An adaptive water level regulation method, system and device

Abstract

This invention provides an adaptive water level regulation method, system, and device. The method includes initializing the regulation system and determining the initial position of the electric actuator; collecting current environmental parameters, including the flow rate of the effluent storage well, the water quality of the effluent storage well, and meteorological and ecological parameters at the current location; calculating the optimal water level control strategy based on the environmental parameters and a wetland water level intelligent regulation algorithm; and regulating the electric actuator according to the optimal water level control strategy. This invention completely eliminates reliance on manual operation and maintenance, achieving fully intelligent operation with autonomous perception, decision-making, and execution, significantly improving operational efficiency. The fully automated regulation process reduces the workload of maintenance personnel by more than 70%, avoiding the lag and subjectivity of traditional manual inspections. It achieves dynamic adaptation of water level regulation to the wetland microenvironment, realizing high-precision, all-condition, low-energy-consumption, and high-efficiency operation goals.

Description

Technical Field

[0001] This invention belongs to the field of intelligent control technology, specifically relating to an adaptive water level regulation method, system, and equipment. Background Technology

[0002] Constructed wetlands, as a low-cost and eco-friendly wastewater treatment and ecological restoration technology, have been widely used in fields such as rural wastewater treatment, municipal and industrial effluent deep treatment, and non-point source pollution control due to their high efficiency in removing pollutants such as COD, ammonia nitrogen, and total phosphorus. They have become a key technological unit in the water environment governance system.

[0003] Currently, China has established relatively mature technical specifications and research foundations in the early design phase of constructed wetlands. Regarding core functional components, there are clear design guidelines for optimal plant selection (such as the suitability screening of aquatic plants like reeds and cattails) and filler gradation (such as the combination optimization of gravel, zeolite, and activated carbon) for different climate zones and pollution types. In terms of operational parameter design, the determination of parameters such as hydraulic retention time of influent load and pollutant concentration thresholds is supported by comprehensive calculation methods and engineering case studies. However, the long-term stable operation of constructed wetlands highly depends on dynamically adaptable operation and management strategies, and related research lags significantly behind the development of design technologies, especially in the field of refined operation and control technology under complex operating conditions, where there are significant shortcomings.

[0004] In practice, constructed wetlands often face four typical and complex operating conditions, and existing operation and management methods are insufficient to effectively address these challenges: First, when the influent flow or water quality is too low, the continuous drop in water level within the wetland leads to water shortage and oxygen imbalance in the roots of aquatic plants. Simultaneously, the activity of microorganisms drops sharply due to insufficient nutrients, significantly reducing pollutant degradation efficiency. Second, when the influent flow or water quality is too high, the short-term shock load exceeds the wetland's purification capacity, and excessively high water levels can easily cause plant stems to collapse, damaging the pore structure of the packing material on which microorganisms attach. Third, during heavy rains, the influx of large amounts of rainwater causes a sudden rise in wetland water levels, not only triggering sewage overflow and secondary pollution but also eroding the wetland substrate and damaging the functional layer. Fourth, in the freezing conditions of winter in both northern and southern regions, the surface water of the wetland freezes, blocking the plant's aeration channels, and the low temperature causes microbial metabolism to stagnate, almost completely eliminating the wetland's purification function under traditional operating conditions. These operational difficulties directly result in the actual pollutant removal rate of constructed wetlands decreasing by 30% to 60% compared to the design value, severely restricting the full realization of their technological value.

[0005] In summary, existing technologies suffer from insufficient control flexibility, low adjustment precision, weak adaptability to multiple operating conditions, and lag in response, which prevents them from achieving precise and dynamic control of water levels in constructed wetlands, thus limiting their pollutant removal efficiency and ecological stability in complex operating scenarios. Summary of the Invention

[0006] To address the shortcomings of the existing technologies, this invention provides an adaptive water level regulation method, system, and device that can effectively and precisely control the inflow and outflow of water in the pool, the quality of the inflow and outflow water, rainfall, and severe weather, and build an AI self-management platform based on the Internet of Things, thereby effectively addressing the problem of malfunction under various complex working conditions.

[0007] In a first aspect, the present invention proposes an adaptive water level regulation method, comprising:

[0008] Initialize the adjustment system and determine the initial position of the electric actuator.

[0009] Collect current environmental parameters; the environmental parameters include the flow rate of the effluent storage well, the water quality of the effluent storage well, the meteorological conditions at the current location, and the ecological parameters at the current location.

[0010] Based on the environmental parameters, the optimal water level control strategy is calculated using a wetland water level intelligent regulation algorithm.

[0011] The electric actuator is controlled according to the optimal water level control strategy.

[0012] The wetland water level intelligent control algorithm includes a multi-source data sensing and preprocessing module, a prediction model module, and a multi-objective rolling optimization decision module.

[0013] The multi-source data sensing and preprocessing module is used to preprocess the collected environmental parameters.

[0014] The dynamic prediction model predicts water quality and quantity over a future time period based on current environmental parameters.

[0015] The multi-objective rolling optimization decision module predicts the optimal water level based on the prediction results of the dynamic prediction model and outputs the optimal water level control strategy.

[0016] Preferably, determining the initial position of the electric actuator includes:

[0017] Retrieve historical operation data and filter historical scenarios with an environmental feature similarity greater than 85% at the current initialization time;

[0018] For the selected similar historical scenarios, extract the initial position data of the electric linear actuator in the corresponding scenario;

[0019] Calculate the position mean and standard deviation of the initial position data. If the standard deviation is less than a preset value, the position mean is directly used as the initial position reference value; otherwise, after removing outliers, the position mean is recalculated and used as the initial position reference value.

[0020] Based on the environmental characteristics and historical scene data at the current initialization moment, flow weight coefficient, water quality weight coefficient, ecological weight coefficient and meteorological weight coefficient are divided. Based on the initial position benchmark value, flow weight coefficient, water quality weight coefficient, ecological weight coefficient and meteorological weight coefficient, the final initial position of the electric actuator is calculated.

[0021] Preferably, the multi-source data sensing and preprocessing module includes:

[0022] By capturing the spatial correlation and temporal dependence of environmental parameters through parallel spatiotemporal branching, TCN feature extraction and LSTM sequence modeling are performed on the current environmental parameters.

[0023] The correlation between the parameters of TCN features and LSTM sequences is calculated by mutual information entropy to construct joint features; at the same time, residual connection processing is performed on TCN features and LSTM sequences.

[0024] An attention weight network is introduced to dynamically adjust the weights of each feature layer based on historical scene data, thereby obtaining a standardized feature vector for the current time, which serves as the input to the dynamic prediction model.

