A sluice gate control method and device for a water conservancy system

Abstract

The application provides a sluice gate control method and device for a water conservancy system, and relates to the technical field of automatic control of water conservancy projects.The method comprises the following steps: obtaining multi-source data through a comprehensive acquisition unit, inputting the multi-source data into a feature extraction layer to perform feature extraction and feature splicing, obtaining an initial feature vector, performing feature weighting on the initial feature vector through an attention feature weighting layer to obtain a weighted feature vector, performing dynamic feature mapping on the weighted feature vector through a dynamic ReLU activation function embedded in a first hidden layer to obtain a hydraulic feature vector, performing feature expansion on the hydraulic feature vector through an adaptive LeakyReLU activation function embedded in a second hidden layer and a residual connection layer to obtain a hydraulic equipment expanded feature vector, and calculating a target opening degree of the sluice gate according to the hydraulic equipment expanded feature vector through an adaptive Sigmoid activation function embedded in an opening degree calculation layer.A sluice gate execution unit generates an opening degree control instruction according to the target opening degree of the sluice gate, and adjusts the opening degree of the sluice gate according to the opening degree control instruction.

Description

Technical Field

[0001] This application relates to the field of automated control technology for water conservancy projects, and in particular to a method and device for controlling sluice gates in a water conservancy system. Background Technology

[0002] In the field of automated control of water conservancy projects, sluice gates refer to sluice gates or gates used to control water flow. They are commonly found in river and canal projects. Adjusting the opening degree of sluice gates is a core link in realizing the rational allocation of water resources and ensuring stable irrigation water supply.

[0003] However, existing sluice gate control methods typically employ fixed activation functions for feature mapping, failing to adjust activation characteristics based on dynamic conditions such as river flow velocity and motor load. Furthermore, the deep integration of equipment parameters and scheduling decisions is neglected during complex feature modeling, resulting in sluggish opening calculations that do not consider the physical state of the equipment. Additionally, the opening calculations do not adequately account for differences in power supply modes, leading to a simplistic adjustment strategy. These deficiencies result in insufficient decision-making accuracy, operational condition adaptability, and long-term stability in sluice gate control. Insufficient decision-making accuracy can easily lead to water supply imbalances in irrigation zones; insufficient operational condition adaptability can cause sluice gates to be unable to respond to dynamic changes in river flow velocity, solar power, and other dynamic conditions; and insufficient long-term stability can lead to a gradual decline in sluice gate control precision, thus restricting the water resource utilization efficiency and operational reliability of the water conservancy system. Summary of the Invention

[0004] To address the aforementioned problems, in a first aspect, the present invention provides a method for controlling the gates of a water conservancy system, comprising: An integrated sluice gate control system is constructed, which includes a comprehensive acquisition unit, a neural network scheduling and control unit, and a sluice gate execution unit. The comprehensive acquisition unit includes a water level sensor, a flow velocity sensor, an absolute encoder, a photovoltaic power sensor, and a camera. The neural network scheduling and control unit includes a feature extraction layer, an attention feature weighting layer, a first hidden layer, a second hidden layer, and an opening degree calculation layer. Multi-source data is acquired through a comprehensive acquisition unit, and then input into a feature extraction layer for feature extraction and feature concatenation to obtain an initial feature vector. The initial feature vector is then weighted by an attention feature weighting layer to obtain a weighted feature vector. By using the dynamic ReLU activation function embedded in the first hidden layer, dynamic feature mapping is performed on the weighted feature vector to obtain the hydraulic feature vector; By embedding an adaptive LeakyReLU activation function and a residual connection layer in the second hidden layer, the hydraulic feature vector is extended to obtain the extended feature vector of the hydraulic equipment. The target opening of the sluice gate is obtained by using an adaptive sigmoid activation function embedded in the opening calculation layer, based on the extended feature vector of the hydraulic equipment. The gate execution unit generates an opening control command based on the target opening degree of the gate, and adjusts the opening degree of the gate according to the opening control command.

[0005] Optionally, the integrated acquisition unit includes a local monitoring unit, a digital twin unit, and an irrigation zoning unit. The acquisition of multi-source data through the integrated acquisition unit includes: Monitoring data of the water conservancy system are collected through on-site monitoring units; The monitoring data is input into the digital twin unit to construct a three-dimensional dynamic model and simulate water flow, thereby obtaining hydraulic characteristic data; The monitoring data is input into the irrigation zoning unit to calculate the irrigation zoning and water demand, and obtain water demand data. The data consists of monitoring data, hydraulic characteristic data, and water demand data. The multi-source data includes total river inflow data, current water level data for each zone, flow velocity data for each zone, current sluice gate opening data, solar power data, water demand gap data for each zone, irrigation zone priority data, average flow velocity data of the river cross section, equipment parameters, motor current data, and the measured opening of the sluice gate at the previous moment.

[0006] Optionally, the step of inputting multi-source data into the feature extraction layer for feature extraction and feature concatenation to obtain an initial feature vector, and then weighting the initial feature vector through an attention feature weighting layer to obtain a weighted feature vector, includes: The total inflow data of the river channel, the current water level data of each zone, the flow velocity data of each zone, the current opening degree of the sluice gates, the solar power data, and the water demand gap data of each zone are input into the feature extraction layer for feature extraction and feature concatenation to obtain the initial feature vector. The attention feature weighting layer weights each feature in the initial feature vector based on the irrigation zone priority data to obtain a weighted feature vector.

