions, forming a complete link from data input to feature mining to strategy generation. This avoids the subjectivity of traditional decision-making, improves the correlation between prediction and decision, and mines the temporal patterns of spatiotemporal features through LSTM networks. This allows scheduling strategies to be generated based on dynamic prediction results rather than static rules, making them more adaptable to hydrological changes and ensuring the model's real-time performance. The pre-trained model can directly analyze the pre-processed data, reducing on-site training time and meeting the timeliness requirements of water resource scheduling.

[0056] Extracting spatiotemporal features from multi-source heterogeneous datasets specifically includes: The feature extraction layer is used to map local ground monitoring equipment and water conservancy facilities into network nodes, and dynamically define the neighborhood relationships between each network node according to the node dynamic neighborhood determination algorithm, forming a dynamic graph structure.

[0057] For each network node in the dynamic graph structure, the raw data features collected at the current moment of each network node are used as the initial spatial features of each network node.

[0058] The raw data features include water level data, water flow data, water quality data, and seepage data.

[0059] Obtain the spatial features of the neighborhood nodes of the target node from the dynamic graph structure after the update in the previous iteration.

[0060] The spatial features of neighboring nodes are aggregated, and the aggregation result is then weighted and fused with the spatial features of the target node.

[0061] Iterative aggregation and weighted fusion are used to dynamically update the initial spatial features of each network node. At the same time, the updated initial spatial features are weighted along the time dimension to generate spatiotemporal features.

[0062] The aggregation results are weighted and fused with the spatial features of the target node. Weights are set according to the importance of node function and data correlation. The initial spatial features of the target node itself are assigned basic weights according to its monitoring accuracy and engineering coreness. The aggregation results of the features of neighboring nodes (i.e., the weighted sum of the mean of monitoring data) are assigned correlation weights according to the spatial distance and data correlation between the neighboring nodes and the target node (the weight of the aggregation result is set to 0.5 for neighbors with a distance < 500 meters and a data correlation > 0.8). The numerical fusion of the two is completed by multiplying the aggregation result by the correlation weight and the target node spatial features by the basic weight, so as to obtain the updated target node spatial features, which lays the foundation for subsequent iterations and time-dimensional weighting.

[0063] The neighborhood determination algorithm is based on Euclidean distance. Monitoring devices with a distance of less than 1 kilometer are defined as neighbors. If the water quality data of a certain device deviates from the surrounding area by more than 20%, then monitoring devices with a distance of less than 2 kilometers are defined as neighbors.

[0064] The time-weighted rule assigns a weight of 0.6 to data from the last 24 hours, 0.3 to data from 24 to 72 hours, and 0.1 to data older than 72 hours. Time-series features are then integrated according to this ratio.

[0065] It accurately captures data correlations and aggregates neighboring nodes through a dynamic graph structure to explore spatial relationships between different monitoring devices and facilities, avoiding isolated data analysis. It integrates spatiotemporal information and performs weighted processing along the time dimension while iterating and updating spatial features, allowing features to reflect both the current state and historical trends, improving the accuracy of subsequent predictions and adapting to dynamically changing scenarios. The neighboring relationships of nodes are dynamically defined by the algorithm and can be adjusted according to changes in hydrological conditions and facility status, avoiding feature lag caused by fixed association rules.

[0066] The prediction result sequence obtained by using an LSTM network to analyze the extracted spatiotemporal features specifically includes: Based on spatiotemporal characteristics, key hydrological parameters within a preset time period are predicted using an LSTM network to obtain the final prediction result sequence.

[0067] Key hydrological parameters include water quantity, water demand, and water level change rate.

[0068] The spatiotemporal feature vector of the current time step T0 and the LSTM output vector of the previous historical time step (T-1) are concatenated to obtain the concatenated input vector.

[0069] The concatenated input vector is multiplied by the forget gate weights and then added to the forget gate bias. The forgetting coefficient is then generated using the sigmoid function.

[0070] The input coefficients are calculated by multiplying the concatenated input vector by the input gate weights and adding it to the input gate bias, and then generating the input coefficients using the sigmoid function.