[0025] Preferably, before performing TCN feature extraction and LSTM sequence modeling on the current environmental parameters, the method further includes:

[0026] Given the current environmental parameters, construct the distance matrix between the low-frequency parameter time series and the baseline parameter time series, and solve the alignment path with the minimum cumulative distance through dynamic programming.

[0027] Along the alignment path, interpolate the low-frequency parameters in the current environmental parameters into a time series consistent with the high-frequency parameters;

[0028] The standard deviation of each parameter is calculated for the aligned time series data, and a Gaussian noise sequence is generated. The Gaussian noise sequence is then superimposed on the original time series data to obtain the enhanced time series data.

[0029] The time-series data after noise enhancement is arranged in reverse chronological order to generate flipped subsequences, and the environmental scene labels of the flipped subsequences are retained to expand the dataset of current environmental parameters.

[0030] Preferably, the dynamic prediction model includes:

[0031] A residual TCN-bidirectional LSTM-attention fusion layer backbone network is constructed, and the network is trained using MSE completion loss and cross-entropy weight alignment loss as joint loss functions.

[0032] The system receives standardized feature vectors and reshapes them according to the region-time dimension to match the spatial distribution of wetlands. It uses depthwise separable convolution combined with TCN spatial feature weights to extract parameter transmission features of the inlet area, core purification area and outlet area. It uses causal convolution combined with LSTM temporal feature weights to capture short-term temporal trends and outputs spatiotemporal basic coding features.

[0033] Construct water quality target vectors and water quantity target vectors, weight them according to attention weights, and concatenate them with spatiotemporal basic coding features. Then, obtain target-related spatiotemporal coding features through convolution compression.

[0034] Spatial separation convolution and Sigmoid gating are used, and the TCN feature correlation of the preprocessing module is combined to filter key regions and output pure spatial features;

[0035] Temporal separable convolution and Tanh gating are used, combined with LSTM sequence trend capture in the preprocessing module to output pure temporal features;

[0036] Calculate the time-step differences for pure time features, generate a difference sequence, and capture the rate of change of parameters;

[0037] Calculate the mutual information entropy between the difference features and the prediction target, and generate self-attention weights;

[0038] The weighted difference features are concatenated with the pure temporal features and input into a two-layer bidirectional LSTM to output enhanced temporal features.

[0039] The pure spatial features are extended with a fully connected layer and concatenated with the enhanced temporal features to obtain spatiotemporal enhanced features;

[0040] The mutual information correlation between spatiotemporal augmentation features and water quality indicators is calculated, step-size attention weights are generated, and after weighted pooling, they are activated by three fully connected layers of LeakyReLU to output the predicted water quality value.

[0041] The spatiotemporal enhanced features are input to a two-layer bidirectional LSTM decoding layer for decoding. The decoded output is matched with the input feature dimension and then added. After processing by a fully connected layer, the predicted water volume value is output.

[0042] Preferably, the optimal water level prediction based on the prediction results of the dynamic prediction model, and the output of the optimal water level control strategy, includes:

[0043] A two-dimensional data table is constructed based on the water quality prediction, water quantity prediction and confidence level. Low confidence prediction points are removed according to the confidence level threshold. Data from multiple time granularities are merged according to the control response priority to obtain short-term prediction dataset and medium- and long-term prediction dataset.

[0044] For short-term prediction datasets, determine whether the water quality meets the standards at each time point based on the predicted water quality values.

[0045] For medium- and long-term forecast datasets, determine whether the water volume threshold has been reached based on the predicted water volume values.

[0046] Based on the predicted water quality, water volume, or fixed period, determine the required water level adjustment range. Combined with the energy consumption characteristic curve of the electric actuator, determine the extension and retraction of the electric actuator.

[0047] Secondly, based on the same inventive concept, this application also proposes an adaptive water level regulation system, the system comprising: an electric actuator used in conjunction with a water storage well, the electric actuator being connected to a control cabinet, the control cabinet being externally connected to a meteorological data acquisition system, an ecological parameter acquisition system, a water flow monitoring probe, and a water quality monitoring probe.

[0048] The control cabinet calculates the optimal water level control strategy based on the collected environmental parameters and the wetland water level intelligent regulation algorithm, and regulates the electric push rod according to the optimal water level control strategy.

[0049] The environmental parameters include the flow rate of the effluent storage well, the water quality of the effluent storage well, the meteorological conditions at the current location, and the ecological parameters at the current location.

[0050] The intelligent wetland water level control algorithm includes a multi-source data sensing and preprocessing module, a prediction model module, and a multi-objective rolling optimization decision module.

[0051] The multi-source data sensing and preprocessing module is used to preprocess the collected environmental parameters.

[0052] The dynamic prediction model predicts water quality and quantity over a future time period based on current environmental parameters.

[0053] The multi-objective rolling optimization decision module predicts the optimal water level based on the prediction results of the dynamic prediction model and outputs the optimal water level control strategy.

[0054] Preferably, the system includes: the water flow rate monitoring probe and the water quality monitoring probe are installed at the inlet of the outlet water storage well;

[0055] The electric actuator is connected to a retractable plastic hose, which is connected to the water outlet. The control cabinet determines the required water level adjustment range based on the predicted water quality, water volume, or fixed cycle. Combined with the energy consumption characteristic curve of the electric actuator, the extension and retraction of the electric actuator is determined. When the electric actuator controls the retractable plastic hose to be lower than the water surface, the water in the water storage well is discharged through the retractable plastic hose.

[0056] Preferably, the system includes: further comprising:

[0057] The control cabinet can also switch between different operating modes based on flow load and weather conditions.

[0058] Thirdly, based on the same inventive concept, this application also proposes a computer device, the computer device including a processor and a memory, the memory storing at least one instruction, the at least one instruction being loaded and executed by the processor to implement an adaptive water level adjustment method as described in the first aspect.

[0059] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0060] This invention completely eliminates reliance on manual operation and maintenance, achieving fully intelligent operation and maintenance processes with autonomous perception, decision-making, and execution, significantly improving operational efficiency. The fully automated control process reduces the workload of operation and maintenance personnel by more than 70%, avoiding the lag and subjectivity of traditional manual inspections.

[0061] This invention innovatively designs a residual TCN-bidirectional LSTM fusion prediction model, which completely solves the problems of spatiotemporal feature fragmentation, delayed response to abrupt changes, and lack of confidence in traditional prediction models. For four typical complex operating conditions—low load, high load, heavy rain, and freezing—it can achieve precise positioning, dynamic prediction and early warning, and adaptive decision-making and control, realizing dynamic adaptation of water level control to the wetland microenvironment, and achieving the operational goals of high precision, all operating conditions, low energy consumption, and high efficiency. Attached Figure Description

[0062] Figure 1 This is a diagram of an adaptive water level regulation system shown in an embodiment of the present invention.