[0007] Optionally, the step of dynamically mapping the weighted feature vector to obtain the hydraulic feature vector through the dynamic ReLU activation function embedded in the first hidden layer includes: The first linearly transformed eigenvector is obtained by performing a linear transformation on the weighted eigenvector using the first weight matrix and the first bias vector. The first dynamic activation slope is calculated based on the average flow velocity data of the river cross-section. The first dynamic activation slope and the first linear transformation feature vector are input into the dynamic ReLU activation function to calculate the hydraulic feature vector.

[0008] Optionally, the step of extending the hydraulic feature vector through the adaptive LeakyReLU activation function and residual connection layer embedded in the second hidden layer to obtain the extended feature vector of the hydraulic equipment includes: The hydraulic eigenvectors are linearly transformed by the second weight matrix and the second bias vector to obtain the second linearly transformed eigenvectors. Input the equipment parameters into the residual connection layer and calculate the residual branch characteristics. The second dynamic activation slope is calculated based on the motor current data and the second linear transformation feature vector. The second dynamic activation slope, the second linear transformation feature vector, and the residual branch feature are input into the adaptive LeakyReLU activation function to calculate the extended feature vector of the hydraulic equipment.

[0009] Optionally, the step of calculating the target opening of the gate based on the extended feature vector of the hydraulic equipment using the adaptive Sigmoid activation function embedded in the opening calculation layer includes: The third linear transformation feature vector is obtained by performing a linear transformation on the extended feature vector of the hydraulic equipment using the third weight matrix and the third bias vector. Learnable parameters are obtained, and the solar power data is linearly transformed using the learnable parameters to obtain the first offset. The second offset is calculated based on the learnable parameters and the regional water demand gap data. The third linear transformation feature vector, the first offset, and the second offset are input into the adaptive Sigmoid activation function to calculate the final offset. The target opening of the sluice gate is calculated based on the measured opening and final offset of the sluice gate at the previous moment.

[0010] Optionally, after adjusting the opening of the sluice gate according to the opening control command, the following may also be included: The system obtains the actual water resource utilization rate, predicted water resource utilization rate, actual water supply satisfaction rate, and predicted water supply satisfaction rate of the water conservancy system, as well as the actual energy consumption, predicted energy consumption, rated energy consumption, and the measured opening degree of the sluice gate at the current moment. Water resource utilization rate loss is calculated based on actual and predicted water resource utilization rates; water supply satisfaction rate loss is calculated based on actual, predicted, and irrigation zone priority data; motor energy consumption loss is calculated based on actual, predicted, and rated energy consumption; and closed-loop feedback loss is calculated based on the measured opening and target opening of the sluice gate at the current moment. The total loss is calculated based on water resource utilization loss, water supply satisfaction loss, motor energy consumption loss, and closed-loop feedback loss. The parameters of the neural network scheduling and control unit are then optimized based on the total loss using the Adam gradient descent algorithm.

[0011] Secondly, the present invention provides a gate control device for a water conservancy system, used to implement the gate control method of the aforementioned water conservancy system, the device comprising: The control system construction module is used to build an integrated sluice gate control system. The integrated sluice gate control system includes a comprehensive acquisition unit, a neural network scheduling control unit, and a sluice gate execution unit. The comprehensive acquisition unit includes a water level sensor, a flow velocity sensor, an absolute encoder, a photovoltaic power sensor, and a camera. The neural network scheduling control unit includes a feature extraction layer, an attention feature weighting layer, a first hidden layer, a second hidden layer, and an opening degree calculation layer. The weighted feature vector acquisition module is used to acquire multi-source data through the comprehensive acquisition unit, input the multi-source data into the feature extraction layer for feature extraction and feature concatenation to obtain an initial feature vector, and then use the attention feature weighting layer to weight the initial feature vector to obtain a weighted feature vector. The hydraulic feature vector acquisition module is used to perform dynamic feature mapping on the weighted feature vector through the dynamic ReLU activation function embedded in the first hidden layer to obtain the hydraulic feature vector; The hydraulic equipment extended feature vector acquisition module is used to extend the hydraulic feature vector by using the adaptive LeakyReLU activation function and residual connection layer embedded in the second hidden layer to obtain the hydraulic equipment extended feature vector; The sluice gate target opening acquisition module is used to calculate the sluice gate target opening based on the hydraulic equipment extended feature vector by using the adaptive Sigmoid activation function embedded in the opening calculation layer. The gate adjustment module is used by the gate execution unit to generate opening control commands based on the target opening degree of the gate, and to adjust the opening degree of the gate according to the opening control commands.

[0012] Thirdly, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the gate control method of the water conservancy system.

[0013] Fourthly, the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the gate control method of the water conservancy system.