[0071] The concatenated input vector is fed into the candidate cell state weights and candidate cell state biases, and candidate coefficients are generated using the tanh function.

[0072] The candidate coefficients are multiplied by the input coefficients to obtain the information to be fused.

[0073] Multiply the cell state of the previous historical time step (T-1) by the forgetting coefficient, and sum the result of the multiplication with the information to be fused to obtain the updated cell state.

[0074] The input vector is concatenated with the output gate weights and biases, and the output coefficients are generated using the sigmoid function. The updated cell state is then multiplied by the output coefficients to obtain the predicted values ​​of the key hydrological parameters corresponding to the next time step T1.

[0075] The rolling prediction iteration obtains the predicted values ​​of key hydrological parameters for all time steps within the preset future time period, and then summarizes them to obtain the final prediction result sequence.

[0076] The predicted values ​​of key hydrological parameters are the results of predictions of core water conservancy parameters for a future preset time period by an LSTM network based on spatiotemporal characteristics. Water content, water demand, and water level change rate are presented as vector groups (multiple time step vector groups constitute the prediction sequence). Water content reflects the water body's storage / excess flow, water demand reflects the water demand of various fields, and water level change rate predicts the risk of water level rise and fall. The three together provide a basis for the genetic algorithm to generate the optimal scheduling strategy.

[0077] The preset time period is set to 24 hours for daily dispatch, 48 hours for flood season, and 12 hours if a rainstorm is forecast.

[0078] The processing logic for rolling prediction iteration includes: First, one time step iteration is performed. After the above calculations, the predicted values ​​of key hydrological parameters corresponding to the next time step T1 are obtained. When the time progresses to the next time step T1, the same calculations are performed to obtain the predicted values ​​of key hydrological parameters corresponding to the next two time steps T2. Specifically: When the time progresses to the next time step T1 (i.e., T1 becomes the new current time), the input data is updated. The monitoring data at time T1 is obtained through the wireless mesh self-organizing network. The monitoring data at time T1 is then concatenated with the predicted values ​​of key hydrological parameters at time step T1 to obtain the new concatenated input vector at time T1.

[0079] Repeat the above calculation process of forget gate → input gate → cell state update → output gate to output the predicted value of key hydrological parameters corresponding to time step T2.

[0080] After completing one time step iteration, the iteration is repeated continuously (i.e., rolling prediction iteration) until the prediction of all time steps within the preset future time period is completed. The predicted values ​​of key hydrological parameters of each time step are summarized to obtain the final prediction result sequence.

[0081] To improve the accuracy of hydrological forecasting, the LSTM network's forget gate, input gate, and output gate mechanisms effectively handle the long-term dependencies of hydrological data, avoiding the short-term forecasting limitations of traditional models. The forecasting iteration logic is clearly defined, and rolling forecasting generates a complete parameter sequence for the future preset time period by continuously updating time step data. This provides detailed future operating conditions for scheduling strategies, ensuring forecast interpretability. The calculation process of each gate coefficient is clear, and the generation logic of the forecast results can be traced, facilitating subsequent verification and adjustment.

[0082] Based on the predicted sequence and current status data, the water resource allocation strategy is generated, specifically including: An optimization method based on genetic algorithm (GA) is adopted to generate the optimal water resource scheduling strategy according to the final prediction result sequence and the current state data. The processing logic includes: Obtain current status data of water conservancy facilities, including the opening degree of each gate, the start / stop status of each water pump, the power of each water pump, and the current water storage of the reservoir.

[0083] Population initialization is performed based on genetic algorithms, and water resource scheduling strategies are encoded into gene chromosomes to randomly generate the initial population.

[0084] Water resource allocation strategies include the operational objectives, timing, and scale of operations for each water conservancy facility within a pre-defined future time period.

[0085] The genetic algorithm is used to perform simulation processing to deduce the scheduling strategy represented by each chromosome. The execution result parameters are obtained based on the scheduling strategy represented by each chromosome. The parameters after execution include safety score parameters, water supply guarantee score parameters, and operating cost parameters.