[0063] Figure 2 This is a flowchart illustrating the adaptive water level regulation method according to an embodiment of the present invention.

[0064] The components are as follows: 1- Temporary water storage well; 2- Inlet pipe; 3- Outlet pipe; 4- Through-wall sleeve; 5- Electric push rod; 6- Telescopic plastic hose; 7- Ring hose fastener; 8- Doppler flow monitoring system (including probe); 9- Multifunctional water quality monitoring system (including probe); 10- Meteorological environment acquisition system; 11- Ecological parameter acquisition system; 12- Angle steel; 13- Integrated monitoring and electrical control cabinet; 14- Data transmission and reception device; 15- Solar panel; 16- Support frame; 17- Cover plate. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0066] Example 1: This invention proposes an adaptive water level regulation method, which is mainly applied to the water level management of temporary storage wells in constructed wetlands, such as... Figure 1 As shown, constructed wetlands mainly include: temporary effluent storage wells, inlet pipes, outlet pipes, wall sleeves, electric actuators, retractable plastic hoses, annular hose fasteners, Doppler flow monitoring systems (including probes), multi-functional water quality monitoring systems (including probes), meteorological and environmental data acquisition systems, ecological parameter acquisition systems, angle steel, integrated monitoring and electrical control cabinets, data transmission and reception devices, solar panels, and an IoT-based AI management and control platform.

[0067] It should be noted that:

[0068] (1) Electric linear actuator, with telescopic characteristics, made of stainless steel or aluminum alloy, with a stroke of 10-1500mm, powered by 24V DC, with a speed of 10-180mm / s, and an IP65 waterproof rating. The control system and power supply of the electric linear actuator are located in the integrated monitoring and electrical control cabinet.

[0069] (2) A retractable plastic hose with the function of being able to extend and retract, and its diameter is matched with the outlet pipe of the artificial wetland.

[0070] (3) Ring-shaped hose fastener, made of stainless steel, with a circular clamp type. One end is fastened to the artificial wetland outlet pipe by bolts, and the other end is connected to the electric push rod by screws.

[0071] (4) Doppler flow monitoring system with a range of 0-1000 m3 / d and signal output via RS485 or USB interface. The system and power supply are located in an integrated monitoring and electrical control cabinet. The probe is placed in the inlet regulating tank and the outlet temporary storage well to monitor the flow.

[0072] (5) Multifunctional water quality monitoring system, capable of monitoring COD Cr It displays functions such as ammonia nitrogen, total phosphorus, and total nitrogen concentration. COD CrThe monitoring range is 0-500 mg / L, ammonia nitrogen is 0-100 mg / L, total phosphorus is 0-50 mg / L, and total nitrogen is 0-100 mg / L. Signal output is via RS485 or USB interface. The system and power supply are located in an integrated monitoring and electrical control cabinet. Probes are placed in the inlet regulating tank and the outlet storage well to monitor water quality.

[0073] (6) Meteorological environment acquisition system: Integrating a rain gauge (0.2 mm resolution), wind speed and direction sensor, atmospheric pressure sensor and photosynthetically active radiation sensor to construct a micro weather station. This unit is specially equipped with a rainfall prediction function for the next 2 hours, which predicts the arrival of rainstorms by analyzing the sudden drop in air pressure and the rising trend of humidity.

[0074] (7) Ecological parameter acquisition system: including plant growth monitoring camera (equipped with multispectral lens), benthic animal activity sensor and bird call recognition microphone, to quantify wetland ecological status. Plant leaf area index and root development status are analyzed through image recognition algorithm.

[0075] (8) Angle steel, L40*3.5, use bolts to fix the electric push rod.

[0076] (9) Integrated monitoring and electrical control cabinet, which integrates a Doppler flow monitoring system, a multi-functional water quality monitoring system, a solar energy control system and a storage battery.

[0077] (10) Data transmission and reception device, located at the top of the integrated monitoring and electrical control cabinet, uses GPRS for transmission.

[0078] (11) Solar panels: a single solar panel is powered by 12-36V with a peak power of 100-500W. Multiple panels can be connected in parallel for power supply. The control system and battery are located in an integrated monitoring and electrical control cabinet.

[0079] (12) An IoT-based AI management and control platform, which features online monitoring, fault management, remote control, operation recording and management, AI self-diagnosis program, and AI self-analysis program. The platform can be used in three ways: PC, web, and mobile APP. The platform will be used to manage electric actuators, monitoring systems, and solar energy systems, and has two control modes: manual and automatic.

[0080] It should be noted that the connection relationship between the above parts is as follows:

[0081] The outlet pipe at the bottom of the constructed wetland is connected to the outlet storage well via a wall sleeve. The wall sleeve is a prefabricated product and waterproofing measures are implemented using standard methods.

[0082] The retractable plastic hose is connected at one end to the water outlet pipe at the bottom of the artificial wetland using a ring-shaped hose fastener, and the other end is connected to the electric push rod using bolts.

[0083] One end of the electric actuator is fixed to the angle steel with bolts, and the other end is connected to the screw of the annular hose fixing component.

[0084] The angle steel is connected to the outlet storage well using expansion bolts.

[0085] One end of the Doppler flow monitoring probe and the multi-functional water quality monitoring probe are placed in the outlet storage well, and the other end is connected to the control system located in the integrated monitoring and electrical control cabinet.

[0086] Both the meteorological and environmental data acquisition system and the ecological parameter acquisition system are powered by solar panels, with mains power as a backup. They can connect in real time to an IoT-based AI management and control platform via signal transmission.

[0087] One end of the solar panel connects to the control system, and the other end connects to the battery. Both the control system and the battery are located within an integrated monitoring and electrical control cabinet. The battery supplies power to all electrical equipment within the cabinet, including the control systems, electric actuators, and monitoring systems. To ensure 24-hour normal operation, mains power is used as a backup power source.

[0088] The IoT-based AI control platform collects data from Doppler flow monitoring probes, multi-functional water quality monitoring probes, meteorological and environmental data acquisition systems, and ecological parameter data acquisition systems. It then organizes, analyzes, and makes decisions based on the data, and issues commands to the electric actuator control system for regulation.

[0089] Before system initialization, it also includes:

[0090] Electric actuator adjustment: The shortest extension distance required by the electric actuator stroke design is the highest water level of the constructed wetland, and the longest distance is the lowest operating water level of the constructed wetland. Before starting operation of the constructed wetland, adjust the electric actuator extension rod to the designed outlet water level height.