[0014] The present invention has the following beneficial effects: 1. After standardization, denoising, and concatenation of multi-source data by the feature extraction layer, an initial feature vector is obtained. Then, the attention feature weighting layer dynamically allocates feature weights according to the priority of irrigation zones to improve the targeting of decisions. Subsequently, the dynamic ReLU activation function of the first hidden layer can adaptively adjust the activation slope according to the river flow velocity to accurately capture the dynamic characteristics of water flow and generate a reliable hydraulic feature vector. The second hidden layer uses an adaptive LeakyReLU activation function combined with a residual connection layer to effectively transmit equipment physical parameters to avoid feature forgetting, and can also adjust the slope according to the motor load to achieve feature expansion under safety constraints, forming an expanded feature vector that takes into account both hydraulic characteristics and equipment status. The adaptive Sigmoid activation function of the opening calculation layer combines the energy supply mode and water demand priority, and incorporates mechanical limits and inertia coefficients to calculate a stable and reasonable target opening of the gate. Finally, the gate execution unit accurately executes the opening control command through real-time deviation correction, dual power supply mode switching, and multiple safety protection mechanisms to achieve stable opening adjustment, greatly improving the decision accuracy, working condition adaptability, and long-term stability of gate control.

[0015] 2. The first hidden layer maps the weighted feature vector to a high-dimensional space through a weight matrix, which can more accurately capture the complex nonlinear relationship between the total water inflow of the river and the water level and flow velocity of the zone, breaking through the limitations of low-dimensional space feature expression and making the hydraulic features more discriminative. The slope of the first dynamic activation is adaptively adjusted according to the average flow velocity of the river cross section, which effectively suppresses the monitoring noise interference in low flow velocity scenarios and completely solves the gradient vanishing problem of the traditional fixed ReLU activation function at low flow velocities, so that the feature mapping always adapts to the dynamic changes of water flow and improves the reliability of hydraulic features.

[0016] 3. The residual connection layer of the second hidden layer directly transmits equipment parameters, forming independent residual branch features, avoiding feature forgetting during deep neural network training, and ensuring that scheduling decisions are deeply bound to the physical state of the equipment; the second dynamic activation slope is adjusted in real time according to the motor current, ensuring scheduling flexibility and achieving optimal scheduling balance under safety constraints; after feature expansion, an extended feature vector of hydraulic equipment is formed, which not only retains the hydraulic characteristics of water flow evolution, but also incorporates equipment operating state constraints, so that subsequent opening calculations can simultaneously meet water demand and protect equipment safety, improving the comprehensiveness and feasibility of decision-making.

[0017] 4. The opening calculation layer adopts a dual-offset design to accurately adapt to dual power supply modes. The first offset is based on solar power adjustment. When the photovoltaic power is sufficient, the opening adjustment is smoother and the energy consumption is reduced. When the mains power is supplied, the response is faster and the water supply stability is guaranteed. The second offset focuses on the water demand gap of high-priority zones to ensure that the critical water demand is met first, and achieves dual optimization of energy adaptation and demand orientation. The first and second offsets are combined with the measured opening at the previous moment to calculate the target opening, so that the opening adjustment is in a smooth transition state, avoiding waterlogging or water shortage in the irrigation area caused by sudden changes in water flow velocity and water level, and ensuring the stability of the water conservancy system operation. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of a method according to an embodiment of the present invention; Figure 2 This is a structural diagram of the device according to an embodiment of the present invention. Detailed Implementation

[0020] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0021] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to and includes any or all possible combinations of one or more of the listed items.

[0022] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0023] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0024] Reference Figure 1 This invention provides a method for controlling the gates of a water conservancy system, comprising: The S100 constructs an integrated sluice gate control system, which includes a comprehensive acquisition unit, a neural network scheduling control unit, and a sluice gate execution unit. The comprehensive acquisition unit includes a water level sensor, a flow velocity sensor, an absolute encoder, a photovoltaic power sensor, and a camera. The neural network scheduling control unit includes a feature extraction layer, an attention feature weighting layer, a first hidden layer, a second hidden layer, and an opening degree calculation layer.

[0025] In some embodiments, the feature extraction layer takes into account six types of core data, such as the total inflow of the river and the current water level of the zone. First, it is normalized to the [0,1] interval through min-max standardization. Then, it uses convolution kernels to extract the local correlation features of each data. Finally, the multi-dimensional features are concatenated into an initial feature vector of a unified dimension, laying the foundation for subsequent weighted calculation.

[0026] The attention feature weighting layer uses irrigation zoning priority data as a basis to assign differentiated weights to the corresponding features of different zoning in the initial feature vector. The higher the priority of the zoning, the greater the feature weight, ensuring that scheduling decisions are tilted towards critical water demand.

[0027] The first hidden layer embeds a dynamic ReLU activation function. First, a linear transformation is performed on the weighted feature vector using a 32×12 dimensional weight matrix and a 32 dimensional bias vector. Then, the dynamic activation slope is calculated based on the average flow velocity of the river cross-section. The activation is more sensitive when the flow velocity is high, and noise is suppressed when the flow velocity is low. The output is a 32 dimensional hydraulic feature vector.

[0028] The second hidden layer combines the adaptive LeakyReLU activation function with the residual connection layer. It performs a linear transformation through a 16×32-dimensional weight matrix and a 16-dimensional bias vector. The residual branch directly transmits equipment parameters to avoid feature forgetting. At the same time, it adjusts the dynamic slope according to the motor current to prevent motor overload. Finally, it outputs a 16-dimensional extended feature vector of the hydraulic equipment.

[0029] The opening calculation layer adopts an adaptive Sigmoid activation function. After linear transformation, it integrates solar power and the water demand gap of the zone to calculate the dual offset, adds mechanical limit and opening inertia constraint, and finally combines the measured opening at the previous moment to output the target opening value of the sluice gate.

