Wetland group dynamic water distribution method, device and equipment based on multi-modal perception

Abstract

The invention provides a wetland group dynamic water distribution method, device and equipment based on multi-modal perception, and relates to the technical field of dynamic water distribution, and the method comprises the following steps: collecting environmental data of a wetland group in real time through a multi-source sensor network subjected to layout optimization in advance; preprocessing the environment data to obtain standardized environment data; processing the standardized environmental data based on a pre-trained federal learning model, and outputting a water demand predicted value of each wetland; fusing the water demand predicted value and the real-time data into a multi-dimensional feature vector, and dynamically adjusting the feature weight to obtain fused data; and generating a water distribution scheme based on the fused data, and executing the water distribution scheme after simulation verification of the digital twin platform. In the mode, the ecological relevance is improved, the ecological balance is ensured, and resources are saved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of dynamic water distribution, in particular to a wetland group dynamic water distribution method, device and equipment based on multi-modal perception. BACKGROUND

[0002] Artificial wetlands are ecological engineering systems constructed by simulating natural wetlands, which use plants, microorganisms and substrates to cooperatively purify water bodies, and have the functions of water quality improvement, ecological restoration and water resource regulation. Artificial wetland management has an important influence on improving water purification efficiency, optimizing resource utilization and ensuring flood control and drought resistance. The core of artificial wetland management lies in dynamic water distribution, ensuring ecological balance, water purification and flood control and drought resistance, improving resource efficiency and responding to environmental mutations.

[0003] In the related art water distribution, a single wetland independent management mode is usually adopted, ignoring the ecological correlation between wetland groups, resulting in local ecological degradation or resource waste. SUMMARY

[0004] Therefore, the present application aims to provide a wetland group dynamic water distribution method, device and equipment based on multi-modal perception to improve ecological correlation, ensure ecological balance and save resources.

[0005] In the first aspect, the present application provides a wetland group dynamic water distribution method based on multi-modal perception, which collects environmental data of the wetland group in real time through a multi-source sensor network with pre-optimized layout; pre-processes the environmental data to obtain standardized environmental data; processes the standardized environmental data based on a pre-trained federated learning model to output water demand prediction values of each wetland; fuses the water demand prediction values with real-time data into a multi-dimensional feature vector and dynamically adjusts the feature weights to obtain fused data; generates a water distribution scheme based on the fused data, and executes after simulation and verification through a digital twin platform.

[0006] In a preferred embodiment of the present application, the layout optimization method of the multi-source sensor network includes: obtaining geographic coordinates of the wetland area to form a wetland plane; randomly selecting a preset number of initial cluster centers within the range of the wetland plane; the preset number is the number of sensor nodes in the multi-source sensor network; assigning each wetland area geographic coordinate to the nearest cluster center to form multiple clusters, and determining the mean of the geographic coordinates of each point in each cluster as the target cluster center; determining the center change between the target cluster center and the cluster center; when the center change is less than a pre-set change threshold or reaches a maximum iteration number, outputting the optimal arrangement position of the preset number of sensor nodes.

[0007] In the preferred embodiment of the present application, the above-mentioned step of assigning each wetland region geographic coordinate to the nearest cluster center to form multiple clusters and determining the mean of the geographic coordinates of each point in each cluster as the target cluster center comprises: determining, for each wetland region geographic coordinate, the Euclidean distance between the wetland region geographic coordinate and all cluster centers; assigning the wetland region geographic coordinate to the cluster corresponding to the cluster center with the smallest distance based on the Euclidean distance; and determining, for each cluster, the mean of the geographic coordinates of each point in the cluster as the target cluster center.

[0008] In the preferred embodiment of the present application, the above-mentioned step of determining the center change between the target cluster center and the cluster center comprises: re-determining the centroid coordinates of each cluster; and determining the center change between the target cluster center and the cluster center after each re-computation.

[0009] In the preferred embodiment of the present application, the above-mentioned step of preprocessing the environmental data to obtain standardized environmental data comprises: In the preferred embodiment of the present application, the above-mentioned step of preprocessing the environmental data to obtain standardized environmental data comprises:

[0010] In the preferred embodiment of the present application, the above-mentioned step of fusing the water demand prediction value and real-time data into a multi-dimensional feature vector and dynamically adjusting the feature weight to obtain fused data comprises: normalizing and combining the water demand prediction value and real-time data into a multi-dimensional feature vector; and dynamically adjusting the feature weight based on the wetland type to obtain the fused data.

[0011] In the preferred embodiment of the present application, the above-mentioned step of fusing the water demand prediction value and real-time data into a multi-dimensional feature vector and dynamically adjusting the feature weight to obtain fused data comprises: normalizing and combining the water demand prediction value and real-time data into a multi-dimensional feature vector; and dynamically adjusting the feature weight based on the wetland type to obtain the fused data.

[0012] In the preferred embodiment of the present application, the above-mentioned step of fusing the water demand prediction value and real-time data into a multi-dimensional feature vector and dynamically adjusting the feature weight to obtain fused data comprises: normalizing and combining the water demand prediction value and real-time data into a multi-dimensional feature vector; and dynamically adjusting the feature weight based on the wetland type to obtain the fused data.

