Water conservancy project water delivery capacity adjusting method and system based on artificial intelligence

Abstract

The application relates to the technical field of hydraulic engineering, and discloses a water delivery capacity regulation method and system for hydraulic engineering based on artificial intelligence. The method comprises the following steps: acquiring multi-source real-time data in the hydraulic engineering, extracting local features in the input data based on a CNN (convolutional neural network), processing long-term dependence of a time sequence through an LSTM (long short-term memory) neural network, paying attention to water delivery variables by using an Attention mechanism, inputting the initial multi-source real-time data into a demand prediction model for prediction to obtain a water delivery demand curve, setting a target function as minimization of energy consumption and maximization of supply-demand matching degree through a PPO (proximal policy optimization) reinforcement learning algorithm according to the water delivery demand curve, and obtaining water delivery control instructions by simulating different control actions. The method significantly improves the accuracy of water delivery demand prediction and can more accurately capture the relationship between complex factors in the hydraulic engineering and water delivery demand.

Description

Technical Field

[0001] The present invention relates to the technical field of water conservancy projects, and in particular to an artificial intelligence-based water delivery regulation method and system for water conservancy projects. Background Art

[0002] Traditionally, water flow regulation in hydraulic projects has relied primarily on manual experience and simple automated equipment. This approach has numerous drawbacks. Limited by individual experience and knowledge, manual operations struggle to accurately and accurately respond to complex and changing hydraulic environments in real time. In the face of extreme weather conditions like sudden rainfall and drought, rapid and effective water flow regulation decisions cannot be made, leading to water waste or insufficient supply. Simple automated equipment operates solely according to pre-set, fixed rules and is unable to adapt to the complex multi-factor coupling of hydraulic projects. Summary of the Invention

[0003] The purpose of the present invention is to solve the above problems and to design a water delivery regulation method and system for water conservancy projects based on artificial intelligence.

[0004] To achieve the above-mentioned purpose, the technical solution of the present invention is as follows: further, in the above-mentioned artificial intelligence-based water conservancy project water delivery adjustment method, the water conservancy project water delivery adjustment method includes the following steps:

[0005] Acquiring multi-source real-time data from a water conservancy project, and performing data preprocessing on the multi-source real-time data to obtain initial multi-source real-time data;

[0006] The CNN convolutional neural network extracts local features from the input data, the LSTM neural network processes the long-term dependencies of the time series, and the attention mechanism is used to focus on the water delivery variables, resulting in a CNN-LSTM-Attention demand forecasting model.

[0007] Inputting the initial multi-source real-time data into the CNN-LSTM-Attention demand forecasting model for prediction to obtain a water delivery demand curve;

[0008] According to the water delivery demand curve, the objective function is set to minimize energy consumption and maximize supply-demand matching through the PPO reinforcement learning algorithm. By simulating different control actions, water delivery control instructions are obtained, and the gate opening and the start and stop status of the pump station are adjusted according to the water delivery control instructions.

[0009] Furthermore, in the above-mentioned artificial intelligence-based water delivery regulation method for a water conservancy project, the step of obtaining multi-source real-time data from the water conservancy project and performing data preprocessing on the multi-source real-time data to obtain initial multi-source real-time data includes:

[0010] Obtain multi-source real-time data on water conservancy projects, including at least rainfall, evaporation, reservoir water levels, water pipeline pressure, and downstream water demand;

[0011] Combining the Z-score method and the IQR method to perform outlier detection on the multi-source real-time data, and using a linear interpolation method to fill in missing values, to obtain first multi-source real-time data;

[0012] Mapping the first multi-source real-time data to the interval [0, 1] using a Min-Max normalization method, and uniformly converting the collected data in different formats into numerical data to obtain second multi-source real-time data;

[0013] The second multi-source real-time data is time-synchronized based on the time of the data center server, and the data is spatially aligned based on the geographic information of the water conservancy project and the location coordinates of each monitoring point to obtain the initial multi-source real-time data.

[0014] Furthermore, in the aforementioned artificial intelligence-based water delivery regulation method for a water conservancy project, the CNN-based convolutional neural network extracts local features from the input data, the LSTM neural network processes the long-term dependencies of the time series, and the attention mechanism is used to focus on the water delivery variables, thereby obtaining a CNN-LSTM-Attention demand forecasting model, including:

[0015] The first convolution layer of the CNN convolutional neural network uses 64 convolution kernels of size (3,1) to extract local features from the input multi-source time series data;

[0016] The second convolution layer uses 128 convolution kernels of different sizes to further extract complex local features;

[0017] The third convolutional layer uses 256 convolution kernels of different sizes to capture local information. The stride of the convolutional layer is set to 1 and the padding method is the same.

