CN-122010283-A - Urban shallow lake ecological protection restoration method and system integrating intelligent control

CN122010283ACN 122010283 ACN122010283 ACN 122010283ACN-122010283-A

Abstract

The invention discloses an integrated intelligent control urban shallow lake ecological protection restoration method and system, which relate to the technical field of water environment ecological management and intelligent water affairs, and comprise the steps of synchronously acquiring water quality, images and meteorological data by constructing a three-dimensional monitoring network, and generating a multi-mode data set through fusion; the method comprises the steps of utilizing a deep learning model to realize pollution tracing, water quality prediction and algal bloom early warning in parallel, outputting a multi-facility cooperative regulation strategy through a reinforcement learning agent based on a prediction result, converting the strategy into a grading instruction to drive an execution unit, and feeding back the treated environment state to the model and the agent to form closed loop optimization. The invention realizes the whole-course intelligent control from monitoring to execution, effectively improves the systematicness, the accuracy and the foresight of lake treatment, and remarkably enhances the water environment risk coping capacity and the ecological restoration effect.

Inventors

YANG FENGJUAN
HONG CHANGHONG
LIU DA
WANG ZHEN
Mai Yingwen
Xiao Gengfeng
LIU SHUAI

Assignees

广东省水利水电科学研究院

Dates

Publication Date: 20260512
Application Date: 20260403

Claims (9)