[0013] In a fourth aspect, the embodiments of the present application further provide a computer readable storage medium, which stores computer executable instructions. When the computer executable instructions are invoked and executed by a processor, the computer executable instructions cause the processor to implement the multi-modal perception based wetland group dynamic water distribution method of the first aspect.

[0014] The embodiments of the present application bring the following beneficial effects: The embodiments of the present application provide a multi-modal perception based wetland group dynamic water distribution method, device and equipment. The multi-source sensor network with pre-optimized layout is used to collect the environmental data of the wetland group in real time. The environmental data is preprocessed to obtain standardized environmental data. The standardized environmental data is processed based on a pre-trained federated learning model to output the water demand prediction value of each wetland. The water demand prediction value and real-time data are fused into a multi-dimensional feature vector and the feature weight is dynamically adjusted to obtain the fused data. The water distribution scheme is generated based on the fused data and is executed after simulation and verification by a digital twin platform. In this way, the ecological correlation is improved, the ecological balance is ensured, and resources are saved.

[0015] Other features and advantages of the present disclosure will be described in the following description, or can be inferred from the description or determined without doubt, or can be known by implementing the above-mentioned technologies of the present disclosure.

[0016] In order to make the above-mentioned purposes, features and advantages of the present disclosure more obvious and easy to understand, the following preferred embodiments are described in detail below, and the accompanying drawings are described as follows. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the following will briefly introduce the drawings needed to be used in the specific embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application. For those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0018] Figure 1 A flowchart of a multi-modal perception based wetland group dynamic water distribution method provided by the embodiments of the present application is shown in the figure; Figure 2 A flowchart of another multi-modal perception based wetland group dynamic water distribution method provided by the embodiments of the present application is shown in the figure; Figure 3 A flowchart of another multi-modal perception based wetland group dynamic water distribution method provided by the embodiments of the present application is shown in the figure; Figure 4 A structural schematic diagram of a multi-modal perception based wetland group dynamic water distribution device provided by the embodiments of the present application is shown in the figure; Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Constructed wetlands are ecological engineering systems built to simulate natural wetlands. They utilize plants, microorganisms, and substrates to purify water, combining water quality improvement, ecological restoration, and water resource regulation. Constructed wetland management has a significant impact on improving water purification efficiency, optimizing resource utilization, and ensuring flood and drought resistance. The core of constructed wetland management lies in dynamic water allocation to ensure ecological balance, synergistic effects between water purification and flood and drought control, improve resource efficiency, and respond to sudden environmental changes.

[0021] In the distribution of water using relevant technologies, a single wetland is usually managed independently, neglecting the ecological connections between wetland groups, which leads to local ecological degradation or waste of resources.

[0022] Based on this, the present invention provides a method, apparatus, and equipment for dynamic water allocation in wetland clusters based on multimodal perception. This method utilizes a pre-optimized multi-source sensor network to collect environmental data from wetland clusters in real time. The environmental data is preprocessed to obtain standardized environmental data. A pre-trained federated learning model processes the standardized environmental data, outputting predicted water demand values ​​for each wetland. These predicted water demand values ​​are then fused with real-time data to form a multi-dimensional feature vector, with dynamic adjustments to the feature weights to obtain fused data. A water allocation scheme is generated based on this fused data and executed after simulation verification using a digital twin platform. This approach improves ecological connectivity, ensures ecological balance, and saves resources.

[0023] To facilitate understanding of this embodiment, a detailed description of a dynamic water allocation method for wetland groups based on multimodal sensing, as disclosed in this embodiment of the invention, will be provided first.

[0024] Example 1 This invention provides a dynamic water allocation method for wetland communities based on multimodal sensing. Figure 1 This is a flowchart illustrating a dynamic water allocation method for wetland communities based on multimodal sensing, provided as an embodiment of the present invention. Figure 1 As shown, the dynamic water allocation method for wetland groups based on multimodal sensing may include the following steps: Step S101, collecting environment data of the wetland group in real time through a multi-source sensor network which is pre-optimized in layout.

[0025] The multi-source sensor network can include a soil humidity monitoring device, a water quality monitoring device, a meteorological monitoring device, and the like.

[0026] The environment data can include soil humidity data, water quality data, meteorological data, and the like.

[0027] Step S102, pre-processing the environment data to obtain standardized environment data.

[0028] Specifically, the pre-processing of the environment data to obtain standardized environment data can include: removing high-frequency noise from the environment data by wavelet decomposition, and reconstructing the de-noised data; after reconstructing the de-noised data, filling in missing data points by cubic polynomial interpolation to obtain standardized data.

[0029] In order to improve the data quality, the environment data can be pre-processed.

[0030] The wavelet decomposition of the environment data to remove high-frequency noise and the reconstruction of the de-noised data can include: removing high-frequency noise from the environment data by wavelet decomposition through the following formula: ; wherein, is the original signal, indicating the input data to be decomposed; k is a translation parameter, indicating the position of the wavelet function on the time axis; is a scale coefficient, generated by a scale function , reflecting the low-frequency component of the signal, ; is a detail coefficient, generated by a wavelet function , .

[0031] The soft threshold value can be applied to remove high-frequency noise from the detail coefficient; wherein, is a noise standard deviation, and N is the number of data points. The de-noised data is reconstructed by using the processed coefficient .