[0086] The fitness score of each chromosome is calculated based on the post-execution parameters using the comprehensive fitness function. The calculation expression is as follows: F = a1*S1 + a2*S2 - a3*S3; Wherein: S1 represents the safety score item, S2 represents the water supply guarantee score item, and S3 represents the operating cost item.

[0087] a1 represents the weight coefficient of the safety scoring item, a2 represents the weight coefficient of the water supply guarantee scoring item, and a3 represents the weight coefficient of the operating cost item.

[0088] Genetic operations are performed on the current population using a genetic algorithm, including selection, crossover, and mutation, to obtain the next generation population.

[0089] Based on the iterative optimization of the genetic algorithm, the process of repeated deduction and genetic operations is carried out until the preset number of iterations or the highest fitness score of the population reaches the convergence threshold is met.

[0090] The chromosome with the highest fitness score after the iteration is decoded to obtain the optimal water resource allocation strategy.

[0091] Chromosome encoding uses real-number encoding to ensure continuous adaptation with scheduling strategy parameters. The chromosome structure is divided according to the facility-operation dimension, with each gene segment corresponding to the scheduling parameters of a single water conservancy facility. The first two genes map to the operation target, the middle three genes map to the operation timing, and the last three genes map to the operation magnitude. In this embodiment, the flood control strategy of a certain gate can be encoded as 10-003-040. Through clear encoding and mapping logic, the scheduling strategy and chromosome are accurately matched, ensuring that population initialization and subsequent decoding can be implemented.

[0092] The acquisition of current status data is based on the principle of real-time adaptation to scheduling requirements. Gate opening and pump start / stop / power are collected once every 1 minute, and reservoir water storage is collected once every 5 minutes to ensure that the data reflects the facility's operating status in real time. The accuracy standard must match the engineering control requirements. The accuracy of gate opening is controlled within ±1%, pump power within ±2%, and reservoir water storage within ±0.5 million cubic meters. Through clear timeliness and accuracy requirements, an accurate data foundation is provided for the simulation and deduction of scheduling strategies, avoiding strategy failure due to data deviation.

[0093] The values ​​of a1, a2, and a3 are determined by combining the core objectives of the project with the actual scenario. The basic value logic is based on the project priority. For example, under normal operating conditions, safety (a1), water supply guarantee (a2), and operating cost (a3) ​​can be set to 0.4, 0.4, and 0.2, respectively, to match the general requirements of prioritizing safety and water supply while taking cost into account. At the same time, they are dynamically adjusted according to the scenario. When focusing on risk prevention and control during the flood season, a1 is increased to 0.5 and a2 is decreased to 0.3 to strengthen the weight of safety objectives. When water resources are scarce during the dry season, a2 is increased to 0.5 and a1 is decreased to 0.3 to prioritize water demand. By clarifying the value selection and dynamic adjustment rules, the multi-objective optimization balance is ensured to fit the actual operating conditions.

[0094] Safety score item S1 is calculated by deducting 10 points for a seepage rate exceeding the dam by 0.01 m / d and 15 points for a water level exceeding the warning level for more than 1 hour, for a total score of 100 points. Water supply guarantee S2 is weighted by 0.4 for agriculture and 40 points for an irrigation guarantee rate of 95% or higher; 0.3 for industry and 10 points for an interruption exceeding 1 hour; and 0.3 for domestic water and 30 points for a water pressure compliance rate of 98% or higher. Operating cost S3 is calculated by adding 5 points for equipment energy consumption exceeding 0.5 kWh / m³ and 8 points for maintenance frequency (more than 2 failures per month). The fitness score is calculated based on these factors.

[0095] To achieve multi-objective optimization, a fitness score is calculated using a comprehensive fitness function (safety, water supply security, cost) to balance the safety, effectiveness, and economy of scheduling, avoiding decision-making bias caused by a single objective and improving strategy optimality. The selection, crossover, and mutation operations of the genetic algorithm can iteratively screen from a large number of candidate strategies and finally output the solution with the highest fitness, which is more scientific than manual decision-making. It is also adapted to dynamic operating conditions, combining the prediction result sequence with the current equipment status (gate opening, water storage, etc.) to generate strategies, ensuring that the strategies meet the real-time operating conditions of actual engineering projects.