1. An integrated intelligent control ecological protection and restoration method for a shallow urban lake is characterized by comprising the following steps of: Constructing a three-dimensional monitoring network, synchronously acquiring water quality parameters, water body images and meteorological data of a target lake, and carrying out space-time alignment and fusion on the acquired multi-source heterogeneous data to generate a lake ecological state multi-mode data set; Analyzing the multi-mode data set by using a deep learning model, and executing tasks including dynamic tracing pollution based on a graph neural network, generating a water quality distribution map of a preset period based on a space-time prediction model, and generating algal bloom occurrence probability and path early warning based on sequence learning in parallel; based on the prediction and early warning results, decision optimization is carried out through the reinforcement learning intelligent agent, the state space of the reinforcement learning intelligent agent comprises prediction water quality and algal bloom risk information, the action space comprises control instructions for at least one of aeration equipment, an ecological water gate regulating pump and a carbon source adding device, and an optimal strategy of multi-facility cooperative regulation is output through interaction with environment simulation; And converting the optimal strategy into a hierarchical control instruction, sending the hierarchical control instruction to a corresponding execution unit, collecting the executed environmental state change through a monitoring network, feeding back to a deep learning model and a reinforcement learning intelligent agent, and performing closed-loop optimization.
2. The integrated intelligent control urban shallow lake ecological protection and restoration method according to claim 1, wherein the method is characterized by constructing a three-dimensional monitoring network, synchronously collecting water quality parameters, water body images and meteorological data of a target lake, performing space-time alignment and fusion on the collected multi-source heterogeneous data, and generating a lake ecological state multi-mode data set, and comprises the following steps: simulating by using historical hydrologic data and a hydrodynamic model, identifying key hydrologic nodes of a target lake, taking the key hydrologic nodes as basic anchor points of a three-dimensional monitoring network, and arranging integrated intelligent monitoring buoys at each anchor point to construct intelligent sensing nodes; Establishing communication links for the intelligent sensing node and the weather station node through the mobile self-organizing network, generating a dynamic self-organizing network which is decentralised and has self-healing capacity, and constructing a three-dimensional monitoring network; The method comprises the steps that multispectral image data of a target lake are obtained based on satellites, satellite downlink synchronous beacon signals are sent to a ground receiving station, the ground receiving station broadcasts the synchronous beacon signals to a three-dimensional monitoring network, a node is triggered to sample, and an unmanned aerial vehicle group is instructed to cooperatively lift up and shoot, so that three-dimensional laser data and multispectral imaging data are obtained; Generating a live-action three-dimensional model comprising lake bottom topography, a shoreline structure and engineering facility positions through the three-dimensional laser data and the multispectral imaging data, generating a digital twin base of a target lake, identifying and dividing a visual characteristic region based on the multispectral imaging data and the water body image, wherein the visual characteristic region comprises an algae gathering region, a turbid water group and a clear water region, acquiring information of the visual characteristic region, correlating and checking corresponding water quality parameters, and establishing a quantitative mapping relation library of visual characteristics and water quality parameters; Binding water quality parameters, water body images, meteorological data, time stamps and coordinates acquired by the nodes, integrating the data in a quantitative mapping relation library into the live-action three-dimensional model, constructing a three-dimensional geographic information layer, a space-time continuous water quality parameter distribution layer, a water body visual characteristic and event labeling layer, a meteorological driving factor layer and an ecological engineering setting state layer, and generating a structured lake ecological state multi-mode data set.
3. The method for ecologically protecting and repairing the urban shallow lake by integrated intelligent control according to claim 1, wherein the method for dynamically tracing pollution based on the graph neural network comprises the following steps: meshing and dividing the target lake, taking each grid as a graph node, arranging key nodes at a river mouth, a reservoir area center, a sensitive area and known endogenous pollution points which are directly put in storage, and taking water quality parameters, underwater image characteristics and meteorological data of each grid as dynamic attribute vectors; Inverting a flow field path according to the buoy track, screening nodes with hydrodynamic relations based on the flow field path, establishing directed edges, and constructing dynamic edge weights according to real-time wind speed, wind direction and node distances; Modeling the constructed graph structure space dependence through a graph attention network, performing attention guidance by taking the current wind field and flow field direction as priori knowledge, judging a possible pollutant diffusion path according to the priori knowledge, giving higher attention weight to neighbor nodes on the pollutant diffusion path, and acquiring the polluted space dependence characteristic according to the graph attention network output; analyzing time sequence data of each node by using a gating time sequence convolution network, acquiring a time sequence mode related to an endogenous pollution point, and acquiring time dependence characteristics of pollution; And dynamically simulating the pollution process in the lake by using the space-dependent features and the time-dependent features, training by taking water quality data of key nodes as supervision signals, calculating dynamic contribution rates by using a trained network based on a shielding method, respectively shielding estuary nodes and node groups marked as potential endogenous areas, and calculating contribution rates of estuary exogenous input and endogenous loads of different lake areas by the change of predicted values.
4. The integrated intelligent control urban shallow lake ecological protection restoration method according to claim 1, wherein generating a water quality distribution map of a preset period based on a space-time prediction model comprises: Meshing and dividing the target lake, and encoding sensor data, underwater images and meteorological data of the meshing blocks to generate physical feature vectors comprising space contexts, visual feature vectors representing regional image features and meteorological driving feature vectors representing meteorological fields; Inside each grid, importance weighting is carried out on different data sources of the same module, cross-modal attention is introduced, dynamic fusion weights are calculated according to information correlation, and data fusion is carried out according to the dynamic fusion weights to obtain high-dimensional characteristics; The method comprises the steps of importing a high-dimensional characteristic sequence of a historical period into a condition to generate an countermeasure network, acquiring a time-space evolution trend of a lake state by using downsampling, importing the time-space evolution trend into a decoder, recovering space details by upsampling and jump connection, and generating a super-resolution water quality distribution map of each time step of a future preset period by adopting a rolling prediction mechanism.