[0032] Specifically, the missing data points are filled in by cubic polynomial interpolation to obtain standardized data, which can include: for the missing data points , based on the data and of the adjacent time points , , a cubic polynomial is constructed: ; the boundary conditions and are satisfied, and the first and second derivatives are continuous.

[0033] Therefore, for the missing points , the interpolation is calculated using the polynomial of the corresponding interval: . After supplementing the missing points, the standardized data is obtained.

[0034] Step S103, processing the standardized environmental data based on the pre-trained federated learning model, outputting the water demand prediction value of each wetland.

[0035] Among them, the federated learning model (LSTM+FedAvg) can be used to predict the water demand of each wetland in the future target time according to the input historical water level, rainfall and evaporation.

[0036] Specifically, the overall structure of the local model can be: ; wherein, , , , are the weight matrices of the LSTM unit, respectively corresponding to the forget gate, the input gate, the output gate and the candidate memory unit; , , , are the bias terms of the LSTM unit; , , are the activation values of the forget gate, the input gate and the output gate, respectively; , is the candidate memory unit and the memory unit at the current time; is the hidden state at the current time.

[0037] Among them, the loss function is represented by mean square error: , which is used to measure the deviation of the predicted water demand from the actual value .

[0038] Step S104, fusing the water demand prediction value and the real-time data into a multi-dimensional feature vector and dynamically adjusting the feature weight to obtain the fused data.

[0039] Specifically, fusing the water demand prediction value and the real-time data into a multi-dimensional feature vector and dynamically adjusting the feature weight to obtain the fused data can include: normalizing and combining the water demand prediction value and the real-time data into a multi-dimensional feature vector; dynamically adjusting the feature weight based on the wetland type to obtain the fused data.

[0040] Among them, each wetland node uploads the local model parameters , and the following formula is weighted and averaged according to the data volume ratio to ensure that the global model reflects the data distribution of all wetlands: ; wherein, Let k be the number of data points in the k-th node. .

[0041] By parameters Add Gaussian noise differential privacy protection. Noise levels are adjusted according to privacy budgets, such as =1.

[0042] Specifically, the preprocessed environmental data is input into the global LSTM model, which outputs a water demand sequence: The Monte Carlo Dropout method is used to calculate the confidence intervals of the predicted values: ;in, To predict the average water demand, Here, M is the standard deviation of the predicted value, M is the number of Monte Carlo samplings, and z is the Z-score. For example, z=1.96 corresponds to a 95% confidence level.

[0043] Specifically, data such as water demand forecasts, real-time water levels, and weather forecasts are normalized and merged into a multi-dimensional feature vector: Then, based on the wetland type, such as ecological zone / agricultural zone / flood control zone, the feature weights are dynamically adjusted.

[0044] Step S105: Generate a water distribution plan based on the fused data, and execute it after simulation and verification through a digital twin platform.

[0045] Among these features, a high-precision three-dimensional hydrological model of the wetland cluster can be constructed in real time using digital twin technology. This model integrates sensor networks (water level, meteorology, water quality) with external data (weather forecasts, remote sensing images) to dynamically perceive the environmental status. After generating a water allocation plan based on a multi-objective optimization model, the plan's effects are simulated using a digital twin platform to verify its feasibility and risks. Water allocation commands are then precisely executed through intelligent gates (PID control) and solar-powered water pumps. In conjunction with the isolated forest algorithm, anomalies (such as equipment failure or sudden rainstorms) are detected in real time, triggering an emergency mode (activating backup water sources and adjusting weights to prioritize the core ecological area). At the same time, a closed-loop feedback mechanism continuously calibrates model parameters to ensure that the system dynamically optimizes with environmental changes, achieving precise, intelligent, and highly robust water resource allocation.

[0046] The dynamic water allocation method for wetland clusters based on multimodal sensing provided in this invention can collect environmental data of wetland clusters in real time through a pre-optimized multi-source sensor network. The environmental data is preprocessed to obtain standardized environmental data. A pre-trained federated learning model processes the standardized environmental data, outputting predicted water demand values ​​for each wetland. These predicted water demand values ​​are then fused with real-time data to form a multi-dimensional feature vector, and the feature weights are dynamically adjusted to obtain fused data. A water allocation scheme is generated based on the fused data and executed after simulation verification through a digital twin platform. This approach improves ecological connectivity, ensures ecological balance, and saves resources.

[0047] Example 2 This invention also provides another dynamic water allocation method for wetland groups based on multimodal sensing; this method is implemented on the basis of the method in the above embodiments; this method focuses on describing the specific implementation of the layout optimization method of the multi-source sensor network.

[0048] Figure 2 A flowchart of another dynamic water allocation method for wetland groups based on multimodal sensing provided in an embodiment of the present invention is shown below. Figure 2 As shown, the layout optimization method for this multi-source sensor network can include the following steps: Step S201: Obtain the geographical coordinates of the wetland area to form a wetland plane.

[0049] The geographical coordinates of the wetland area can be latitude and longitude or two-dimensional plane coordinates.