[0091] Solar power system commissioning: Connect and test the solar panels, control system, and battery. After successful commissioning, connect the electric actuator, monitoring probe, and other electrical equipment to the battery to supply power. Simultaneously, connect a bypass to the mains power supply to ensure that mains power can be switched on if solar power is insufficient.

[0092] Commissioning of flow and water quality monitoring probes: During the commissioning of the constructed wetland, the two monitoring probes were commissioned and calibrated to ensure that the probes were sensitive and that data acquisition was normal.

[0093] Debugging the meteorological environment data acquisition system: Test whether the rain gauge can record the rainfall for each rainfall event correctly. Check whether the wind speed and direction sensor, atmospheric pressure sensor, and photosynthetically active radiation sensor are functioning properly.

[0094] Ecological parameter acquisition system: Testing the functionality of the plant growth monitoring camera (equipped with a multispectral lens), benthic animal activity sensor, and bird call recognition microphone. Testing its ability to analyze plant leaf area index and root development using image recognition algorithms.

[0095] Debugging of the IoT-based AI control platform: After all the aforementioned components have been debugged and are functioning correctly, the platform itself is debugged to test its ability to receive data from the monitoring probes, electric actuators, solar energy system, meteorological and environmental data acquisition system, and ecological parameter data acquisition system. Simultaneously, remote / manual switching operation is performed to ensure the system operates normally.

[0096] In this embodiment, as Figure 2 As shown, the adaptive water level regulation method specifically includes:

[0097] Step 1: Initialize the adjustment system and determine the initial position of the electric actuator.

[0098] Step 2: Collect current environmental parameters; the environmental parameters include the flow rate of the effluent storage well, the water quality of the effluent storage well, the meteorological conditions at the current location, and the ecological parameters at the current location.

[0099] Preferably, the sensing unit is activated to collect environmental parameters at a preset frequency:

[0100] High-frequency parameters: effluent storage well flow rate (3 minutes / time), effluent storage well water quality (5 minutes / time);

[0101] Low-frequency parameters: Current location weather (every 10 minutes / time), current location ecological parameters (every 8 minutes / time);

[0102] Preferably, the collected data is transmitted to the control unit in real time to generate a raw parameter time series dataset.

[0103] Step 3: Based on the environmental parameters, calculate the optimal water level control strategy using the wetland water level intelligent regulation algorithm.

[0104] Step 4: Adjust the electric push rod according to the optimal water level control strategy.

[0105] The wetland water level intelligent control algorithm includes a multi-source data sensing and preprocessing module, a prediction model module, and a multi-objective rolling optimization decision module.

[0106] The multi-source data sensing and preprocessing module is used to preprocess the collected environmental parameters.

[0107] The dynamic prediction model predicts water quality and quantity over a future time period based on current environmental parameters.

[0108] The multi-objective rolling optimization decision module predicts the optimal water level based on the prediction results of the dynamic prediction model and outputs the optimal water level control strategy.

[0109] In this embodiment, determining the initial position of the electric actuator includes:

[0110] Retrieve historical operation data and filter historical scenarios with an environmental feature similarity greater than 85% at the current initialization moment.

[0111] For the selected similar historical scenarios, extract the initial position data of the electric actuator in the corresponding scenario.

[0112] Calculate the position mean and standard deviation of the initial position data. If the standard deviation is less than a preset value, the position mean is directly used as the initial position reference value; otherwise, after removing outliers, the position mean is recalculated and used as the initial position reference value.

[0113] Based on the environmental characteristics and historical scene data at the current initialization moment, flow weight coefficient, water quality weight coefficient, ecological weight coefficient and meteorological weight coefficient are divided. Based on the initial position benchmark value, flow weight coefficient, water quality weight coefficient, ecological weight coefficient and meteorological weight coefficient, the final initial position of the electric actuator is calculated.

[0114] For example, operational data from the past three hydrological cycles (including the rainy season, dry season, and freezing period of 2021-2023) of the wetland were retrieved, totaling 10,800 samples. Using the environmental characteristics at the current initialization time (9:00 AM, May 10, 2024) as a baseline: effluent storage well flow rate 4.2 m³ / h, wastewater COD 38 mg / L, ammonia nitrogen 3.6 mg / L, TP 0.21 mg / L, rainfall 0 mm, wind speed 1.2 m / s, temperature 22℃, vegetation coverage 88%, soil moisture content 35%, and aquatic organism population density 0.8 individuals / m³, the similarity with historical samples was calculated.

[0115] Meteorological similarity: calculated using Euclidean distances of rainfall, wind speed, and temperature.

[0116] ;

[0117] Water quality similarity: Calculate the relative error mean of COD, ammonia nitrogen, and TP, and screen samples with a mean of <8%.

[0118] Ecological similarity: Samples with a similarity greater than 0.9 are selected by calculating the cosine similarity between vegetation coverage and soil moisture content.

[0119] Ultimately, 210 historical scene data points with an environmental feature similarity greater than 85% were obtained.

[0120] Initial position reference value determined:

[0121] Initial position data of electric linear actuators from 210 similar scenarios were extracted, ranging from 720-750mm. The calculated position mean was 735mm, with a standard deviation of 8.2mm. The preset standard deviation threshold was 10mm. Since 8.2mm < 10mm, 735mm was directly used as the initial position reference value.

[0122] Based on the current environmental characteristics (stable flow, water quality close to design value, clear weather, and good ecological condition), the following weighting coefficients are assigned: flow weight 0.3, water quality weight 0.4, ecological weight 0.1, and meteorological weight 0.2.

[0123] Retrieve the historical optimal initial positions for each weighted dimension: optimal flow rate position 730mm, optimal water quality position 740mm, optimal ecological position 738mm, and optimal meteorological position 735mm.

[0124] ;

[0125] Final initial position:

[0126]

[0127] Control the electric actuator to move to the 735.8mm position.

[0128] Preferably, the multi-source data sensing and preprocessing module includes:

[0129] By capturing the spatial correlation and temporal dependence of environmental parameters through parallel spatiotemporal branching, TCN feature extraction and LSTM sequence modeling are performed on the current environmental parameters.

[0130] The correlation between the parameters of TCN features and LSTM sequences is calculated by mutual information entropy to construct joint features; at the same time, residual connection processing is performed on TCN features and LSTM sequences.

[0131] An attention weight network is introduced to dynamically adjust the weights of each feature layer based on historical scene data, thereby obtaining a standardized feature vector for the current time, which serves as the input to the dynamic prediction model.