5. The integrated intelligent control urban shallow lake ecological protection restoration method according to claim 1, wherein the sequence learning-based algal bloom occurrence probability and path early warning comprises the following steps: Cutting a plurality of complete algal bloom life cycle sequences from historical data by taking an algal bloom event as a center, and marking scene meta-features comprising event types, active driving factors and event intensities based on the algal bloom life cycle sequences in each algal bloom time; based on a multi-task learning model of an algal bloom event training encoder-decoder framework with scene element characteristics, inputting current and past multi-mode ecological environment sequences, and extracting comprehensive state vectors containing algal bloom potential driving forces by using an encoder; The comprehensive state vector is shared by an occurrence probability decoder, a scale strength decoder and a diffusion path decoder, the occurrence probability of algal bloom at each time point in the future is judged through the occurrence probability decoder, the area occupation ratio and the chlorophyll a peak concentration which are possibly influenced in the future are output through the scale strength decoder, the predicted wind field and flow field data are combined through the diffusion path decoder, and the dominant drifting direction and the influence range of algae Chinese group are simulated through a space-time attention module.
6. The integrated intelligent control urban shallow lake ecological protection restoration method according to claim 1, wherein decision optimization is performed by a reinforcement learning agent based on prediction and early warning results, and a state space of the reinforcement learning agent contains predicted water quality and algal bloom risk information, comprising: introducing pollution tracing prediction results, a water quality distribution diagram of a preset period, algal bloom occurrence probability and a path early warning result into a reinforcement learning intelligent body to construct a lake state space, wherein an action space is defined as a discretization or piecewise continuous control instruction combination of aeration equipment, an ecological water gate regulating pump and a carbon source adding device, and a hierarchical structure is arranged on the control instruction in the action space; Constructing a multi-objective rewarding function according to the water quality rewarding, the ecological risk rewarding, the economic cost rewarding and the engineering toughness rewarding, adopting Dueling DQN architecture, decomposing the Q value into a state value function and a dominant function, aggregating the output of the state value function and the dominant function to obtain a final Q value, and determining the critical degree of the current lake ecological state and the treatment action capable of bringing the most extra improvement by the reinforcement learning agent through the final Q value; in the training process, calculating the time difference error of each experience, setting sampling priority according to the time difference error of an experience sample, and strengthening learning of an intelligent agent in wrong experience and unexpected experience through priority experience playback so as to accelerate convergence to an optimal strategy; Setting a safety rule in the trained reinforcement learning intelligent agent, checking the generated action by using the safety rule, and outputting an optimal strategy of multi-facility cooperative regulation after checking.
7. The integrated intelligent control urban shallow lake ecological protection restoration method according to claim 1, wherein the method is characterized in that the optimal strategy is converted into a hierarchical control instruction and sent to a corresponding execution unit, and comprises the following steps: Presetting a semantic action dictionary, mapping original action values in an optimal strategy output by an reinforcement learning agent to a predefined standardized operation mode, generating a hierarchical control instruction, introducing fuzzy logic smoothing, defining a critical area of an input variable as a transition area, and carrying out smooth weighted transition on the instruction in the transition area; Constructing a dynamic multistage response threshold library, dynamically adjusting a threshold according to seasonality, hydrology period and equipment state, and periodically updating the semantic action dictionary; And issuing a hierarchical control instruction to each execution unit through an industrial Internet of things protocol, automatically associating subsequent monitoring data after issuing the instruction, analyzing the change trend of a preset monitoring index, comparing the change trend with an expected effect, and generating an abnormal alarm according to a comparison result.
8. The integrated intelligent control ecological protection and restoration method for urban shallow lakes according to claim 1, characterized in that the environmental state changes after execution are collected through a monitoring network and fed back to a deep learning model and a reinforcement learning agent for closed-loop optimization, comprising: creating an intervention event log according to the executed instruction content, the starting and ending time, the acting geographical range and the expected treatment target, and taking each intervention event as a center, and extracting a preposed state sequence, a postposed response sequence and comparison area data before and after intervention as environment feedback data; The method comprises the steps of constructing intervention events and environmental feedback data in a preset period as a fine adjustment task set, fine adjusting deep learning models of different tasks, evaluating the importance of model parameters in the fine adjustment process, and applying elastic constraint to the model parameters higher than an importance threshold value to prevent the parameters from being changed severely; The agent model in the simulator is subjected to fine adjustment by using a fine adjustment task set, an intervention event is taken as input, the environment response sequence output by the simulator is required to be close to the response sequence actually monitored, and the simulator is calibrated; And in the experience playback library of the reinforcement learning intelligent agent, storing the real decision and the execution process as special experiences in the playback library, setting the corresponding priority as the highest, and performing periodic retraining of the reinforcement learning intelligent agent.
9. An integrated intelligent control ecological protection and restoration system for the urban shallow lake is characterized by being used for realizing the integrated intelligent control ecological protection and restoration method for the urban shallow lake according to any one of claims 1-8, and comprises a three-dimensional monitoring network unit, a multi-mode data fusion and preprocessing unit, a multi-task deep learning prediction unit, an reinforcement learning intelligent decision unit and a hierarchical execution and control unit; The three-dimensional monitoring network unit integrates a water quality monitoring buoy for collecting water temperature, dissolved oxygen, total phosphorus and chlorophyll a, a camera unit for obtaining water appearance and underwater images, and a weather station for collecting wind speed and illumination, and collecting water quality parameters, water images and weather data of a target lake; the multi-modal data fusion and preprocessing unit is responsible for carrying out space-time alignment and fusion on the collected multi-source heterogeneous data to generate a lake ecological state multi-modal data set; The multi-mode data set is analyzed by the multi-task deep learning prediction unit, and tasks including dynamic tracing pollution based on a graph neural network, generation of a water quality distribution map of a preset period based on a space-time prediction model and algal bloom occurrence probability and path early warning based on sequence learning are executed in parallel; The reinforced learning intelligent decision unit performs decision optimization through a deep Q network reinforced learning intelligent body based on prediction and early warning results, a state space of the reinforced learning intelligent body comprises predicted water quality and algal bloom risk information, an action space comprises a control instruction for at least one of aeration equipment, an ecological water gate pump and a carbon source adding device, and an optimal strategy of multi-facility cooperative regulation is output through interaction with environment simulation; The hierarchical execution and control unit converts the generated intelligent decision strategy into specific executable physical actions, and performs fine control on the execution unit according to a preset multilevel risk threshold.