[0018] Furthermore, in the aforementioned artificial intelligence-based water delivery regulation method for a water conservancy project, the CNN-based convolutional neural network extracts local features from the input data, the LSTM neural network processes the long-term dependencies of the time series, and the attention mechanism is used to focus on the water delivery variables, thereby obtaining a CNN-LSTM-Attention demand forecasting model, including:

[0019] The first LSTM layer of the LSTM neural network receives the local feature data output by the CNN convolutional neural network and converts it into a time series data format;

[0020] The second layer LSTM layer is used for further processing the output of the first layer LSTM layer, enhancing the model's ability to capture long-term dependencies of time series, and the dropout rate of the LSTM layer is set to 0.2, and the return sequence is set to True.

[0021] Further, in the above-mentioned water conveyance capacity regulation method for water conservancy projects based on artificial intelligence, the CNN convolutional neural network is used to extract local features in the input data, the LSTM neural network is used to process long-term dependencies of time series, and the Attention mechanism is used to pay attention to the water conveyance variables to obtain a CNN-LSTM-Attention demand prediction model, which includes:

[0022] The hidden state output by the LSTM neural network is taken as Query, Key and Value, the dot product of Query and Key is calculated and divided by a scaling factor to obtain an attention score;

[0023] The attention score is normalized by using a Softmax function to obtain the attention weight of each input element;

[0024] Finally, the Value is weighted and summed according to the attention weight, and the importance between the water conveyance demand and each input variable is automatically learned through the Attention mechanism.

[0025] Further, in the above-mentioned water conveyance capacity regulation method for water conservancy projects based on artificial intelligence, the PPO reinforcement learning algorithm is used to set the objective function as minimizing energy consumption and maximizing supply-demand matching degree according to the water conveyance demand curve, and the water conveyance control instruction is obtained by simulating different control actions, which includes:

[0026] The state space and state space are defined, and the state transition equation is established according to the physical model of the water conservancy project and the dynamic characteristics of the water conveyance system;

[0027] The objective function is designed to give positive rewards when energy consumption is reduced and supply-demand matching degree is improved; and negative rewards when energy consumption is increased or supply-demand matching degree is reduced.

[0028] Further, in the above-mentioned water conveyance capacity regulation method for water conservancy projects based on artificial intelligence, the PPO reinforcement learning algorithm is used to set the objective function as minimizing energy consumption and maximizing supply-demand matching degree according to the water conveyance demand curve, and the water conveyance control instruction is obtained by simulating different control actions, which further includes:

[0029] Different control actions are simulated in the state space by using the PPO reinforcement learning algorithm, and the reward value of each action is calculated according to the state transition equation and the reward function;

[0030] The control action that minimizes the objective function is selected as the optimal water conveyance control instruction.

[0031] Further, in an artificial intelligence-based water conservancy project water delivery amount regulation system, the water conservancy project water delivery amount regulation system comprises the following modules:

[0032] A real-time data acquisition module is configured to acquire multi-source real-time data in a water conservancy project, perform data preprocessing on the multi-source real-time data, and obtain initial multi-source real-time data.

[0033] A prediction model establishment module is configured to extract local features in input data based on a CNN convolutional neural network, process long-term dependencies of time series through an LSTM neural network, and use an Attention mechanism to pay attention to water delivery variables to obtain a CNN-LSTM-Attention demand prediction model.

[0034] A water delivery demand prediction module is configured to input the initial multi-source real-time data into the CNN-LSTM-Attention demand prediction model for prediction to obtain a water delivery demand curve.

[0035] A water delivery control and regulation module is configured to set a target function as minimizing energy consumption and maximizing supply-demand matching degree through a PPO reinforcement learning algorithm according to the water delivery demand curve, obtain water delivery control instructions by simulating different control actions, and adjust gate opening and pump station start-stop states according to the water delivery control instructions.

[0036] Further, in an artificial intelligence-based water conservancy project water delivery amount regulation system, the water delivery control and regulation module comprises the following sub-modules:

[0037] A definition sub-module is configured to define state space and state space, establish a state transition equation according to a physical model of a water conservancy project and dynamic characteristics of a water delivery system.