[0096] Step S5 specifically includes: The water resource scheduling strategy generates water resource scheduling instructions, which include target equipment, operation type, operation parameters and execution time. The water resource scheduling instructions are then distributed to the water conservancy engineering equipment bound to the target equipment through a wireless mesh self-organizing network and executed according to the instructions to complete the water resource allocation and management.

[0097] The device transmits the result back via the Mesh network within 10 minutes after execution. If the instruction times out and is not executed, it will automatically retry twice. If it still fails, manual intervention will be triggered.

[0098] To achieve seamless integration from strategy to execution, scheduling strategies are transformed into explicit instructions containing target devices, operating parameters, and execution times. This avoids misunderstandings during execution, ensures timely instruction transmission, and delivers instructions via a Mesh network. Combined with the network's self-healing capabilities, instructions are ensured to be quickly delivered to target devices even at faulty nodes, avoiding scheduling delays, clarifying execution responsibilities, and binding instructions to target devices for easy follow-up tracking of execution status, thus improving management traceability.

[0099] Example 2, refer to Figure 2 The water resource allocation and management system for water conservancy projects based on artificial intelligence provided by the present invention includes a monitoring module, an inspection module, a processing module, an analysis module, and a scheduling module.

[0100] The monitoring module is used to deploy mobile ground monitoring equipment and build a wireless mesh self-organizing network to acquire ground monitoring data.

[0101] The inspection module is used to obtain inspection data by using drones to inspect water conservancy facilities and integrate it into the wireless Mesh self-organizing network.

[0102] The processing module is used to preprocess ground monitoring data and inspection data to obtain multi-source heterogeneous datasets.

[0103] The analysis module is used to analyze multi-source heterogeneous datasets through a water resource scheduling model and generate water resource scheduling strategies.

[0104] The scheduling module is used to generate water resource scheduling instructions based on water resource scheduling strategies, and to send the water resource scheduling instructions to water conservancy engineering equipment to complete water resource allocation and management.

[0105] This invention achieves dynamic coverage of the entire water conservancy project area by constructing an air-ground collaborative mobile monitoring network and a wireless mesh self-organizing network, effectively solving the problem of blind spots in fixed monitoring network coverage. Through multi-source heterogeneous data fusion and artificial intelligence analysis, it overcomes the shortcomings of traditional static models in mining hydrological time-series characteristics. The intelligent decision-making mechanism based on genetic algorithms realizes dynamic scheduling with multi-objective optimization, significantly improving the accuracy and adaptability of water resource scheduling. The self-healing network architecture ensures communication reliability, and the integrated design realizes intelligent management of the entire link from data acquisition to command execution, thereby achieving optimized allocation and efficient utilization of water resources.

[0106] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0107] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A water resource allocation and management method for water conservancy projects based on artificial intelligence, characterized in that: Includes the following steps: Step S1: Deploy mobile ground monitoring equipment and build a wireless mesh self-organizing network to acquire ground monitoring data; Step S2: Inspection data is obtained by using drones to inspect water conservancy facilities and then incorporated into the wireless Mesh self-organizing network; Step S3: Preprocess the ground monitoring data and inspection data to obtain a multi-source heterogeneous dataset; Step S4: Analyze the multi-source heterogeneous dataset using the water resource scheduling model to generate a water resource scheduling strategy; Step S5: Generate water resource scheduling instructions based on the water resource scheduling strategy, and send the water resource scheduling instructions to water conservancy engineering equipment to complete water resource allocation and management.