[0050] Step S202: Randomly select a preset number of initial cluster centers within the wetland plane area; the preset number is the number of sensor nodes in the multi-source sensor network.

[0051] K initial cluster centers are randomly selected within the wetland plane, where K is the number of sensor nodes, which can be determined based on the wetland area and resolution requirements.

[0052] Step S203: Assign the geographic coordinates of each wetland area to the nearest cluster center to form multiple clusters, and determine the mean geographic coordinates of each point in each cluster as the target cluster center.

[0053] The geographic coordinates of each wetland area can be calculated using the following formula. Determine the geographical coordinates of the wetland area With all cluster centers Euclidean distance: The following formula is used to assign the geographic coordinates of wetland areas to the clusters corresponding to the cluster centers with the smallest distances, based on Euclidean distance: For each cluster, the mean geographic coordinates of each point in the cluster are determined as the target cluster center.

[0054] Step S204, determine the center change between the target cluster center and the cluster center.

[0055] Specifically, determining the center change between the target cluster center and the cluster center can include: re-determining the centroid coordinates of each cluster by the following formula: ; wherein, represents the data point set of the jth cluster, and represents the target cluster center; after each re-computation, the center change between the target cluster center and the cluster center is determined by the following formula: ; wherein, and represents the cluster center.

[0056] Step S205, when the center change is less than the pre-set change threshold or reaches the maximum number of iterations, output the optimal arrangement position of a preset number of sensor nodes.

[0057] Wherein, the change threshold can be set to 1-1.5m.

[0058] The wetland group dynamic water distribution method based on multi-modal perception provided by the embodiment of the application adopts the K-means clustering algorithm to optimize the sensor network layout, analyzes the geographic coordinates of the wetland area, distributes the sensor nodes according to the optimal position, maximizes the monitoring coverage range and minimizes the data redundancy. By iteratively updating the cluster center, it ensures that the nodes evenly cover the key areas, improves the data collection efficiency, reduces the communication energy consumption and equipment redundancy, and meets the dynamic monitoring needs of the wetland.

[0059] Embodiment 3 The embodiment of the application also provides another wetland group dynamic water distribution method based on multi-modal perception; the method is implemented on the basis of the above-mentioned embodiment method; the method focuses on the specific implementation mode of generating a water distribution scheme based on fused data and executing after simulation and verification by a digital twin platform.

[0060] Figure 3 The flowchart of another wetland group dynamic water distribution method based on multi-modal perception provided by the embodiment of the application is shown in Figure 3 , which generates a water distribution scheme based on fused data and executes after simulation and verification by a digital twin platform, and can include the following steps: Step S301, generate a Pareto optimal solution set based on the fused data combined with the NSGA-III algorithm.

[0061] The fused data includes real-time data and prediction data. The real-time data can include current water level, soil humidity and water quality indicators, such as PH, dissolved oxygen, etc. The prediction data can include water demand in the future specified time output by the federated learning model, weather forecast, etc. The weather forecast can be rainfall, evaporation, etc. The future specified time can be 24 hours in the future.

[0062] Specifically, population initialization is performed: N candidate solutions are randomly generated, for example, N=100, and each solution includes: water distribution amount , gate opening , water distribution timing .

[0063] Non-dominated sorting is then performed: the solution set is layered based on the objective function value by the following formula: , , , the first layer of non-dominated solutions is screened out: all candidate solutions are traversed, and each solution is compared with other solutions one by one. If solution A is not inferior to solution B in all objective functions, and is strictly superior in at least one objective, solution A is said to dominate solution B. The solution that is not dominated by any other solution is the non-dominated solution. All non-dominated solutions are classified into the first layer of Pareto front, forming the initial optimal solution set.

[0064] Reference points are then generated: reference points are uniformly distributed in the three-dimensional target space (ecology, economy, and society) to ensure solution set diversity. The number of reference points M is represented as: ; wherein H is the number of partitions in each target direction (i.e. the number of uniformly divided intervals), for example: H=4 means that each target dimension is divided into 4 segments; D is the number of objective functions (dimension number) of the optimization problem, for example: D=3 corresponds to three objectives of ecology, economy and society; M is the total number of generated reference points.

[0065] Therefore, when H=4, D=3, , a total of 20 uniformly distributed reference points are generated.

[0066] Finally, crossover and mutation are performed: crossover and mutation generate offspring by simulating binary crossover (SBX) combined with the excellent characteristics of the parent generation to ensure the inheritance of the solution set; polynomial mutation introduces random disturbance to break through local optimum and explore new solution space. The two cooperate to maintain population diversity, avoid premature convergence, and ensure the universality and balanced distribution of the Pareto front solution. Simulated binary crossover (SBX) is used to generate offspring solutions.

[0067] Crossover probability refers to the probability of gene exchange between two parent individuals to generate offspring in genetic algorithms, which is usually 0.5-1; high probability is conducive to promoting the transmission of excellent characteristics and accelerating convergence.

[0068] The mutation probability refers to the probability of random change of individual genes, and the mutation probability (n is the variable dimension) is usually between 0.001 and 0.01; the low probability is conducive to maintaining population diversity and avoiding local optimum.

[0069] In step S302, the target weight is dynamically adjusted through deep reinforcement learning.