[0038] A judgment sub-module is configured to design a target function, give a positive reward when energy consumption is reduced and supply-demand matching degree is improved, and give a negative reward when energy consumption is increased or supply-demand matching degree is reduced.

[0039] Further, in an artificial intelligence-based water conservancy project water delivery amount regulation system, the water delivery control and regulation module comprises the following sub-modules:

[0040] A simulation sub-module is configured to simulate different control actions in state space through a PPO reinforcement learning algorithm, and calculate a reward value of each action according to a state transition equation and a reward function.

[0041] A selection sub-module is configured to select a control action that minimizes the target function as an optimal water delivery control instruction.

[0042] The beneficial effects are that, by acquiring multi-source real-time data in a water conservancy project, the multi-source real-time data are preprocessed to obtain initial multi-source real-time data; a CNN convolutional neural network is used to extract local features in input data, an LSTM neural network is used to process long-term dependence of time series, an Attention mechanism is used to pay attention to water delivery variables, and a CNN-LSTM-Attention demand prediction model is obtained; the initial multi-source real-time data are input into the CNN-LSTM-Attention demand prediction model for prediction, and a water delivery demand curve is obtained; according to the water delivery demand curve, a PPO reinforcement learning algorithm is used to set a target function as minimizing energy consumption and maximizing supply-demand matching degree, water delivery control instructions are obtained by simulating different control actions, and gate opening and pump station start-stop states are adjusted according to the water delivery control instructions. 1. Compared with a single model or a traditional prediction method, the accuracy of water delivery demand prediction is significantly improved, the relationship between complex factors in a water conservancy project and water delivery demand can be more accurately captured, a reliable prediction basis is provided for water delivery control, and unreasonable allocation of water resources caused by prediction deviation is reduced. 2. The influence of various operations on energy consumption and supply-demand matching degree can be comprehensively evaluated, and the optimal water delivery control instructions are generated. Compared with a traditional water delivery control mode, the scheme can dynamically adjust the gate opening and the pump station start-stop state according to real-time data and prediction results, meet the downstream water demand, effectively reduce the energy consumption, realize efficient use of water resources and optimization of water conservancy operation cost, and improve the economic benefit of the water conservancy project. BRIEF DESCRIPTION OF DRAWINGS

[0043] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments with reference made to the accompanying drawings. The drawings are for purposes of illustration only and are not intended to be limiting of the present application.

[0044] Figure 1 The first embodiment schematic diagram of a water conservancy water delivery amount adjustment method based on artificial intelligence in an embodiment of the present application;

[0045] Figure 2 The second embodiment schematic diagram of a water conservancy water delivery amount adjustment method based on artificial intelligence in an embodiment of the present application;

[0046] Figure 3 The first embodiment schematic diagram of a water conservancy water delivery amount adjustment system based on artificial intelligence in an embodiment of the present application. DETAILED DESCRIPTION

[0047] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0048] Those skilled in the art can understand that the singular forms "a", "an" and "the" used herein include plural forms, unless specifically stated otherwise. It should be further understood that the use of the term "comprise" in the specification of the present application means that the features, integers, steps, operations, elements and / or components are present, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0049] The present application will be specifically described below with reference to the accompanying drawings, Figure 1 As shown, an artificial intelligence-based water conservancy project water delivery capacity adjusting method comprises the following steps:

[0050] Step 101, acquiring multi-source real-time data in a water conservancy project, performing data preprocessing on the multi-source real-time data to obtain initial multi-source real-time data;

[0051] Specifically, in the present embodiment, the multi-source real-time data in the water conservancy project includes at least rainfall, evaporation, reservoir water level, water delivery pipeline pressure and downstream water demand;

[0052] The Z-score method and the IQR method are combined to detect outliers of the multi-source real-time data, and the linear interpolation method is used to fill in missing values to obtain first multi-source real-time data;

[0053] The Min-Max normalization method is used to map the first multi-source real-time data to the interval [0, 1], and different formats of data collected are uniformly converted into numerical data to obtain second multi-source real-time data;

[0054] Taking the time of the data center server as the reference, the second multi-source real-time data is processed for time synchronization, and according to the geographical information of the water conservancy project and the position coordinates of each monitoring point, the data is processed for spatial alignment to obtain initial multi-source real-time data.

[0055] Specifically,

[0056] (I) Multi-source real-time data acquisition;

[0057] Data acquisition equipment and installation:

[0058] Rainfall: A tipping bucket rain gauge is installed in an open, unobstructed area within the water conservancy project basin to ensure accurate collection of natural rainfall data. The sensor collects rainfall data once per minute and transmits the data in real time to the data center server via a 4G wireless communication module.