2. The water resource allocation and management method for water conservancy projects based on artificial intelligence as described in claim 1, characterized in that: Step S1 specifically includes: Deploy mobile ground monitoring equipment to form a mobile monitoring network, and use the mobile monitoring network to monitor the hydrological environment and safety status of water conservancy projects and obtain monitoring data; The monitoring data includes data collected by buoy monitoring stations, data collected by shore-based mobile monitoring robots, and data collected by vehicle-mounted monitoring equipment; The mobile ground monitoring equipment includes buoy-type monitoring stations, shore-based mobile monitoring robots, and vehicle-mounted monitoring equipment; The buoy-type monitoring station is used to monitor the cross-section of reservoirs and rivers to obtain water flow parameters, including water level, flow velocity and water quality; The shoreline mobile monitoring robot is used to inspect along the dike or canal and monitor abnormal conditions, including seepage, deformation and cracks. The vehicle-mounted monitoring equipment is used for temporary monitoring tasks; The construction of the wireless mesh self-organizing network includes: Mesh routing nodes with multi-hop communication capabilities are deployed on each ground monitoring device, each drone, and key locations to form a mesh topology. When any Mesh routing node in the wireless Mesh self-organizing network fails, the remaining Mesh routing nodes can automatically recalculate the route and generate a new path that bypasses the failed Mesh routing node. The key locations include water conservancy hubs and control facilities, key sections of rivers and canals, and communication and geographical high points.

3. The water resource allocation and management method for water conservancy projects based on artificial intelligence as described in claim 2, characterized in that: Step S2 specifically includes: The drone is controlled to inspect water conservancy facilities according to a preset flight path, and the inspection data is collected and then incorporated into the wireless Mesh self-organizing network. The inspection data includes image data and point cloud data; The water conservancy facilities include dams, spillways, gates, and channels.

4. The water resource allocation and management method for water conservancy projects based on artificial intelligence as described in claim 3, characterized in that: Step S3 specifically includes: The monitoring data and inspection data are subjected to spatiotemporal alignment, data cleaning and normalization to obtain a multi-source heterogeneous dataset.

5. The water resource allocation and management method for water conservancy projects based on artificial intelligence as described in claim 4, characterized in that: Step S4 specifically includes: Based on the aforementioned multi-source heterogeneous dataset, a pre-trained water resource scheduling model is used for real-time analysis to generate water resource scheduling strategies. The real-time analysis using a pre-trained water resource scheduling model specifically includes: The water resource scheduling model includes an input layer, a feature extraction layer, a decision layer, and an output layer. The multi-source heterogeneous dataset is received through the input layer; Spatiotemporal features are extracted from the multi-source heterogeneous dataset through the feature extraction layer; The decision layer uses an LSTM network to analyze the extracted spatiotemporal features to obtain a prediction result sequence. Based on the prediction result sequence and the current state data, a water resource scheduling strategy is generated. The output layer is used to output water resource scheduling strategies.

6. The water resource allocation and management method for water conservancy projects based on artificial intelligence as described in claim 5, characterized in that: Extracting spatiotemporal features from the multi-source heterogeneous dataset specifically includes: The feature extraction layer is used to map local ground monitoring equipment and water conservancy facilities into network nodes, and dynamically define the neighborhood relationship between each network node according to the node dynamic neighborhood determination algorithm to form a dynamic graph structure. For each network node in the dynamic graph structure, the original data features collected at the current moment of each network node are used as the initial spatial features of each network node. The raw data features include water level data, water flow data, water quality data, and seepage data; Obtain the spatial features of the neighboring nodes of the target node in the previous iteration from the dynamic graph structure; The spatial features of the neighboring nodes are aggregated, and the aggregation result is weighted and fused with the spatial features of the target node. The initial spatial features of each network node are dynamically updated by iterative aggregation and weighted fusion, and the updated initial spatial features are weighted along the time dimension to generate spatiotemporal features.