[0132] The preferred TCN spatial feature extraction branch specifically includes:

[0133] Network structure: 2-layer residual TCN, each layer contains a combination of causal convolution + BatchNorm + ReLU, the convolution kernel size is 3×1 (3 regions in the spatial dimension and 1 stride in the temporal dimension), and the number of output channels is 32 and 64 respectively;

[0134] Input adaptation: The aligned data is reshaped into a tensor of [batch_size, 3, 20, 10] according to the region dimension (3), time series dimension (20), and parameter dimension (10: flow, COD, ammonia nitrogen, TP, rainfall, wind speed, temperature, vegetation cover, soil moisture content, and aquatic organism population density).

[0135] Spatial correlation capture: The first layer TCN extracts multi-parameter correlations within a single region (such as the concentration correlation between COD and TP in the core purification zone), and the second layer TCN extracts cross-regional parameter transmission features (such as the hysteretic correlation between influent flow rate and effluent TP). Finally, a TCN spatial feature map of [batch_size,3,20,64] is output.

[0136] Preferred LSTM timing modeling branch:

[0137] Network structure: 2-layer bidirectional LSTM, 128 neurons in each hidden layer, Dropout probability 0.2 (to suppress overfitting), using Adam optimizer (initial learning rate 0.001);

[0138] Input adaptation: Reshape the aligned data into a temporal tensor of [batch_size, 20, 10] according to the temporal dimension (20) - parameter dimension (10);

[0139] Time dependency capture: Forward LSTM captures future time series trends (such as the flow rate increase trend from 9:00 to 9:30), and backward LSTM captures historical time series dependencies (such as the correlation between flow rate at 9:30 and water quality at 9:00). The time series dimension is compressed by global average pooling, and the output is an LSTM time series feature vector of [batch_size, 256] (256 = 2 × 128, bidirectional output concatenation).

[0140] The correlation between parameters of TCN spatial features and LSTM temporal features is calculated using a joint probability distribution. The specific steps are as follows:

[0141] The TCN spatial feature map is compressed to [batch_size,20,64] using a 1×1 convolution, and aligned with the dimensions of the LSTM temporal feature vector (expanded to [batch_size,20,64] via a fully connected layer);

[0142] For each pair of feature dimensions (64 pairs in total), the joint probability distribution P(X,Y) and marginal probability distributions P(X) and P(Y) are calculated using the kernel density estimation method (X is a TCN feature and Y is an LSTM feature).

[0143] Mutual information entropy is calculated using the formula:

[0144] ;

[0145] The correlation was calculated, and strong correlation feature pairs with I(X,Y) > 0.5 were selected, resulting in 28 pairs. For example, the correlation between the COD feature of the core purification area of ​​TCN and the flow time-series trend feature of LSTM was 0.68.

[0146] First, the TCN features and LSTM features are concatenated along the channel dimension, and then compressed to [batch_size,20,64] using a 3×3 convolution kernel to obtain the joint feature map.

[0147] Example: The spatial features of COD from TCN and the temporal features of traffic from LSTM are concatenated and then processed by convolution to output a joint feature of 0.35. The final joint feature integrates spatial correlation and temporal dependence.

[0148] After each TCN convolution, the output feature [batch_size,3,20,64] is adjusted to match the dimension of the input feature [batch_size,3,20,10] by a 1×1 convolution. The residuals are then stacked according to the formula F(x)+x, where F(x) is the convolution output and x is the input, which helps to alleviate the gradient vanishing problem in deep networks.

[0149] The LSTM output features [batch_size,256] are adjusted to [batch_size,20,10] through a fully connected layer and superimposed on the input temporal tensor to preserve the original temporal information.

[0150] The constructed joint feature map [batch_size,20,64] is superimposed with the concatenated features of TCN spatial features + LSTM temporal features to enhance the feature representation capability.

[0151] The wetland's feature layer weights and regulation effect correlation data from the past three hydrological cycles were used, totaling 8000 samples. Each sample contains historical weights from TCN / LSTM / joint features corresponding to the water level regulation error label of the scene. A two-layer fully connected attention network was used, consisting of a 64-dimensional input layer, a 32-dimensional hidden layer, and a 3-dimensional output layer. The output is the weight coefficients of the TCN spatial feature layer w1, the LSTM temporal feature layer w2, and the joint feature layer w3, i.e., w1 + w2 + w3 = 1. Each dimension of the fused features was standardized, resulting in a feature vector with dimensions [batch_size, 20, 64], perfectly matching the input layer dimensions of the dynamic prediction model, allowing it to be directly input into the model for spatiotemporal encoding.

[0152] Preferably, before performing TCN feature extraction and LSTM sequence modeling on the current environmental parameters, the method further includes:

[0153] Given the current environmental parameters, construct the distance matrix between the low-frequency parameter time series and the baseline parameter time series, and solve the alignment path with the minimum cumulative distance through dynamic programming.

[0154] Along the alignment path, interpolate the low-frequency parameters in the current environmental parameters into a time series consistent with the high-frequency parameters;

[0155] The standard deviation of each parameter is calculated for the aligned time series data, and a Gaussian noise sequence is generated. The Gaussian noise sequence is then superimposed on the original time series data to obtain the enhanced time series data.

[0156] The time-series data after noise enhancement is arranged in reverse chronological order to generate flipped subsequences, and the environmental scene labels of the flipped subsequences are retained to expand the dataset of current environmental parameters.

[0157] Preferably, using the flow rate of the temporary storage well at 3 minutes / cycle as the baseline parameter, a distance matrix is ​​constructed for the ecological parameter at 8 minutes / cycle: Let the ecological parameter sequence X=(88,88.2) (9:00, 9:08), and the baseline parameter sequence Y=(4.2,4.3,4.1) (9:00, 9:03, 9:10). Calculate the Euclidean distance matrix D, and solve the alignment path P={(1,1),(1,2),(2,3)} with the minimum cumulative distance through dynamic programming. Use cubic spline interpolation along the path to generate the ecological parameter 88.1% at time 9:03, so that the ecological parameter and the baseline parameter maintain a uniform frequency of 3 minutes / cycle.

[0158] The standard deviation of the aligned flow parameters was calculated as σ = 0.1 m³ / h. A Gaussian noise sequence N = (0.008, −0.005, 0.009) (intensity ≤ 10%σ) was generated and superimposed to obtain the enhanced flow data (4.208, 4.295, 4.109). The flow subsequence from 9:00 to 9:30 was selected, and a reversed subsequence was generated in reverse chronological order and labeled with a stable inflow scenario. The dataset size was expanded to 1.8 times the original.