[0059] Evaporation: An E601B evaporator is installed in a flat, well-ventilated observation area that complies with national meteorological observation standards. Evaporation data is recorded hourly and transmitted to a data center using an NB-IoT wireless communication module.

[0060] Reservoir water level: A radar water level meter is installed at a suitable location upstream of the reservoir dam to avoid being affected by factors such as water flow disturbances and aquatic plants. Reservoir water level is collected every 10 minutes and the data is sent to the data center via a 4G communication module.

[0061] Water pipeline pressure: A pressure transmitter is installed in a straight section of the water pipeline, away from valves, elbows, and other components that could interfere with pressure measurement. The pressure transmitter collects pipeline pressure data once per second and transmits the data to a data center via 4G wireless communication.

[0062] Downstream water demand: This forecast combines real-time user demand reporting data with historical water demand data. Users submit real-time water demand reports through a dedicated water conservancy project water demand reporting system, which updates user demand reporting data hourly. Historical water demand data forecasts are based on downstream user water use records from the past three years and are updated daily at dawn using an ARIMA (Asynchronous Recognition and Integrated Molecular Analyses) algorithm to generate water demand forecasts for the next period.

[0063] Data collection frequency and communication method:

[0064] Each sensor and data source collects data at the specified frequency and transmits the data to the data center server in real time via 4G or NB-IoT wireless communication modules to ensure real-time and reliable data. The data center server has a dedicated database to store and manage the collected multi-source data.

[0065] (2) Data preprocessing;

[0066] Data cleaning:

[0067] Outlier detection: Z-score method and IQR method are combined to detect outliers in the collected data. Z-score method calculates the standard deviation of the data points from the mean to determine outliers. Data points that are more than ±3 times the standard deviation from the mean are considered outliers. IQR method calculates the interquartile range (IQR = Q3 - Q1) and considers data points less than Q1 - 1.5IQR or greater than Q3 + 1.5IQR as outliers. For detected outliers, the average or median of adjacent time points is used to fill in the missing data based on the time series characteristics of the data.

[0068] Missing value processing: Linear interpolation is used to fill in missing data within a short period of time (data missing within a few minutes). For long-term missing data (data missing for more than an hour), historical data and related factors (rainfall, downstream water demand, etc.) are used to make a regression prediction to fill in the missing data.

[0069] Data conversion:

[0070] Normalization method: According to the characteristics of the data and the requirements of the model, Min-Max normalization and Z-score standardization are combined to normalize the data. For data with obvious range such as rainfall, evaporation, and reservoir water level, Min-Max normalization is used to map the data to the interval [0, 1]. For data such as water pipeline pressure and downstream water demand, Z-score standardization is used to make the data have zero mean and unit variance.

[0071] Data format conversion: The collected data of different formats (text format, numerical format, etc.) are converted to numerical data to facilitate subsequent model input and processing.

[0072] Data integration:

[0073] Time synchronization: The data collected by each sensor and data source is time-synchronized based on the time of the data center server to ensure that all data has the same timestamp. Data with a time deviation greater than a certain threshold (1 minute) is considered invalid and is cleaned up.

[0074] Spatial alignment: According to the geographical information of the water conservancy project and the location coordinates of each monitoring point, the multi-source data is spatially aligned to associate the monitoring data of different locations with the water delivery system structure of the water conservancy project, facilitating subsequent model analysis and control command generation.

[0075] Step 102, based on the CNN convolutional neural network, local features in the input data are extracted, long-term dependencies of time series are processed through the LSTM neural network, and the Attention mechanism is used to focus on the water delivery variables to obtain a CNN-LSTM-Attention demand prediction model;

[0076] Specifically, in the CNN convolutional neural network in this embodiment, the first convolutional layer uses 64 convolutional kernels of size (3, 1) to extract local features from the input multi-source time series data.

[0077] The second convolutional layer uses 128 convolutional kernels of size to further extract complex local features.

[0078] The third convolutional layer uses 256 convolutional kernels of size to capture local information, and the step length of the convolutional layer is set to 1 and the padding mode is same.

[0079] The first LSTM layer of the LSTM neural network receives the local feature data output by the CNN convolutional neural network and converts it into time series data format.

[0080] The second LSTM layer is used to further process the output of the first LSTM layer to enhance the model's ability to capture long-term dependencies of time series, and the dropout rate of the LSTM layer is set to 0.2 and the return sequence is set to True.