[0070] Among them, NSGA-III is used to generate a Pareto solution set, and deep reinforcement learning (PPO) is used to adjust the ecological, economic and social target weights in real time, solve multi-dimensional demand conflicts, and improve the adaptability and balance of the water distribution strategy.

[0071] Among them, the target weight can be dynamically optimized according to real-time environmental changes. Deep reinforcement learning (PPO) can learn the optimal strategy through the definition of state space (such as real-time water level, weather forecast), action space (dynamically adjusting ecological / economic / social target weight) and reward function (vegetation coverage, energy consumption, etc.).

[0072] Among them, the target weight can be an ecological target weight, an economic target weight and a social target weight. Initially, the ecological target, the economic target and the social target are assigned initial weights, for example, the ecological target weight =0.6, the economic target weight =0.3, and the social target weight =0.1.

[0073] Among them, the reward function involves ecological reward, economic penalty and social penalty. For ecological reward: for every 1% increase in vegetation coverage, reward +10; for economic penalty: for every 1kwh increase in energy consumption, penalty-5; for every 1 yuan overspending in irrigation cost, penalty-3; for social penalty: for every 1m increase in water level, penalty-20.

[0074] Specifically, interactive data (dynamic-action-reward) can be collected based on the current strategy, and generalized advantage estimation (GAE) can be used to calculate the action advantage value .

[0075] Therefore, the objective function is maximized , which is represented by the following formula: .

[0076] Among them, is the probability of selecting action by the current strategy (neural network) in state ; is the probability of selecting the same action by the old strategy (the strategy before updating), which is used to constrain the updating range of the strategy; is the advantage function, which measures the action​ The pros and cons of the average strategy (such as the comprehensive improvement of ecological, economic, and social goals); The shear threshold value limits the strategy update step size and prevents training instability.

[0077] Finally, the target weights are adjusted based on the optimized strategy, for example, the initial weights of the ecological, economic, and social goals are adjusted to: ecological goal weight =0.65, economic goal weight =0.25, and social goal weight =0.1.

[0078] Step S303, determine the water distribution scheme based on the Pareto optimal solution set and target weights.

[0079] To make the generated water distribution scheme meet the actual constraint conditions, constraint processing and scheme verification are required. For example, when the water volume exceeds the limit, a penalty is imposed on the solution that violates the constraint, and the penalty term is: ; where is the penalty coefficient, .

[0080] The scheme needs to be verified for feasibility. According to the channel length L and flow rate v, the transmission time is calculated to ensure ; check must be greater than the ecological water demand, i.e. , for example, the mangrove ecological water demand .

[0081] Specifically, select the solution with the highest comprehensive score from the Pareto frontier, for example, the ecological score accounts for 60%; determine the water allocation for each wetland , for example, the mangrove area is 1200m 3 / d, and the agricultural area is 800m 3 / d; and determine the execution parameters, such as gate opening =75%, and the water distribution timing is 6:00~8:00 every day; pre-visualize the scheme effect in the digital twin platform and verify the water level change and ecological impact.

[0082] Step S304, execute the water distribution scheme through the digital twin platform and monitor the execution results, and adjust the Pareto optimal solution set based on the execution results.

[0083] The purpose of dynamic feedback and iterative optimization is to continuously optimize the model based on actual execution results. By monitoring indicators such as water level, vegetation coverage, and energy consumption in real time after execution, the model is updated. If the actual effect deviates from the prediction (such as water level error ≥5%), trigger NSGA-III to regenerate the solution set and update the federated learning model parameters to improve the accuracy of water demand prediction.

[0084] Among them, regarding the digital twin modeling of the digital twin platform: a wetland three-dimensional hydrological model (resolution 1 m x 1 m) can be built based on remote sensing and GIS, integrating soil permeability coefficient, vegetation transpiration rate and other parameters, real-time access to sensor, weather and satellite data, and calibrating the model accuracy (water level error ≤5%) through historical data, establishing a virtual-real synchronous twin platform to support dynamic monitoring and closed-loop control.

[0085] Specifically, three-dimensional hydrological modeling is performed: based on remote sensing images and geographic information systems (GIS), a three-dimensional terrain model of the wetland group is established (resolution 1 m x 1 m). Hydrological parameters are integrated: soil permeability coefficient, vegetation transpiration rate, channel slope, etc.; data access: real-time access to sensor data (water level, soil moisture, water quality), weather forecast (rainfall, evaporation), satellite remote sensing data (NDVI vegetation index); model calibration: use historical data to verify model accuracy (such as water level simulation error ≤5%), adjust parameters (such as Manning coefficient).

[0086] Specifically, real-time data monitoring and data synchronization are performed: data collection, sensor network uploads data (water level, soil moisture, equipment status) every 5 minutes; data cleaning, wavelet decomposition and cubic polynomial are used for denoising and interpolation processing; digital twin synchronization, updating real-time water level, vegetation coverage and other parameters in the three-dimensional model, generating a visual interface.

[0087] Specifically, optimal control instructions can be generated based on the digital twin simulation results, including: scenario simulation: input the water distribution scheme generated by the multi-objective optimization model to simulate the water level change, vegetation response and energy consumption within the next 6 hours; risk assessment: detect potential risks (such as water level exceeding or channel blockage); generate instructions: if the simulation results meet the constraint conditions, output the execution instructions (gate opening, pump flow); otherwise, trigger the optimization model to recalculate.