[0159] Preferably, the dynamic prediction model includes:

[0160] A residual TCN-bidirectional LSTM-attention fusion layer backbone network is constructed, and the network is trained using MSE completion loss and cross-entropy weight alignment loss as joint loss functions.

[0161] The system receives standardized feature vectors and reshapes them according to the region-time dimension to match the spatial distribution of wetlands. It uses depthwise separable convolution combined with TCN spatial feature weights to extract parameter transmission features of the inlet area, core purification area and outlet area. It uses causal convolution combined with LSTM temporal feature weights to capture short-term temporal trends and outputs spatiotemporal basic coding features.

[0162] Construct water quality target vectors and water quantity target vectors, weight them according to attention weights, and concatenate them with spatiotemporal basic coding features. Then, obtain target-related spatiotemporal coding features through convolution compression.

[0163] Spatial separation convolution and Sigmoid gating are used, and the TCN feature correlation of the preprocessing module is combined to filter key regions and output pure spatial features;

[0164] Temporal separable convolution and Tanh gating are used, combined with LSTM sequence trend capture in the preprocessing module to output pure temporal features;

[0165] Calculate the time-step differences for pure time features, generate a difference sequence, and capture the rate of change of parameters;

[0166] Calculate the mutual information entropy between the difference features and the prediction target, and generate self-attention weights;

[0167] The weighted difference features are concatenated with the pure temporal features and input into a two-layer bidirectional LSTM to output enhanced temporal features.

[0168] The pure spatial features are extended with a fully connected layer and concatenated with the enhanced temporal features to obtain spatiotemporal enhanced features;

[0169] The mutual information correlation between spatiotemporal augmentation features and water quality indicators is calculated, step-size attention weights are generated, and after weighted pooling, they are activated by three fully connected layers of LeakyReLU to output the predicted water quality value.

[0170] The spatiotemporal enhanced features are input to a two-layer bidirectional LSTM decoding layer for decoding. The decoded output is matched with the input feature dimension and then added. After processing by a fully connected layer, the predicted water volume value is output.

[0171] Preferably, the optimal water level prediction based on the prediction results of the dynamic prediction model, and the output of the optimal water level control strategy, includes:

[0172] A two-dimensional data table is constructed based on the water quality prediction, water quantity prediction and confidence level. Low confidence prediction points are removed according to the confidence level threshold. Data from multiple time granularities are merged according to the control response priority to obtain short-term prediction dataset and medium- and long-term prediction dataset.

[0173] For short-term prediction datasets, determine whether the water quality meets the standards at each time point based on the predicted water quality values.

[0174] For medium- and long-term forecast datasets, determine whether the water volume threshold has been reached based on the predicted water volume values.

[0175] Based on the predicted water quality, water volume, or fixed period, determine the required water level adjustment range. Combined with the energy consumption characteristic curve of the electric actuator, determine the extension and retraction of the electric actuator.

[0176] In this embodiment, based on the above-described adaptive water level regulation method, the conventional operating mode is as follows:

[0177] (1) When the constructed wetland is running at the design flow and design load, start the solar power supply system, Doppler flow and multi-functional water quality monitoring probe, meteorological environment acquisition system and ecological parameter acquisition system to start normal monitoring of the system.

[0178] (2) The electric actuator needs to be pre-adjusted to the designed outlet water level. After that, the sewage is purified by the artificial wetland and then enters the outlet storage well.

[0179] (3) At this time, if the multi-functional water quality monitoring probe detects that the effluent water quality does not meet the standards, the platform will send an instruction to the electric push rod to adjust the height of the effluent pipe to the highest water level, so as to give full play to the absorption of pollutants by plants, the adsorption of pollutants by fillers in wetlands and the degradation of pollutants by microorganisms, and achieve the discharge of sewage in compliance with standards.

[0180] (4) If it still does not meet the standard, repeat the above operation until it reaches the highest water level.

[0181] (5) When the wetland is operating normally for 7-30 days, it can be lowered to the lowest water level in one go using an electric push rod. The water level drops in a short time, flushing the filler material in the wetland, reducing the blockage of the filler material in the wetland, and improving the overall removal effect of the wetland on pollutants.

[0182] In this embodiment, based on the above adaptive water level regulation method, the overload operation mode is as follows:

[0183] (1) When the inflow rate of the constructed wetland exceeds the design parameters by 1.0-1.2 times or the inflow load exceeds the initial design parameters by 1.0-1.5 times, an emergency response should be initiated immediately to slow down the inflow rate. At the same time, the movement distance of the electric push rod can be set in different time periods to allow it to operate between the lowest and highest water levels.

[0184] (2) When the water level is higher than the filler material, the plants, filler material, and microorganisms in the wetland are in an aerobic stage; when the water level is lower than the filler material, they are in an anoxic stage. By continuously alternating between aerobic and anoxic conditions, the wetland's ability to remove pollutants is improved.

[0185] Alternatively, an electric actuator can be used to raise the height of the retractable plastic hose, extending the retention time of wastewater in the constructed wetland, thereby achieving compliant wastewater discharge. When the effluent quality does not meet the standards, the treatment method is the same as described above (1).

[0186] In this embodiment, based on the above-described adaptive water level regulation method, the low-load operation mode is as follows:

[0187] When the influent flow rate of the constructed wetland is less than 0.2-0.8 times the design parameters or the influent load is less than 0.2-1.0 times the initial design parameters, increase the influent flow rate. Alternatively, while maintaining the aforementioned flow rate, lower the electric actuator from the design water level until the effluent just meets the standard.

[0188] In this embodiment, based on the above adaptive water level regulation method, the rainstorm operation mode is as follows:

[0189] (1) When the meteorological environment acquisition system detects in advance that heavy rain is about to fall, it issues an instruction through the platform in advance before the rainstorm arrives. Under the premise of ensuring that the water quality meets the standards, the electric push rod is extended from the designed water level to a certain height. The volume corresponding to this height is the storage capacity during the rainstorm.

[0190] (2) During heavy rain, the electric push rod will retract to a certain height to cope with the impact flow and load generated when the rainstorm arrives. At the same time, it will ensure that the effluent water quality meets the standards.

[0191] (3) After the rain, the water level is gradually restored to the design level. At this time, the sediment flushing program is started, and the water level is lowered to the lowest level in one go using an electric push rod. The water level drops in a short time to flush the filler material in the wetland, reducing the blockage problem caused by the sudden increase in flow during the rainy season.

[0192] In this embodiment, based on the above adaptive water level adjustment method, the freezing operation mode is as follows:

[0193] When the meteorological and environmental data acquisition system detects a sudden change in weather and the onset of freezing conditions, it moves the electric actuator from the designed outlet water level to the highest water level, ensuring that the effluent water quality meets the standards. Once the weather returns to normal, the water level is gradually restored to the designed level.