[0030] The S200 acquires multi-source data through a comprehensive acquisition unit, inputs the multi-source data into a feature extraction layer for feature extraction and feature concatenation to obtain an initial feature vector, and then uses an attention feature weighting layer to weight the initial feature vector to obtain a weighted feature vector.

[0031] In some embodiments, the integrated data acquisition unit includes a local monitoring unit, a digital twin unit, and an irrigation zoning unit. Acquiring multi-source data through the integrated data acquisition unit includes: Monitoring data of the water conservancy system are collected through on-site monitoring units; The monitoring data is input into the digital twin unit to construct a three-dimensional dynamic model and simulate water flow, thereby obtaining hydraulic characteristic data; The monitoring data is input into the irrigation zoning unit to calculate the irrigation zoning and water demand, and obtain water demand data. The data consists of monitoring data, hydraulic characteristic data, and water demand data. The multi-source data includes total river inflow data, current water level data for each zone, flow velocity data for each zone, current sluice gate opening data, solar power data, water demand gap data for each zone, irrigation zone priority data, average flow velocity data of the river cross section, equipment parameters, motor current data, and the measured opening of the sluice gate at the previous moment.

[0032] In some embodiments, the local monitoring unit includes a water level sensor, a flow velocity sensor, an absolute encoder, a photovoltaic power sensor, and a camera.

[0033] Water level sensors collect current water level data for each zone. Flow velocity sensors collect flow velocity data from different zones. The absolute encoder collects the current opening degree data of the sluice gate. The measured opening of Doumen at the previous moment Photovoltaic power sensors collect solar power data. Hall effect current sensors collect motor current data. All raw data were acquired at a sampling frequency of 10s / sample. High-frequency fluctuating data were processed using a 50Hz low-pass filter; outliers were identified using the 3σ criterion, and the moving average of the previous 5 samples was used to replace them when they were outliers; finally, the data were mapped to the [0,1] interval through min-max standardization.

[0034] After receiving standardized monitoring data, the digital twin unit constructs a 3D model based on BIM+GIS technology, including river topology and sluice gate structure, and embeds equipment parameter vectors. Including gate thickness River channel roughness Rated power of motor .

[0035] A numerical model was built based on the Saint-Venant equations, with the total inflow data of the river channel as input. Partition flow rate data Simulates the evolution of water flow and outputs average flow velocity data of the river cross section. , This includes the number of monitoring points at the cross-section and characteristic data such as the hydraulic response coefficient.

[0036] Irrigation zones are divided based on soil type and crop type using the K-means clustering algorithm. Each partition, irrigation partition priority data , where i is the partition number.

[0037] Calculation of crop evapotranspiration based on the Penman-Monteith formula Combined with soil water storage and effective rainfall ,pass Obtain water demand gap data for each region .

[0038] Monitoring data includes , , , , , , Hydraulic characteristic data includes , Water demand data includes , This constitutes multi-source data.

[0039] In some embodiments, the step of inputting multi-source data into a feature extraction layer for feature extraction and feature concatenation to obtain an initial feature vector, and then weighting the initial feature vector through an attention feature weighting layer to obtain a weighted feature vector, includes: The total inflow data of the river channel, the current water level data of each zone, the flow velocity data of each zone, the current opening degree of the sluice gates, the solar power data, and the water demand gap data of each zone are input into the feature extraction layer for feature extraction and feature concatenation to obtain the initial feature vector. The attention feature weighting layer weights each feature in the initial feature vector based on the irrigation zone priority data to obtain a weighted feature vector.

[0040] In some embodiments, the feature extraction layer contains 12 neurons, corresponding to 12-dimensional standardized multi-source data. Neuron 1 is input to the total inflow of the river channel, neurons 2-6 are input to the current water level data of the zone, neurons 7-8 are input to the flow velocity data of the zone, neurons 9-10 are input to the current opening of the sluice gate, neuron 11 is input to the solar power data, and neuron 12 is input to the water demand gap data of the zone. The features corresponding to each data are concatenated to form an initial feature vector X.

[0041] To enhance the scheduling contribution of key features, a single-head attention mechanism is introduced to dynamically allocate input feature weights, as shown in the following formula: in, , , All are 12×12 dimensional weight matrices set according to irrigation zone priority data. , , Both are 12-dimensional bias vectors. The square root of the feature dimension. For dynamic weight vectors, The algorithm uses a weighted feature vector, where Softmax() is the Softmax function and LayerNorm() is the layer normalization operation. It calculates feature similarity through the query-key vector inner product, and then uses the Softmax function to obtain a dynamic weight vector, increasing the weight of high-priority key features such as water demand gaps and motor current, while decreasing the weight of non-key features. To avoid loss of feature information, layer normalization stabilizes the training gradient, ensuring that the core information is fully preserved after the 12-dimensional features are weighted.

[0042] S300 obtains hydraulic feature vectors by dynamically mapping the weighted feature vectors through a dynamic ReLU activation function embedded in the first hidden layer.

[0043] In some embodiments, the step of dynamically mapping the weighted feature vector using a dynamic ReLU activation function embedded in the first hidden layer to obtain a hydraulic feature vector includes: The first linearly transformed eigenvector is obtained by performing a linear transformation on the weighted eigenvector using the first weight matrix and the first bias vector. The first dynamic activation slope is calculated based on the average flow velocity data of the river cross-section. The first dynamic activation slope and the first linear transformation feature vector are input into the dynamic ReLU activation function to calculate the hydraulic feature vector.