[0081] The hidden state output by the LSTM neural network is used as Query, Key and Value, the dot product of Query and Key is calculated and divided by a scaling factor to obtain the attention score.

[0082] The attention score is normalized using the Softmax function to obtain the attention weight of each input element.

[0083] Finally, the Value is weighted and summed according to the attention weight, and the importance of the water delivery demand and each input variable is automatically learned through the Attention mechanism.

[0084] Specifically:

[0085] (I) CNN convolutional neural network;

[0086] Network structure: The CNN convolutional neural network is composed of 3 convolutional layers and 3 max-pooling layers alternately.

[0087] The first convolutional layer uses 64 convolution kernels of size (3, 1) to extract local features from the input multi-source time series data. The convolution kernel has a sliding window size of 3 in the time dimension and 1 in the feature dimension to capture local correlations in the data over time. ReLU is used as the activation function to introduce nonlinearity and enhance the model's expressiveness.

[0088] The first maximum pooling layer: The pooling kernel size is (2,1), which downsamples the data output by the convolutional layer to reduce the data dimension, reduce the computational complexity of the model, and avoid overfitting.

[0089] The second convolutional layer uses 128 convolution kernels of size (5, 1) to further extract more complex local features. Larger convolution kernels can capture local information over a longer time range. The activation function also uses ReLU.

[0090] The second maximum pooling layer: The pooling kernel size is (2,1), and the data output by the second convolutional layer is downsampled.

[0091] The third convolutional layer uses 256 convolution kernels of size (7, 1) to capture richer local information. Increasing the number of convolution kernels improves the model's ability to extract different features. The activation function is ReLU.

[0092] The third maximum pooling layer: The pooling kernel size is (2,1), and the data output by the third convolutional layer is downsampled to obtain the final local feature representation.

[0093] Parameter settings: The step size of the convolution layer is set to 1 to ensure that the local features of the data can be fully extracted; the padding method is same, so that the time dimension of the data remains unchanged after the convolution operation, which facilitates subsequent pooling operations and connection with the LSTM neural network.

[0094] (2) LSTM neural network;

[0095] Network structure: The LSTM neural network consists of two hidden layers, each containing 128 neurons.

[0096] The first LSTM layer receives local feature data output by the CNN convolutional neural network and converts it into a time series data format, allowing the LSTM neural network to process long-term dependencies in the time series. LSTM neurons control the flow of information through forget gates, input gates, and output gates, effectively capturing data dependencies in long time series.

[0097] The second LSTM layer further processes the output of the first LSTM layer to enhance the model's ability to capture long-term dependencies in time series. Both LSTM layers use a bidirectional LSTM structure to consider both past and future time series information, improving the model's prediction accuracy.

[0098] Parameter settings: The dropout rate of the LSTM layer is set to 0.2 to prevent overfitting; the return sequence is set to True to pass the output of the LSTM layer to the Attention mechanism for processing.

[0099] (Three) Attention mechanism;

[0100] Attention type: Scaled Dot-Product Attention is used, which can efficiently calculate the correlation between elements in the input sequence and focus on input variables.

[0101] Operation process: The hidden state output by the LSTM neural network is used as Query, Key and Value. The dot product of Query and Key is calculated and divided by a scaling factor (sqrt(d_k), where d_k is the dimension of Key), to get the attention score. Then use the Softmax function to normalize the attention score to get the attention weight of each input element. Finally, the weighted sum of Value is obtained according to the attention weight, and the final output is obtained. Through the Attention mechanism, the model can automatically learn the importance of water demand and input variables (rainfall, reservoir water level, downstream water demand, etc.), and pay more attention to variables that have a greater impact on water demand.

[0102] (Four) Model integration;

[0103] The local features extracted by the CNN convolutional neural network are input into the LSTM neural network to process the long-term dependencies of time series, and then the Attention mechanism is used to focus on water variables, and finally the CNN-LSTM-Attention demand prediction model is obtained. The output of the model is the predicted value of water demand in the future period.

[0104] Step 103, input the initial multi-source real-time data into the CNN-LSTM-Attention demand prediction model for prediction to obtain the water demand curve;

[0105] Specifically, in this embodiment:

[0106] (I) Data input format;

[0107] The pre-processed multi-source real-time data is arranged in time series into a multi-dimensional array with dimensions [sample number, time step, feature number]. Among them, the sample number is determined according to the size of the training data, the time step is the length of the time window input to the model (the past 24 hours of time data), and the feature number is the feature dimension of the collected multi-source data (including rainfall, evaporation, reservoir water level, water pipeline pressure and downstream water demand, etc.).