[0194] In this embodiment, based on the above adaptive water level adjustment method, the inspection mode is as follows:

[0195] Every 30-60 days, the wetland system will be routinely inspected. The inspection mode mainly monitors the growth of plants and the normal water level in the wetland.

[0196] When cameras in the ecological parameter acquisition system detect slow or no plant growth, the platform can be used to review influent and effluent flow rates and water quality. If the effluent quality meets standards, but the influent flow rate or concentration is far below the initial design value, the influent flow rate or effluent water level can be increased to provide more nutrients for the plants and promote their normal growth. Conversely, if the plants are detected to be growing too fast, the influent flow rate or effluent water level can be reduced, while ensuring that the effluent quality meets standards.

[0197] When the cameras in the ecological parameter acquisition system detect that plants are withering, the platform can issue a work order to the maintenance personnel to replant the plants.

[0198] When the cameras in the ecological parameter acquisition system detect pests and diseases in plants, the platform can issue work orders to maintenance personnel to spray pesticides to treat the plants.

[0199] When cameras in the ecological parameter acquisition system detect that weeds are growing too tall in the wetland, even covering the height of other plants, the platform can issue a work order to the maintenance personnel to clear the weeds.

[0200] When the cameras in the ecological parameter acquisition system detect that the plants have reached the designed height, the platform can issue a work order to the maintenance personnel to harvest the plants.

[0201] When cameras in the ecological parameter acquisition system detect that weeds are growing too tall in the wetland, even covering the height of other plants, the platform can issue a work order to the maintenance personnel to clear the weeds.

[0202] Inspection of normal water levels within the wetland:

[0203] The platform automatically reviews the recorded effluent water level data during wetland operation period. If it finds that the effluent water level remains above the design water level or the maximum water level for an extended period while the wetland is operating under normal conditions, or if the camera detects water seepage on the surface of the wetland system, it indicates that the wetland's packing material is significantly clogged. In this case, the following measures can be taken:

[0204] (1) The platform issues an instruction to lower the water level to the minimum level while ensuring that the water quality meets the standards, and uses the water level difference to flush the filler material in the wetland to alleviate the blockage problem.

[0205] (2) When repeated flushing fails to lower the normal operating water level in the pool, the platform will issue a work order to the operation and management personnel to inspect the wetland site to determine whether the packing material needs to be replaced.

[0206] In Example 2, this application also proposes an adaptive water level regulation system, the system comprising: an electric actuator used in conjunction with a temporary water storage well, the electric actuator being connected to a control cabinet, the control cabinet being externally connected to a meteorological data acquisition system, an ecological parameter acquisition system, a water flow monitoring probe, and a water quality monitoring probe.

[0207] The control cabinet calculates the optimal water level control strategy based on the collected environmental parameters and the wetland water level intelligent regulation algorithm, and regulates the electric push rod according to the optimal water level control strategy.

[0208] The environmental parameters include the flow rate of the effluent storage well, the water quality of the effluent storage well, the meteorological conditions at the current location, and the ecological parameters at the current location.

[0209] The intelligent wetland water level control algorithm includes a multi-source data sensing and preprocessing module, a prediction model module, and a multi-objective rolling optimization decision module.

[0210] The multi-source data sensing and preprocessing module is used to preprocess the collected environmental parameters.

[0211] The dynamic prediction model predicts water quality and quantity over a future time period based on current environmental parameters.

[0212] The multi-objective rolling optimization decision module predicts the optimal water level based on the prediction results of the dynamic prediction model and outputs the optimal water level control strategy.

[0213] Preferably, the system includes: the water flow monitoring probe and the water quality monitoring probe are installed at the inlet of the water storage well.

[0214] The electric actuator is connected to a retractable plastic hose, which is connected to the water outlet. The control cabinet determines the required water level adjustment range based on the predicted water quality, water volume, or fixed cycle. Combined with the energy consumption characteristic curve of the electric actuator, the extension and retraction of the electric actuator is determined. When the electric actuator controls the retractable plastic hose to be lower than the water surface, the water in the water storage well is discharged through the retractable plastic hose.

[0215] Preferably, the system includes: further comprising:

[0216] The control cabinet can also switch between different operating modes based on flow load and weather conditions.

[0217] In embodiment three, this application also proposes a computer device, the computer device including a processor and a memory, the memory storing at least one instruction, the at least one instruction being loaded and executed by the processor to implement an adaptive water level adjustment method as described in embodiment one.

[0218] Although exemplary embodiments have been described herein with reference to the accompanying drawings, it should be understood that the above exemplary embodiments are merely illustrative and are not intended to limit the scope of the invention. Various changes and modifications can be made therein by those skilled in the art without departing from the scope and spirit of the invention. All such changes and modifications are intended to be included within the scope of the invention as claimed in the appended claims.

[0219] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the statement "comprising a…" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0220] Although the description of the invention has been given in conjunction with the specific embodiments described above, it will be apparent to those skilled in the art that many substitutions, modifications, and variations can be made based on the foregoing. Therefore, all such substitutions, modifications, and variations are included within the spirit and scope of the appended claims.

Claims

1. A method of adaptive water level regulation, characterized by, The method comprises the following steps: initializing the adjusting system to determine the initial position of the electric push rod; collecting current environmental parameters; the environmental parameters include the flow of the temporary water storage well, the water quality of the temporary water storage well, the current location meteorology, and the current location ecological parameters; calculating the optimal water level control strategy based on the wetland water level intelligent control algorithm according to the environmental parameters; controlling the electric push rod according to the optimal water level control strategy; wherein the wetland water level intelligent control algorithm comprises a multi-source data sensing and preprocessing module, a prediction model module, and a multi-objective rolling optimization decision module; the multi-source data sensing and preprocessing module is used for preprocessing the collected environmental parameters; the dynamic prediction model is used for predicting the water quality and quantity in the future time period based on the current environmental parameters; the multi-objective rolling optimization decision module is used for predicting the optimal water level based on the prediction result of the dynamic prediction model and outputting the optimal water level control strategy; the multi-source data sensing and preprocessing module comprises: extracting TCN features and LSTM sequence modeling of the current environmental parameters by capturing the spatial correlation and time dependence of the environmental parameters through parallel space-time branches; calculating the correlation between the parameters of the TCN features and the LSTM sequence through mutual information entropy, constructing joint features, and simultaneously performing residual connection processing on the TCN features and the LSTM sequence; introducing an attention weight network, dynamically adjusting the weight of each feature layer according to historical scene data, obtaining a standardized feature vector at the current time as the input of the dynamic prediction model.