[0044] In some embodiments, to adapt to the dynamic changes in river flow, the first hidden layer employs a dynamic ReLU activation function and hydraulic feature vectors. The calculation formula is: in, The first weight matrix is ​​32×12 dimensional; The first bias vector is 32-dimensional; These are weighted eigenvectors; The eigenvectors of the first linear transformation; This is a learnable parameter, with an initial value of 1.0; This represents the average flow velocity data across the river cross-section. The first dynamic activation slope is tanh(); the hyperbolic tangent function is tanh(); and the dynamic ReLU activation function is max(), used to select the appropriate value. and The maximum value between.

[0045] First linear transformation eigenvectors Mapping 12-dimensional weighted feature vectors to a 32-dimensional high-dimensional space captures complex hydraulic correlations; the first dynamic activation slope Adaptive adjustment based on water flow conditions, flow velocity >0.6 ≈0.996, activating more sensitive response and enhancing dispatching response under water flow impact; flow velocity <0.2 ≈0.679, suppressing noise interference and solving the gradient vanishing problem of traditional fixed ReLU activation function in low flow rate scenarios.

[0046] The S400 extends the hydraulic feature vector by embedding an adaptive LeakyReLU activation function and a residual connection layer in the second hidden layer, thereby obtaining the extended feature vector of the hydraulic equipment.

[0047] In some embodiments, the step of extending the hydraulic feature vector through the adaptive LeakyReLU activation function and residual connection layer embedded in the second hidden layer to obtain the extended feature vector of the hydraulic equipment includes: The hydraulic eigenvectors are linearly transformed by the second weight matrix and the second bias vector to obtain the second linearly transformed eigenvectors. Input the equipment parameters into the residual connection layer and calculate the residual branch characteristics. The second dynamic activation slope is calculated based on the motor current data and the second linear transformation feature vector. The second dynamic activation slope, the second linear transformation feature vector, and the residual branch feature are input into the adaptive LeakyReLU activation function to calculate the extended feature vector of the hydraulic equipment.

[0048] In some embodiments, to strengthen equipment safety constraints and improve feature transfer efficiency, an adaptive LeakyReLU activation function and a residual connection layer are introduced in the second hidden layer to expand the feature vector of the hydraulic equipment. The calculation formula is: in, The second weight matrix is ​​16×32 dimensional; The second bias vector is 16-dimensional; The eigenvectors of the second linear transformation; 3D device parameters; It is a 16×3 dimensional residual branch weight matrix; R represents the 16-dimensional residual bias; R is the residual branch characteristic. () represents the Sigmoid function; This is the motor current data; This is the second dynamic activation slope; This represents the adaptive LeakyReLU activation function, in and Take the maximum value between the two values ​​and add it to R.

[0049] The residual branch feature R directly transmits device parameters, avoiding the deep network's forgetting of the correlation between scheduling decisions and device states; the second dynamic activation slope Adjust the load according to the motor load. >0.8 ≈0.802, suppressing the decision to increase the opening degree and avoid motor overload; load <0.3 ≈0.505, ensuring scheduling flexibility; after feature expansion, a 16-dimensional hydraulic equipment extended feature vector with hydraulic-equipment dual-dimensional support is formed.

[0050] The S500 calculates the target opening of the sluice gate based on the extended feature vector of the hydraulic equipment by using an adaptive sigmoid activation function embedded in the opening calculation layer.

[0051] In some embodiments, the step of calculating the target opening of the gate based on the extended feature vector of the hydraulic equipment using the adaptive Sigmoid activation function embedded in the opening calculation layer includes: The third linear transformation feature vector is obtained by performing a linear transformation on the extended feature vector of the hydraulic equipment using the third weight matrix and the third bias vector. Learnable parameters are obtained, and the solar power data is linearly transformed using the learnable parameters to obtain the first offset. The second offset is calculated based on the learnable parameters and the regional water demand gap data. The third linear transformation feature vector, the first offset, and the second offset are input into the adaptive Sigmoid activation function to calculate the final offset. The target opening of the sluice gate is calculated based on the measured opening and final offset of the sluice gate at the previous moment.

[0052] In some embodiments, to adapt to the mechanical characteristics and dual power supply mode of the gate execution unit, the opening calculation layer adopts an adaptive Sigmoid activation function with multiple constraints, and the gate target opening is... The calculation formula is: in, The third weight matrix is ​​of dimension n×16; Let be the third bias vector in n dimensions; The eigenvectors of the third linear transformation; β and γ are learnable parameters with initial values ​​of 0.5, 0.1, and 0.2, respectively. This represents solar power data; T is the first offset. This represents the highest priority water demand gap in the regional water demand gap data. This is the second offset; This is the final offset output by the adaptive Sigmoid activation function; =0.08 is the opening inertia coefficient; The measured opening of the sluice gate at the previous moment; max() is the maximum value function; min() is the minimum value function.

[0053] The first offset T adapts to dual power supply modes: when solar power is sufficient, the opening adjustment is smooth, reducing energy consumption; when powered by mains power, the response is faster, ensuring water supply. The second offset... Prioritize water demand in high-priority zones; adapt mechanical limits and inertial constraints of the adaptive Sigmoid activation function to match the mechanical characteristics of the dam gate, directly driving the dam gate execution unit.