[0108] (II) Model training process;

[0109] Data set division: The historical data is divided into training set, validation set and test set, of which the training set accounts for 70%, the validation set accounts for 20%, and the test set accounts for 10%. The training set is used for parameter learning of the model, the validation set is used for hyperparameter adjustment and overfitting detection of the model, and the test set is used for evaluating the final performance of the model.

[0110] Optimizer selection: Adam optimizer is adopted, which combines momentum method and adaptive learning rate method, and can automatically adjust learning rate during training process to improve training efficiency and convergence speed of the model.

[0111] Loss function definition: Mean Squared Error (MSE) is used as the loss function to measure the difference between the predicted value and the actual water demand value.

[0112] Training process: The training set is used to iteratively train the model, and the loss function is calculated each time. The model parameters are updated through the backpropagation algorithm. During the training process, the performance of the model is evaluated regularly using the validation set, and the hyperparameters (learning rate, number of convolution kernels, number of LSTM neurons, etc.) are adjusted according to the loss value of the validation set to avoid overfitting and improve the generalization ability of the model. When the loss values of the training set and the validation set no longer decrease significantly, the training is stopped, and the optimal model parameters are obtained.

[0113] (III) Model prediction;

[0114] The pre-processed initial multi-source real-time data is input into the trained CNN-LSTM-Attention demand prediction model according to the input format, and the model outputs the water demand prediction value in the future period. The prediction values are connected to obtain the water demand curve.

[0115] Step 104, according to the water demand curve, set the objective function as minimizing energy consumption and maximizing supply-demand matching degree through PPO reinforcement learning algorithm, obtain water control instructions through simulating different control actions, and adjust the gate opening and pump station start-stop state according to the water control instructions.

[0116] Specifically, in this embodiment, the state space and state space are defined, and the state transition equation is established according to the physical model of the water conservancy project and the dynamic characteristics of the water conveyance system.

[0117] The objective function is designed to give positive rewards when energy consumption is reduced and supply-demand matching degree is improved, and negative rewards when energy consumption is increased or supply-demand matching degree is reduced.

[0118] Different control actions are simulated in the state space through the PPO reinforcement learning algorithm, and the reward value of each action is calculated according to the state transition equation and the reward function.

[0119] The control action that minimizes the objective function is selected as the optimal water delivery control instruction.

[0120] Specifically:

[0121] (1) PPO reinforcement learning algorithm;

[0122] Objective function setting: The objective function is to minimize the weighted sum of energy consumption and maximize the supply-demand matching degree. By adjusting the weight coefficient, the importance of energy consumption and supply-demand matching degree can be balanced. Energy consumption mainly includes the power consumption of the pumping station and the mechanical energy consumption of the gate regulation, etc. The supply-demand matching degree is measured by calculating the matching degree between the actual water delivery amount and the predicted water demand value. The reciprocal of the root mean square error (RMSE) is used as an indicator of the supply-demand matching degree, i.e. supply-demand matching degree = 1 / RMSE.

[0123] State space definition: The state space includes real-time data such as current reservoir water level, water delivery pipeline pressure, downstream water demand, gate opening, pump station start-stop state, etc., as well as predicted values of water demand curve information. The dimension of the state vector is determined according to the actual water conservancy engineering system parameters.

[0124] Action space definition: Control actions include gate opening and pump station start-stop state. The gate opening value ranges from 0 to 100%, adjusted in steps of 1%. The pump station start-stop state is 0 or 1, 0 indicating that the pump station is stopped, and 1 indicating that the pump station is started.

[0125] State transition equation: According to the physical model of water conservancy engineering and the dynamic characteristics of water delivery system, the state transition equation is established to describe the changes in the state space after executing a certain control action. The state transition equation takes into account factors such as the continuity equation of water flow, the energy equation, and the control characteristics of the pump station and the gate.

[0126] Reward function definition: The reward function is designed according to the objective function. When energy consumption is reduced and supply-demand matching degree is improved, positive rewards are given. When energy consumption is increased or supply-demand matching degree is reduced, negative rewards are given.

[0127] (2) Simulate different control actions;

[0128] The PPO reinforcement learning algorithm is used to simulate different control actions in the state space, the reward value of each action is calculated according to the state transition equation and the reward function, and the control action that can minimize the objective function (i.e., maximize the reward value) is selected as the optimal water delivery control instruction.