2. The self-adapting water level regulating method according to claim 1, wherein, The method for determining the initial position of the electric push rod comprises the following steps: retrieve historical operation data and filter the historical scenes with an environmental feature similarity greater than 85% at the current initialization time; extract the initial position data of the electric push rod in the corresponding scene from the filtered similar historical scenes; calculate the mean and standard deviation of the initial position data, and if the standard deviation is less than a preset value, directly use the mean as the initial position reference value; otherwise, remove the outliers and recalculate the mean as the initial position reference value; divide the flow weight coefficient, the water quality weight coefficient, the ecological weight coefficient, and the meteorological weight coefficient according to the environmental features at the current initialization time and the historical scene data, and calculate the final initial position of the electric push rod based on the initial position reference value, the flow weight coefficient, the water quality weight coefficient, the ecological weight coefficient, and the meteorological weight coefficient.

3. The method of claim 2, wherein, Before performing TCN feature extraction and LSTM sequence modeling on the current environmental parameters, the method further comprises the following steps: construct a distance matrix of the low-frequency parameter time series and the reference parameter time series in the current environmental parameters, and solve the alignment path with the minimum cumulative distance through dynamic programming; interpolate the low-frequency parameters in the current environmental parameters into time series consistent with the high-frequency parameters along the alignment path; calculate the standard deviation of each type of parameter for the aligned time series data, generate a Gaussian noise sequence, and superimpose the Gaussian noise sequence and the original time series data to obtain enhanced time series data; arrange the noise-enhanced time series data in reverse order of time to generate a flip sub-sequence, retain the environmental scene label of the flip sub-sequence, and expand the data set of the current environmental parameters.

4. The method of claim 3, wherein, The dynamic prediction model comprises: The residual TCN-bidirectional LSTM-attention fusion layer backbone network is constructed, and the network is trained by taking the MSE completion loss and the cross-entropy weight alignment loss as the joint loss function. The standardized feature vector is received, reshaped according to the area-time sequence dimension to match the spatial distribution of the wetland, and the deep separable convolution is combined with the TCN spatial feature weight to extract the parameter conduction characteristics of the inlet area, the core purification area and the outlet area; the causal convolution is combined with the LSTM time sequence feature weight to capture the short-term time sequence trend, and the spatio-temporal basic encoding feature is output; The water quality target vector and the water quantity target vector are constructed, the water quality target vector and the water quantity target vector are weighted according to the attention weight, and are spliced with the spatio-temporal basic encoding feature, and the target related spatio-temporal encoding feature is obtained through convolution compression; The spatial separable convolution and the Sigmoid gate are used to combine the TCN feature correlation degree of the preprocessing module to filter the key area and output the pure space feature; The time separable convolution and the Tanh gate are used to combine the LSTM sequence trend of the preprocessing module to capture the time sequence mutation and output the pure time feature; The difference sequence is generated by calculating the difference between the pure time features, and the parameter change rate is captured; The mutual information entropy of the difference feature and the prediction target is calculated to generate the self-attention weight; The weighted difference feature and the pure time feature are spliced and input into the 2-layer bidirectional LSTM to output the enhanced time feature; The pure space feature is expanded through the full connection layer and spliced with the enhanced time feature to obtain the spatio-temporal enhanced feature; The mutual information correlation degree between the spatio-temporal enhanced feature and the water quality index is calculated to generate the step attention weight, which is weighted and pooled through the 3-layer full connection layer LeakyReLU activation to output the water quality prediction value; The spatio-temporal enhanced feature is input into the 2-layer bidirectional LSTM decoding layer for decoding, and the decoding output is added after matching the input feature dimension, and the water quantity prediction value is output after processing through the full connection layer.

5. The method of claim 4, wherein, The prediction result based on the dynamic prediction model is used for optimal water level prediction, and an optimal water level control strategy is output, including: According to the water quality prediction value, the water quantity prediction value and the confidence, a two-dimensional data table is constructed, low confidence prediction points are removed according to the confidence threshold, multi-time granularity data is combined according to the control response priority, and short-term prediction data set and medium-long term prediction data set are obtained; For the short-term prediction data set, whether the water quality at each time point meets the standard is judged according to the water quality prediction value; For the medium-long term prediction data set, whether the water quantity threshold is reached is judged according to the water quantity prediction value; According to the water quality prediction value, the water quantity prediction value or the fixed period, the required water level adjustment range is determined, and the extension amount of the electric push rod is determined according to the electric push rod energy consumption characteristic curve.

6. A regulating system employing an adaptive water level regulating method according to any one of claims 1 to 5, characterized in that, The system comprises an electric push rod matched with the outlet temporary storage well, an electric push rod connected control cabinet, an external meteorological link acquisition system, an ecological parameter acquisition system, a water flow monitoring probe and a water quality monitoring probe; The control cabinet calculates the optimal water level control strategy based on the wetland water level intelligent control algorithm according to the collected environmental parameters, and controls the electric push rod according to the optimal water level control strategy. The environmental parameters include water flow of the temporary water storage well, water quality of the temporary water storage well, current location meteorology, and current location ecological parameters. The wetland water level intelligent regulation algorithm comprises a multi-source data sensing and preprocessing module, a prediction model module, and a multi-objective rolling optimization decision module. The multi-source data sensing and preprocessing module is configured to preprocess the collected environmental parameters. The dynamic prediction model is configured to predict water quality and water quantity in a future time period based on current environmental parameters. The multi-objective rolling optimization decision module is configured to perform optimal water level prediction based on the prediction result of the dynamic prediction model, and output an optimal water level control strategy.

7. The regulation system of claim 6, wherein The water flow monitoring probe and the water quality monitoring probe are arranged at a water inlet pipe opening of the temporary water storage well. The electric push rod is connected to a telescopic plastic hose, the telescopic plastic hose is connected to a water outlet, a control cabinet determines a current required water level regulation range according to a water quality prediction value or a water quantity prediction value or a fixed period, determines a telescopic amount of the electric push rod in combination with an energy consumption characteristic curve of the electric push rod, and when the electric push rod controls the telescopic plastic hose to be below the water surface, water in the temporary water storage well is discharged through the telescopic plastic hose.

8. The conditioning system of claim 7, wherein, Further comprising: The control cabinet further switches different operation modes according to flow load and meteorological conditions.

9. A computer device, comprising: The computer device comprises a processor and a memory, the memory stores at least one instruction, the at least one instruction is loaded and executed by the processor to implement the adaptive water level regulation method of any one of claims 1 to 5.

Images