[0054] The S600 gate execution unit generates opening control commands based on the target gate opening, and adjusts the gate opening according to the opening control commands.

[0055] In some embodiments, the gate actuator first obtains the current opening degree through an absolute encoder, calculates the deviation value between the current opening degree and the target opening degree, and generates an opening control command containing the adjustment direction, step size and duration in combination with the mechanical adjustment rate limit to avoid mechanical shock.

[0056] The system uses a servo motor with a reduction gear transmission mechanism to execute commands and automatically switches the power supply mode according to the solar power data. When the solar power is sufficient, photovoltaic power is activated, and when it is insufficient, the mains power is switched. The motor speed is controlled by PWM pulse signals to achieve smooth adjustment of the opening degree.

[0057] During the adjustment process, the absolute encoder collects the actual opening data 10 times per second and compares it with the target opening in real time. If the deviation is greater than 1%, the control command is dynamically corrected until the deviation between the actual opening and the target opening is ≤0.5%.

[0058] A series Hall current sensor monitors the motor current in real time. If the current exceeds 1.2 times the rated value, it immediately triggers overload protection, suspends adjustment, and sends an alarm. At the same time, a mechanical limit switch is set to prevent the opening degree from exceeding the 0-100% range and avoid equipment damage.

[0059] In some embodiments, after adjusting the opening of the gate according to the opening control command, the method further includes: The system obtains the actual water resource utilization rate, predicted water resource utilization rate, actual water supply satisfaction rate, and predicted water supply satisfaction rate of the water conservancy system, as well as the actual energy consumption, predicted energy consumption, rated energy consumption, and the measured opening degree of the sluice gate at the current moment. Water resource utilization rate loss is calculated based on actual and predicted water resource utilization rates; water supply satisfaction rate loss is calculated based on actual, predicted, and irrigation zone priority data; motor energy consumption loss is calculated based on actual, predicted, and rated energy consumption; and closed-loop feedback loss is calculated based on the measured opening and target opening of the sluice gate at the current moment. The total loss is calculated based on water resource utilization loss, water supply satisfaction loss, motor energy consumption loss, and closed-loop feedback loss. The parameters of the neural network scheduling and control unit are then optimized based on the total loss using the Adam gradient descent algorithm.

[0060] In some embodiments, to balance the three objectives of water resource utilization rate, water supply satisfaction rate, and motor energy consumption, closed-loop feedback calibration is incorporated, and a dynamically weighted multi-objective loss function is designed, with a total loss... The formula is as follows: in, , , For dynamic weights, =0.3 is the weight of the closed-loop feedback loss, ∑=1 / +1 / +1 / , For water resource utilization loss, For the loss of water supply satisfaction rate, For motor energy loss, This is the loss from closed-loop feedback.

[0061] The definitions of each sub-loss are as follows: Water resource utilization loss =0.5×MAE( , )+0.5×max(0, - )², For actual water resource utilization rate, To predict water resource utilization, MAE() is the mean absolute error function, and max() is the maximum value function; strong penalties are imposed for predicted utilization rates that are lower than the actual values, thus avoiding water resource waste. Water supply satisfaction rate loss : = ×[0.5×Huber( , )+0.5×QuantileLoss( , [,τ=0.9)], where N is the total number of partitions. This is priority data for irrigation zones. To represent the actual water supply satisfaction rate, To predict water supply satisfaction rate; Huber() is a robust loss function, which is equivalent to squared loss for small errors to ensure smoothness, and equivalent to absolute loss for large errors to suppress outlier interference; QuantileLoss() is a quantile loss function used to control the quantile of the predicted value, allowing the predicted value to cover a specified proportion of the actual value, focusing on high satisfaction rate scenarios, and prioritizing and weighting to ensure water supply for key crops.

[0062] Motor energy loss =0.3×MSE(E,Ê)+0.7×max (0,Ê-1.2 )², where E is the actual energy consumption and Ê is the predicted energy consumption. The rated energy consumption is denoted as MSE(), which is the mean square error function. A strong penalty is applied when the rated energy consumption exceeds 120% to prevent equipment overload.

[0063] Closed-loop feedback loss =MSE( - )+δ², This represents the measured opening of the Doumen Gate at the current moment. δ represents the target opening of Doumen, and δ represents the digital twin bias. By integrating execution feedback and twin calibration, the consistency between the model and the physical entity is improved.

[0064] The Adam gradient descent algorithm is used to optimize network parameters, and a dynamic learning rate and dual optimization mechanism are introduced, as shown in the following formula: in, For the network parameters before the update, Here are the updated network parameters, η(t) is the dynamic learning rate, η0=0.001 is the initial learning rate, and β1=0.9 and β2=0.999 are the momentum parameters. The number of iterations. The gradient represents the total loss. Regularly optimize and update all weights; replace old weights when the loss decreases by ≥5%. Freeze irrelevant weights through emergency optimization and rapidly improve accuracy through 30 iterations in a localized manner, ensuring the network adapts to changing operating conditions over the long term.