[0129] (Three) adjust the gate opening and the start-stop state of the pump station;

[0130] According to the generated water delivery control instruction, the gate opening and the start-stop state of the pump station are adjusted in real time to realize accurate control of the water delivery quantity of the water conservancy project to meet the downstream water demand while minimizing energy consumption.

[0131] The beneficial effects are that, by acquiring multi-source real-time data in the water conservancy project, the multi-source real-time data is preprocessed to obtain initial multi-source real-time data; based on the CNN convolutional neural network, local features in the input data are extracted, the LSTM neural network is used to process long-term dependencies of time series, and the Attention mechanism is used to focus on the water delivery variables to obtain a CNN-LSTM-Attention demand prediction model; the initial multi-source real-time data is input into the CNN-LSTM-Attention demand prediction model for prediction to obtain a water delivery demand curve; according to the water delivery demand curve, the PPO reinforcement learning algorithm is used to set the objective function as minimizing energy consumption and maximizing supply-demand matching degree, different control actions are simulated to obtain water delivery control instructions, and the gate opening and the start-stop state of the pump station are adjusted according to the water delivery control instructions. 1. Compared with a single model or a traditional prediction method, the accuracy of water delivery demand prediction is significantly improved, the relationship between complex factors in the water conservancy project and water delivery demand can be more accurately captured, reliable prediction basis is provided for water delivery control, and unreasonable allocation of water resources caused by prediction deviation is reduced. 2. The influence of various operations on energy consumption and supply-demand matching degree can be comprehensively evaluated to generate optimal water delivery control instructions. Compared with the traditional water delivery control method, this scheme can dynamically adjust the gate opening and the start-stop state of the pump station according to real-time data and prediction results, meet the downstream water demand while effectively reducing energy consumption, realize efficient use of water resources and optimization of water conservancy project operation cost, and improve the economic benefit of the water conservancy project.

[0132] Please refer to Figure 2 In a water conservancy project water delivery quantity adjustment method based on artificial intelligence, point cloud data and unmanned aerial vehicle image data are registered and modeled by using an improved NeRF neural radiation field framework to generate a geographic real scene three-dimensional model, including the following steps:

[0133] Step 201, acquiring multi-source real-time data in the water conservancy project, at least including rainfall, evaporation, reservoir water level, water delivery pipeline pressure and downstream water demand;

[0134] Step 202, combining Z-score method and IQR method to detect outliers of multi-source real-time data, using linear interpolation method to fill in missing values, obtaining first multi-source real-time data;

[0135] Step 203, using Min-Max normalization method to map the first multi-source real-time data to the interval [0, 1], and converting the collected data of different formats into numerical data, obtaining second multi-source real-time data;

[0136] Step 204, taking the time of the data center server as the reference, processing the second multi-source real-time data for time synchronization, and according to the geographic information of the water conservancy project and the position coordinates of each monitoring point, processing the data for spatial alignment, obtaining initial multi-source real-time data.

[0137] The above describes the embodiment of the water conservancy project water delivery capacity adjusting method based on artificial intelligence, please refer to Figure 3 In a water conservancy project water delivery capacity adjusting system based on artificial intelligence, the water conservancy project water delivery capacity adjusting system comprises the following modules:

[0138] A real-time data acquisition module is configured to acquire multi-source real-time data in a water conservancy project, perform data preprocessing on the multi-source real-time data, and obtain initial multi-source real-time data.

[0139] A prediction model establishing module is configured to extract local features in input data based on a CNN convolutional neural network, process long-term dependencies of time series through an LSTM neural network, and focus on water delivery variables using an Attention mechanism to obtain a CNN-LSTM-Attention demand prediction model.

[0140] A water delivery demand prediction module is configured to input the initial multi-source real-time data into the CNN-LSTM-Attention demand prediction model for prediction to obtain a water delivery demand curve.

[0141] A water delivery control and adjustment module is configured to set a target function as minimizing energy consumption and maximizing supply-demand matching degree according to the water delivery demand curve through a PPO reinforcement learning algorithm, obtain water delivery control instructions by simulating different control actions, and adjust gate opening and pump station start-stop states according to the water delivery control instructions.