[0088] Specifically, decision instructions can be converted into physical device actions, including: intelligent gate control, set target water level , calculate error by the following formula , control output by the following formula: , where , , , adjust gate opening (accuracy 1 cm) to make the actual water level approach the target value; solar water pump control, adjust flow according to water distribution : , flow = Qitirrigation (such as 3 = 1200 m / d, t = 8 h, flow = 150 m 3 / d).3 Dynamic adjustment of water pump speed (precision 5%), matching water distribution timing.

[0089] Specifically, anomaly detection and emergency response can be performed, specifically including: Regarding anomaly detection, the Isolation Forest algorithm is adopted: build Isolation Tree (iTree): randomly select features and split values, recursively divide the data space into sub-regions until each sub-region contains only one sample or reaches the tree height limit; anomaly point characteristics: due to large differences from most data, anomaly points are usually isolated at a shallow level (shorter path); calculate path length: record the average path length required to isolate each sample in all isolation trees ; the shorter the path length, the higher the sample anomaly probability; calculate anomaly score: analyze sensor data stream, calculate anomaly score: , where is the average path length of the data point in the isolation tree, is the path length normalization factor, and a score close to 1 is determined as an anomaly, and a score close to 0 is normal. Specifically, if the anomaly score is > 0.7, it is determined as an abnormal event (such as affected by heavy rain, equipment failure, etc.).

[0090] Regarding emergency mode triggering: start backup water source, automatically open underground water well or water storage tank to ensure core ecological area water supply; conservation strategy switching, mangrove protection area: maintain minimum ecological water demand =1000m3 / d, agricultural area: water distribution amount reduced to 50% (such as =400m3 / d); manual intervention notification, send alarm to administrator, such as through APP push and SMS sending.

[0091] Specifically, system parameters can be optimized according to actual execution effect: Effect evaluation: compare the deviation of digital twin prediction value and actual water level, vegetation coverage (allowable error ≤5%); parameter calibration: if the deviation is out of limit, adjust the hydrological model parameters (such as permeability coefficient) or federal learning model weight; model retraining: collect new data, update LSTM water demand prediction model (full training once every 7 days).

[0092] Specifically, human-computer interaction interface can be provided to support decision monitoring: Cloud platform: WebGL three-dimensional interface displays real-time water level, water distribution scheme simulation, and device status; historical data backtracking (such as water level change curve in the past 30 days). Mobile APP: real-time receive warning (such as "mangrove area water level below threshold"); remote manual control (such as forced closing of a gate).

[0093] The wetland group dynamic water distribution method based on multi-modal perception provided by the embodiment of the present application adopts NSGA-III to generate a Pareto solution set, combines deep reinforcement learning (PPO) to adjust the ecological, economic and social target weights in real time, solves the multi-dimensional demand conflict, improves the adaptability and balance of the water distribution strategy, constructs a high-precision three-dimensional hydrological model, integrates the PID algorithm and the solar equipment to accurately execute the water distribution instruction, dynamically calibrates the model through real-time data feedback, realizes the whole-process closed-loop management of "perception-decision-execution-optimization", and realizes the whole-process closed-loop management of "perception-decision-execution-optimization". Based on the isolated forest algorithm, the data anomalies (such as heavy rain and equipment failure) are detected in real time, the standby water source is automatically started, and the water distribution strategy is adjusted, so that the core ecological area is supplied with water, and the risk resistance and emergency response efficiency of the system are enhanced.

[0094] Embodiment 4 Corresponding to the method embodiment described above, the embodiment of the present application provides a wetland group dynamic water distribution device based on multi-modal perception, Figure 4 The structure diagram of the wetland group dynamic water distribution device based on multi-modal perception provided by the embodiment of the present application is shown as Figure 4 The wetland group dynamic water distribution device based on multi-modal perception can include: The environmental data real-time acquisition module 401 is used to acquire the environmental data of the wetland group in real time through the multi-source sensor network optimized in advance.

[0095] The environmental data preprocessing module 402 is used to preprocess the environmental data to obtain standardized environmental data.

[0096] The water demand prediction value output module 403 is used to process the standardized environmental data based on the pre-trained federated learning model, and output the water demand prediction value of each wetland.

[0097] The data fusion module 404 is used to fuse the water demand prediction value and real-time data into a multi-dimensional feature vector and dynamically adjust the feature weight to obtain the fused data.

[0098] The water distribution scheme generation module 405 is used to generate a water distribution scheme based on the fused data, and execute after simulation and verification through a digital twin platform.

[0099] The wetland group dynamic water distribution device based on multi-modal perception provided by the embodiment of the application can collect environmental data of the wetland group in real time through a multi-source sensor network that is pre-optimized in layout, pre-process the environmental data to obtain standardized environmental data, process the standardized environmental data based on a pre-trained federated learning model, output water demand prediction values of each wetland, fuse the water demand prediction values and real-time data into a multi-dimensional feature vector and dynamically adjust feature weights to obtain fused data, generate a water distribution scheme based on the fused data, and execute the water distribution scheme after simulation and verification through a digital twin platform. In this way, the ecological correlation is improved, the ecological balance is ensured, and resources are saved.