[0065] Reference Figure 2This invention provides a gate control device 20 for a water conservancy system, used to implement a gate control method for a water conservancy system. The device includes: The control system construction module 21 is used to construct an integrated sluice gate control system. The integrated sluice gate control system includes a comprehensive acquisition unit, a neural network scheduling control unit, and a sluice gate execution unit. The comprehensive acquisition unit includes a water level sensor, a flow velocity sensor, an absolute encoder, a photovoltaic power sensor, and a camera. The neural network scheduling control unit includes a feature extraction layer, an attention feature weighting layer, a first hidden layer, a second hidden layer, and an opening degree calculation layer. The weighted feature vector acquisition module 22 is used to acquire multi-source data through the comprehensive acquisition unit, input the multi-source data into the feature extraction layer for feature extraction and feature concatenation to obtain an initial feature vector, and then perform feature weighting on the initial feature vector through the attention feature weighting layer to obtain a weighted feature vector. The hydraulic feature vector acquisition module 23 is used to perform dynamic feature mapping on the weighted feature vector through the dynamic ReLU activation function embedded in the first hidden layer to obtain the hydraulic feature vector; The hydraulic equipment extended feature vector acquisition module 24 is used to extend the hydraulic feature vector by using the adaptive LeakyReLU activation function and residual connection layer embedded in the second hidden layer to obtain the hydraulic equipment extended feature vector. The sluice gate target opening acquisition module 25 is used to calculate the sluice gate target opening based on the hydraulic equipment extended feature vector by using the adaptive Sigmoid activation function embedded in the opening calculation layer. The gate adjustment module 26 is used by the gate execution unit to generate an opening control command based on the target opening of the gate, and to adjust the opening of the gate according to the opening control command.

[0066] This application provides an electronic device, including a processor and a memory; the memory stores a computer program, wherein the computer program, when executed by the processor, implements the gate control method of the water conservancy system according to any of the above schemes.

[0067] Specifically, the processor may include, for example, a general-purpose microprocessor, an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor may also include onboard memory for caching purposes. The processor may be a single processing unit or multiple processing units for performing different actions of the method flow according to embodiments of this application.

[0068] Memory can be any medium capable of containing, storing, transmitting, propagating, or transmitting instructions. For example, memory can include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, instruments, or propagation media. Specific examples of memory include: magnetic storage devices such as magnetic tape or hard disk drives (HDDs); optical storage devices such as optical discs (CD-ROMs); and also random access memory (RAM) or flash memory; and / or wired / wireless communication links.

[0069] This application also provides a computer-readable medium storing a computer program thereon, which, when executed by a processor, implements the gate control method of the water conservancy system according to any of the above-described schemes. This computer-readable medium may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The aforementioned computer-readable medium carries one or more programs, which, when executed, implement the method as described in the embodiments of this application.

[0070] According to embodiments of this application, a computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wired, optical fiber, radio frequency signals, etc., or any suitable combination thereof.

[0071] Those skilled in the art will understand that the features described in the various embodiments and / or claims of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments and / or claims of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application. Therefore, the scope of this application should not be limited to the above embodiments, but should be defined not only by the appended claims, but also by their equivalents. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for controlling the gates of a water conservancy system, characterized in that, include: An integrated sluice gate control system is constructed, which includes a comprehensive acquisition unit, a neural network scheduling and control unit, and a sluice gate execution unit. The comprehensive acquisition unit includes a water level sensor, a flow velocity sensor, an absolute encoder, a photovoltaic power sensor, and a camera. The neural network scheduling and control unit includes a feature extraction layer, an attention feature weighting layer, a first hidden layer, a second hidden layer, and an opening degree calculation layer. Multi-source data is acquired through a comprehensive acquisition unit, and then input into a feature extraction layer for feature extraction and feature concatenation to obtain an initial feature vector. The initial feature vector is then weighted by an attention feature weighting layer to obtain a weighted feature vector. By using the dynamic ReLU activation function embedded in the first hidden layer, dynamic feature mapping is performed on the weighted feature vector to obtain the hydraulic feature vector; By embedding an adaptive LeakyReLU activation function and a residual connection layer in the second hidden layer, the hydraulic feature vector is extended to obtain the extended feature vector of the hydraulic equipment. The target opening of the sluice gate is obtained by using an adaptive sigmoid activation function embedded in the opening calculation layer, based on the extended feature vector of the hydraulic equipment. The gate execution unit generates an opening control command based on the target opening degree of the gate, and adjusts the opening degree of the gate according to the opening control command.

2. The gate control method for a water conservancy system according to claim 1, characterized in that, The integrated data acquisition unit includes a local monitoring unit, a digital twin unit, and an irrigation zoning unit. The acquisition of multi-source data through the integrated data acquisition unit includes: Monitoring data of the water conservancy system are collected through on-site monitoring units; The monitoring data is input into the digital twin unit to construct a three-dimensional dynamic model and simulate water flow, thereby obtaining hydraulic characteristic data; The monitoring data is input into the irrigation zoning unit to calculate the irrigation zoning and water demand, and obtain water demand data. The data consists of monitoring data, hydraulic characteristic data, and water demand data. The multi-source data includes total river inflow data, current water level data for each zone, flow velocity data for each zone, current sluice gate opening data, solar power data, water demand gap data for each zone, irrigation zone priority data, average flow velocity data of the river cross section, equipment parameters, motor current data, and the measured opening of the sluice gate at the previous moment.