7. The water resource allocation and management method for water conservancy projects based on artificial intelligence as described in claim 6, characterized in that: The specific steps for obtaining the prediction result sequence by analyzing the extracted spatiotemporal features using an LSTM network include: Based on the aforementioned spatiotemporal characteristics, key hydrological parameters within a preset time period are predicted using an LSTM network to obtain the final prediction result sequence. The key hydrological parameters include water quantity, water demand, and water level change rate; The spatiotemporal feature vector of the current time step T0 and the LSTM output vector of the previous historical time step (T-1) are concatenated to obtain the concatenated input vector. The concatenated input vector is multiplied by the forget gate weights and then added to the forget gate bias to calculate the forgetting coefficient using the sigmoid function. The concatenated input vector is multiplied by the input gate weights and then added to the input gate bias. The input coefficients are then generated using the sigmoid function. The concatenated input vector is input into the candidate cell state weights and candidate cell state biases, and candidate coefficients are generated by the tanh function. The candidate coefficients are multiplied by the input coefficients to obtain the information to be fused; Multiply the cell state of the previous historical time step (T-1) by the forgetting coefficient, and sum the result of the multiplication with the information to be fused to obtain the updated cell state. The concatenated input vector is input into the output gate weights and biases, and the output coefficients are generated by the sigmoid function. The updated cell state is multiplied by the output coefficients to obtain the predicted value of the key hydrological parameters corresponding to the next time step T1. The rolling prediction iteration obtains the predicted values ​​of key hydrological parameters for all time steps within the preset future time period, and then summarizes them to obtain the final prediction result sequence.

8. The water resource allocation and management method for water conservancy projects based on artificial intelligence as described in claim 7, characterized in that: The water resource scheduling strategy is generated based on the predicted result sequence and the current state data, specifically including: An optimization method based on the genetic algorithm (GA) is used to generate an optimal water resource scheduling strategy according to the final prediction result sequence and the current state data. The processing logic includes: Obtain current status data of water conservancy facilities, including the opening degree of each gate, the start / stop status of each water pump, the power of each water pump, and the current water storage of the reservoir; Population initialization is performed based on genetic algorithms, and water resource scheduling strategies are encoded into gene chromosomes to randomly generate the initial population. Water resource allocation strategies include the operational objectives, timing, and scale of operations for each water conservancy facility within a pre-defined future time period. The genetic algorithm is used to perform simulation processing to deduce the scheduling strategy represented by each chromosome. The execution result parameters are obtained based on the scheduling strategy represented by each chromosome. The parameters after execution include safety score parameters, water supply guarantee score parameters, and operating cost parameters. The fitness score of each chromosome is calculated based on the post-execution parameters using the comprehensive fitness function. The calculation expression is as follows: F = a1*S1 + a2*S2 - a3*S3; Where: S1 represents the safety score item, S2 represents the water supply guarantee score item, and S3 represents the operating cost item; a1 represents the weight coefficient of the safety scoring item, a2 represents the weight coefficient of the water supply guarantee scoring item, and a3 represents the weight coefficient of the operating cost item. Genetic operations are performed on the current population using a genetic algorithm, including selection, crossover, and mutation, to obtain the next generation population. Based on the iterative optimization of the genetic algorithm, the deduction and genetic operations are repeated until the preset number of iterations or the highest fitness score of the population reaches the convergence threshold. The chromosome with the highest fitness score after the iteration is decoded to obtain the optimal water resource allocation strategy.

9. The water resource allocation and management method for water conservancy projects based on artificial intelligence as described in claim 8, characterized in that: Step S5 specifically includes: The water resource scheduling strategy generates water resource scheduling instructions, which include target equipment, operation type, operation parameters, and execution time. The water resource scheduling instructions are then distributed to water conservancy engineering equipment bound to the target equipment via the wireless mesh self-organizing network, and executed according to the instructions to complete water resource allocation and management.

10. An artificial intelligence-based water conservancy project water resource allocation and management system, applied in the artificial intelligence-based water conservancy project water resource allocation and management method as described in any one of claims 1-9, characterized in that, It includes a monitoring module, an inspection module, a processing module, an analysis module, and a scheduling module; The monitoring module is used to deploy mobile ground monitoring equipment and build a wireless mesh self-organizing network to acquire ground monitoring data; The inspection module is used to obtain inspection data by using drones to inspect water conservancy facilities and to integrate it into the wireless Mesh self-organizing network; The processing module is used to preprocess ground monitoring data and inspection data to obtain multi-source heterogeneous datasets; The analysis module is used to analyze multi-source heterogeneous datasets through a water resource scheduling model and generate water resource scheduling strategies. The scheduling module is used to generate water resource scheduling instructions based on the water resource scheduling strategy, and to send the water resource scheduling instructions to water conservancy engineering equipment to complete water resource allocation and management.

Images