Description

Urban shallow lake ecological protection restoration method and system integrating intelligent control Technical Field The invention relates to the technical field of water environment ecological management and intelligent water affairs, in particular to an integrated intelligent control ecological protection and restoration method and system for urban shallow lakes. Background The urban shallow lake is used as an important component of the urban water ecological system, has the characteristics of shallow water, poor fluidity, easiness in being influenced by urban pollution, and the like, and generally faces the problems of unstable water quality, high eutrophication risk, single habitat, degradation of the ecological system, loss of self-cleaning capacity and the like. Especially under the background of rapid urban development, a large amount of pollutants enter the lake through warehouse-in branches, so that key water quality indexes such as total phosphorus, chemical oxygen demand and the like frequently exceed standards, algae bloom is frequently generated, and the water body functions and urban landscapes are seriously affected. At present, the treatment of the lakes mostly adopts traditional engineering means such as local dredging, artificial aeration, shoreline hardening, single wetland construction and the like, and has the outstanding problems that (1) the existing method is often focused on a certain link or single problem, and lacks a systematic treatment thought of a full chain from 'tributary-river mouth-reservoir area', so that the cooperativity among various measures is poor, and the treatment effect is difficult to last. (2) In the repairing process, the multidimensional coupling relation between river bank, water body and sediment is often ignored, the ecological environment is recovered incompletely, the self-recovery capability of the ecological system is weak, and the external interference is difficult to deal with. (3) Management relies on manual periodic monitoring, cannot master dynamic change of water quality in real time, has insufficient early warning capability on sudden ecological risks such as algal bloom, and response measures are often delayed from pollution events. (4) The treatment facilities are in a fixed operation mode after being built, and cannot be dynamically optimized and regulated according to real-time hydrology and water quality conditions, so that the operation efficiency is low and the cost is high. Although ideas such as 'sponge city', 'intelligent river and lake', etc. are gradually popularized, a technical system covering the whole chain of pollution tracing, engineering integration, intelligent feedback and emergency regulation is not formed at present, and a customized comprehensive treatment scheme for urban shallow lakes is especially lacking. Therefore, a systematic solution capable of comprehensively planning a source-process-end and integrating ecological engineering and intelligent management and control technology is needed to realize sustainable ecological restoration and long-acting operation and maintenance of urban shallow lakes. Disclosure of Invention In order to solve the technical problems, the invention provides the urban shallow lake ecological protection and restoration method and system integrating intelligent management and control, which realize the intelligent management and control of the whole process from monitoring to execution, effectively improve the systematicness, the accuracy and the foresight of lake management and obviously enhance the water environment risk coping capacity and the ecological restoration effect. The invention provides an integrated intelligent control ecological protection and restoration method for a shallow city lake, which comprises the following steps: Constructing a three-dimensional monitoring network, synchronously acquiring water quality parameters, water body images and meteorological data of a target lake, and carrying out space-time alignment and fusion on the acquired multi-source heterogeneous data to generate a lake ecological state multi-mode data set; Analyzing the multi-mode data set by using a deep learning model, and executing tasks including dynamic tracing pollution based on a graph neural network, generating a water quality distribution map of a preset period based on a space-time prediction model, and generating algal bloom occurrence probability and path early warning based on sequence learning in parallel; based on the prediction and early warning results, decision optimization is carried out through the reinforcement learning intelligent agent, the state space of the reinforcement learning intelligent agent comprises prediction water quality and algal bloom risk information, the action space comprises control instructions for at least one of aeration equipment, an ecological water gate regulating pump and a carbon source adding device, and an optimal strategy of multi-facility coop