[0100] In some embodiments, the environmental data real-time acquisition module is further configured to obtain geographical coordinates of the wetland areas to form a wetland plane, randomly select a preset number of initial cluster centers within the wetland plane, set the preset number as a number of sensor nodes in the multi-source sensor network, assign each geographical coordinate of the wetland areas to the nearest cluster center to form a plurality of clusters, and determine a mean value of geographical coordinates of each point in each cluster as a target cluster center. The environmental data real-time acquisition module is further configured to determine a center change amount between the target cluster center and the cluster center, and output an optimal arrangement position of the preset number of sensor nodes when the center change amount is less than a pre-set change threshold or a maximum iteration number is reached.

[0101] In some embodiments, the environmental data real-time acquisition module is further configured to determine, for each geographical coordinate of the wetland areas, an Euclidean distance between the geographical coordinate of the wetland areas and all cluster centers, assign the geographical coordinate of the wetland areas to a cluster corresponding to a cluster center with the smallest distance based on the Euclidean distance, and determine, for each cluster, a mean value of geographical coordinates of each point in the cluster as a target cluster center.

[0102] In some embodiments, the environmental data real-time acquisition module is further configured to re-determine a centroid coordinate of each cluster, and determine a center change amount between the target cluster center and the cluster center after each re-computation.

[0103] In some embodiments, the environmental data preprocessing module is further configured to remove high-frequency noise from the environmental data by wavelet decomposition and reconstruct the denoised data, and fill in missing data points by cubic polynomial interpolation after reconstructing the denoised data to obtain standardized data.

[0104] In some embodiments, the environmental data preprocessing module is further configured to construct a cubic polynomial based on data of adjacent time points for the missing data points, determine an interpolation based on the cubic polynomial and a pre-set boundary condition, and fill in the missing data points by the interpolation to obtain the standardized data.

[0105] In some embodiments, the data fusion module is further configured to normalize and combine the water demand prediction value and real-time data into a multi-dimensional feature vector; dynamically adjust the feature weight based on the wetland type to obtain the fused data.

[0106] The device provided by the embodiments of the present application has the same implementation principle and generated technical effects as the foregoing method embodiments, and for brevity of description, the part not mentioned in the device embodiment can be referred to the corresponding content in the foregoing method embodiments.

[0107] Embodiment 5 The embodiments of the present application further provide an electronic device for running the wetland group dynamic water distribution method based on multi-modal perception. Figure 5 As shown in the structural schematic diagram of an electronic device, the electronic device comprises a memory 500 and a processor 501, wherein the memory 500 is configured to store one or more computer instructions, and the one or more computer instructions are executed by the processor 501 to implement the wetland group dynamic water distribution method based on multi-modal perception.

[0108] Further, Figure 5 The electronic device further comprises a bus 502 and a communication interface 503, and the processor 501, the communication interface 503 and the memory 500 are connected through the bus 502.

[0109] The memory 500 can include a high-speed random access memory (RAM) and can also include a non-volatile memory such as at least one disk memory. The communication between the system network element and at least one other network element is realized through at least one communication interface 503 (which can be wired or wireless), and the Internet, a wide area network, a local area network, a metropolitan area network, etc. can be used. The bus 502 can be an ISA bus, a PCI bus or an EISA bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, Figure 5 Only one bidirectional arrow is used in the figure, but it does not mean that there is only one bus or only one type of bus.

[0110] The processor 501 can be an integrated circuit chip having a signal processing capability. In the implementation process, each step of the above method can be completed by the integrated logic circuit of hardware in the processor 501 or the instruction in the form of software. The processor 501 described above can be a general processor, including a central processing unit (CPU), a network processor (NP), etc.; can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. Each method, step and logic block diagram disclosed in the embodiment of the present application can be implemented or executed. The general processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in combination with the embodiment of the present application can be directly embodied as a hardware code processor for execution, or a combination of hardware and software modules in the code processor for execution. The software module can be located in a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium in the storage medium 500 is read by the processor 501, and the hardware thereof is combined to complete the steps of the method of the above embodiment.

[0111] The embodiment of the present application also provides a computer readable storage medium, the computer readable storage medium stores computer executable instructions, when the computer executable instructions are called and executed by the processor, the computer executable instructions cause the processor to implement the wetland group dynamic water distribution method based on multi-modal perception described above. For specific implementation, please refer to the method embodiment, which will not be repeated here.

[0112] The computer program product for implementing the wetland group dynamic water distribution method based on multi-modal perception provided by the embodiment of the present application includes a computer readable storage medium storing non-volatile program codes executable by the processor. The instructions included in the program codes can be used to execute the method in the foregoing method embodiment. For specific implementation, please refer to the method embodiment, which will not be repeated here.

[0113] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system, device and unit described above can refer to the corresponding process in the foregoing method embodiment, which will not be repeated here.

[0114] In several embodiments provided by the present application, it should be understood that the disclosed system, device and method can be implemented in other manners. The described device embodiments are merely schematic, and for example, the division of the units is only a logical function division, and there can be another division manner in actual implementation; for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between different units, or the among different units, can be indirect couplings or communication connections through some interfaces, devices or units, and can be in electric, mechanical or other forms.