3. The gate control method for a water conservancy system according to claim 2, characterized in that, The process of inputting multi-source data into a feature extraction layer for feature extraction and feature concatenation to obtain an initial feature vector, and then weighting the initial feature vector through an attention feature weighting layer to obtain a weighted feature vector, includes: The total inflow data of the river channel, the current water level data of each zone, the flow velocity data of each zone, the current opening degree of the sluice gates, the solar power data, and the water demand gap data of each zone are input into the feature extraction layer for feature extraction and feature concatenation to obtain the initial feature vector. The attention feature weighting layer weights each feature in the initial feature vector based on the irrigation zone priority data to obtain a weighted feature vector.

4. The gate control method for a water conservancy system according to claim 2, characterized in that, The process of dynamically mapping the weighted feature vector using the dynamic ReLU activation function embedded in the first hidden layer to obtain the hydraulic feature vector includes: The first linearly transformed eigenvector is obtained by performing a linear transformation on the weighted eigenvector using the first weight matrix and the first bias vector. The first dynamic activation slope is calculated based on the average flow velocity data of the river cross-section. The first dynamic activation slope and the first linear transformation feature vector are input into the dynamic ReLU activation function to calculate the hydraulic feature vector.

5. The gate control method for a water conservancy system according to claim 2, characterized in that, The process of extending the hydraulic feature vector by embedding an adaptive LeakyReLU activation function and a residual connection layer in the second hidden layer to obtain the extended feature vector of the hydraulic equipment includes: The hydraulic eigenvectors are linearly transformed by the second weight matrix and the second bias vector to obtain the second linearly transformed eigenvectors. Input the equipment parameters into the residual connection layer and calculate the residual branch characteristics. The second dynamic activation slope is calculated based on the motor current data and the second linear transformation feature vector. The second dynamic activation slope, the second linear transformation feature vector, and the residual branch feature are input into the adaptive LeakyReLU activation function to calculate the extended feature vector of the hydraulic equipment.

6. The gate control method for a water conservancy system according to claim 2, characterized in that, The process of calculating the target opening of the sluice gate using the adaptive Sigmoid activation function embedded in the opening calculation layer, based on the extended feature vector of the hydraulic equipment, includes: The third linear transformation feature vector is obtained by performing a linear transformation on the extended feature vector of the hydraulic equipment using the third weight matrix and the third bias vector. Learnable parameters are obtained, and the solar power data is linearly transformed using the learnable parameters to obtain the first offset. The second offset is calculated based on the learnable parameters and the regional water demand gap data. The third linear transformation feature vector, the first offset, and the second offset are input into the adaptive Sigmoid activation function to calculate the final offset. The target opening of the sluice gate is calculated based on the measured opening and final offset of the sluice gate at the previous moment.

7. The gate control method for a water conservancy system according to claim 2, characterized in that, After adjusting the opening of the sluice gate according to the opening control command, the following is also included: The system obtains the actual water resource utilization rate, predicted water resource utilization rate, actual water supply satisfaction rate, and predicted water supply satisfaction rate of the water conservancy system, as well as the actual energy consumption, predicted energy consumption, rated energy consumption, and the measured opening degree of the sluice gate at the current moment. Water resource utilization rate loss is calculated based on actual and predicted water resource utilization rates; water supply satisfaction rate loss is calculated based on actual, predicted, and irrigation zone priority data; motor energy consumption loss is calculated based on actual, predicted, and rated energy consumption; and closed-loop feedback loss is calculated based on the measured opening and target opening of the sluice gate at the current moment. The total loss is calculated based on water resource utilization loss, water supply satisfaction loss, motor energy consumption loss, and closed-loop feedback loss. The parameters of the neural network scheduling and control unit are then optimized based on the total loss using the Adam gradient descent algorithm.

8. A gate control device for a water conservancy system, used to implement the gate control method for a water conservancy system as described in any one of claims 1 to 7, characterized in that, The device includes: The control system construction module is used to build an integrated sluice gate control system. The integrated sluice gate control system includes a comprehensive acquisition unit, a neural network scheduling control unit, and a sluice gate execution unit. The comprehensive acquisition unit includes a water level sensor, a flow velocity sensor, an absolute encoder, a photovoltaic power sensor, and a camera. The neural network scheduling control unit includes a feature extraction layer, an attention feature weighting layer, a first hidden layer, a second hidden layer, and an opening degree calculation layer. The weighted feature vector acquisition module is used to acquire multi-source data through the comprehensive acquisition unit, input the multi-source data into the feature extraction layer for feature extraction and feature concatenation to obtain an initial feature vector, and then use the attention feature weighting layer to weight the initial feature vector to obtain a weighted feature vector. The hydraulic feature vector acquisition module is used to perform dynamic feature mapping on the weighted feature vector through the dynamic ReLU activation function embedded in the first hidden layer to obtain the hydraulic feature vector; The hydraulic equipment extended feature vector acquisition module is used to extend the hydraulic feature vector by using the adaptive LeakyReLU activation function and residual connection layer embedded in the second hidden layer to obtain the hydraulic equipment extended feature vector; The sluice gate target opening acquisition module is used to calculate the sluice gate target opening based on the hydraulic equipment extended feature vector by using the adaptive Sigmoid activation function embedded in the opening calculation layer. The gate adjustment module is used by the gate execution unit to generate opening control commands based on the target opening degree of the gate, and to adjust the opening degree of the gate according to the opening control commands.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the gate control method of the water conservancy system as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the gate control method of the water conservancy system as described in any one of claims 1 to 7.

Images