[0115] The units described as separated components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed on a plurality of network units. In actual implementation, some or all of the units can be selected according to the actual needs to achieve the purposes of the embodiments of the present application.

[0116] In addition, each functional unit in the various embodiments of the present application can be integrated in one processing unit, or each unit can exist physically as a separate unit, or two or more units can be integrated in one unit.

[0117] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer readable storage medium of a processor. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art, or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in the various embodiments of the present application. The foregoing storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk, and various media that can store program codes.

[0118] Finally, it should be noted that the above examples are merely specific embodiments of the present application, and are used to illustrate the technical solutions of the present application, but are not intended to limit the present application. The protection scope of the present application is not limited thereto. Although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that any person skilled in the art can still modify or easily think of changes to the technical solutions recorded in the foregoing examples, or make equivalent replacements to some of the technical features, within the technical range disclosed by the present application. These modifications, changes or replacements do not cause the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A multi-modal perception based dynamic water distribution method for wetland communities, characterized in that, The method comprises: Real-time collection of environmental data of a wetland group by a multi-source sensor network with pre-optimized layout; Preprocessing of the environmental data to obtain standardized environmental data; Processing of the standardized environmental data based on a pre-trained federated learning model to output a water demand prediction value of each wetland; Fusion of the water demand prediction value and real-time data into a multi-dimensional feature vector and dynamic adjustment of feature weights to obtain fused data; Generation of a water distribution scheme based on the fused data and execution after simulation and verification by a digital twin platform.

2. The method of claim 1, wherein, The layout optimization method of the multi-source sensor network comprises: Obtaining geographical coordinates of a wetland area to form a wetland plane; Randomly selecting a preset number of initial cluster centers within the range of the wetland plane; the preset number is the number of sensor nodes in the multi-source sensor network; Assigning each geographical coordinate of the wetland area to the nearest cluster center to form multiple clusters, and determining the mean of the geographical coordinates of each point in each cluster as a target cluster center; Determining the center change amount between the target cluster center and the cluster center; When the center change amount is less than a pre-set change amount threshold or a maximum iteration number is reached, outputting the optimal arrangement positions of the preset number of sensor nodes.

3. The method of claim 2, wherein, The method of assigning each geographical coordinate of the wetland area to the nearest cluster center to form multiple clusters, and determining the mean of the geographical coordinates of each point in each cluster as a target cluster center, comprises: For each geographical coordinate of the wetland area, determining the Euclidean distance between the geographical coordinate of the wetland area and all cluster centers; Based on the Euclidean distance, assigning the geographical coordinate of the wetland area to the cluster corresponding to the cluster center with the smallest distance; For each cluster, determining the mean of the geographical coordinates of each point in the cluster as a target cluster center.

4. The method of claim 3, wherein, The method of determining the center change amount between the target cluster center and the cluster center comprises: Redetermining the centroid coordinates of each cluster; After each recalculation, determining the center change amount between the target cluster center and the cluster center.

5. The method of claim 1, wherein, The method of preprocessing the environmental data to obtain standardized environmental data comprises: Using wavelet decomposition to remove high-frequency noise from the environmental data, and reconstructing the denoised data; After reconstructing the denoised data, filling in missing data points by cubic polynomial interpolation to obtain standardized data.

6. The method of claim 5, wherein, The method of filling in missing data points by cubic polynomial interpolation to obtain standardized data comprises: For the missing data points, constructing a cubic polynomial based on the data of adjacent time points; Determining the interpolation based on the cubic polynomial and a pre-set boundary condition; Filling in the missing data points by the interpolation to obtain the standardized data.

7. The method of claim 1, wherein, The method of fusing the water demand prediction value and real-time data into a multi-dimensional feature vector and dynamically adjusting feature weights to obtain fused data comprises: Normalizing and combining the water demand prediction value and real-time data into a multi-dimensional feature vector; Dynamically adjusting feature weights based on the type of wetland to obtain fused data.

8. A multi-modal perception based dynamic water distribution device for wetland population, characterized in that, The device comprises: An environmental data real-time acquisition module is configured to collect environmental data of the wetland group in real time through a multi-source sensor network that is pre-optimized in layout; An environmental data preprocessing module is configured to preprocess the environmental data to obtain standardized environmental data; A water demand prediction value output module is configured to process the standardized environmental data based on a pre-trained federated learning model and output water demand prediction values of each wetland; A data fusion module is configured to fuse the water demand prediction values and real-time data into a multi-dimensional feature vector and dynamically adjust feature weights to obtain fused data; A water distribution scheme generation module is configured to generate a water distribution scheme based on the fused data and execute the scheme after simulation verification through a digital twin platform.

9. An electronic device, comprising: A processor and a memory are included, the memory stores computer executable instructions that can be executed by the processor, and the processor executes the computer executable instructions to implement the multi-modal perception based dynamic water distribution method for wetland groups according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer executable instructions, and when the computer executable instructions are called and executed by the processor, the computer executable instructions cause the processor to implement the multi-modal perception based dynamic water distribution method for wetland groups according to any one of claims 1 to 7.

Images