[0142] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A water delivery regulation method for a water conservancy project based on artificial intelligence, characterized in that: The water transfer rate regulation method for a water conservancy project comprises the following steps: Acquiring multi-source real-time data from a water conservancy project, and performing data preprocessing on the multi-source real-time data to obtain initial multi-source real-time data; The first convolution layer based on the CNN convolutional neural network uses 64 convolution kernels to extract local features of the input multi-source time series data; the second convolution layer uses 128 convolution kernels to further extract complex local features; the third convolution layer uses 256 convolution kernels to capture local information, and the step size of the convolution layer is set to 1, and the filling method is the same; the first LSTM layer of the LSTM neural network receives the local feature data output by the CNN convolutional neural network and converts it into a time series data format; the second LSTM layer is used to further process the output of the first LSTM layer to enhance the model's capture of long-term dependencies in the time series. The dropout rate of the LSTM layer is set to 0.2, and the return sequence is set to True. The hidden state output by the LSTM neural network is used as the query, key, and value. The attention score is obtained by calculating the dot product of the query and the key and dividing it by the scaling factor. The attention score is normalized using the Softmax function to obtain the attention weight of each input element. Finally, the value is weighted and summed according to the attention weight. The importance between water supply demand and each input variable is automatically learned through the attention mechanism, resulting in a CNN-LSTM-Attention demand prediction model. Inputting the initial multi-source real-time data into the CNN-LSTM-Attention demand forecasting model for prediction to obtain a water delivery demand curve; According to the water delivery demand curve, the objective function is set to minimize energy consumption and maximize supply-demand matching through the PPO reinforcement learning algorithm. By simulating different control actions, water delivery control instructions are obtained, and the gate opening and the start and stop status of the pump station are adjusted according to the water delivery control instructions.

2. The artificial intelligence-based water supply regulation method for a water conservancy project according to claim 1, characterized in that: The step of acquiring multi-source real-time data from a water conservancy project and performing data preprocessing on the multi-source real-time data to obtain initial multi-source real-time data includes: Obtain multi-source real-time data on water conservancy projects, including at least rainfall, evaporation, reservoir water levels, water pipeline pressure, and downstream water demand; Combining the Z-score method and the IQR method to perform outlier detection on the multi-source real-time data, and using a linear interpolation method to fill in missing values, to obtain first multi-source real-time data; Mapping the first multi-source real-time data to the interval [0, 1] using a Min-Max normalization method, and uniformly converting the collected data in different formats into numerical data to obtain second multi-source real-time data; The second multi-source real-time data is time-synchronized based on the time of the data center server, and the data is spatially aligned based on the geographic information of the water conservancy project and the location coordinates of each monitoring point to obtain the initial multi-source real-time data.

3. The method for regulating water delivery of a water conservancy project based on artificial intelligence according to claim 1, characterized in that: According to the water delivery demand curve, the objective function is set to minimize energy consumption and maximize supply-demand matching through the PPO reinforcement learning algorithm, and the water delivery control instructions are obtained by simulating different control actions, including: Define the state space and state space, and establish the state transfer equation according to the physical model of the water conservancy project and the dynamic characteristics of the water delivery system; Design an objective function that gives positive rewards when energy consumption decreases and the supply-demand match improves; and gives negative rewards when energy consumption increases or the supply-demand match decreases.

4. The artificial intelligence-based water supply regulation method for a water conservancy project according to claim 3, characterized in that: The objective function is set to minimize energy consumption and maximize supply-demand matching through the PPO reinforcement learning algorithm according to the water delivery demand curve, and water delivery control instructions are obtained by simulating different control actions, further comprising: The PPO reinforcement learning algorithm is used to simulate different control actions in the state space, and the reward value of each action is calculated based on the state transition equation and reward function; The control action that minimizes the objective function is selected as the optimal water delivery control instruction.

5. A water delivery regulation system for water conservancy projects based on artificial intelligence, characterized in that: The water delivery regulation system of the water conservancy project includes the following modules: A real-time data acquisition module is used to acquire multi-source real-time data in a water conservancy project, and perform data preprocessing on the multi-source real-time data to obtain initial multi-source real-time data; The prediction model establishment module is used to extract local features of the input multi-source time series data using 64 convolution kernels in the first convolution layer of the CNN convolutional neural network; the second convolution layer uses 128 convolution kernels to further extract complex local features; the third convolution layer uses 256 convolution kernels to capture local information, and the step size of the convolution layer is set to 1, and the padding method is the same; the first LSTM layer of the LSTM neural network receives the local feature data output by the CNN convolutional neural network and converts it into a time series data format; the second LSTM layer is used to further process the output of the first LSTM layer to enhance the model's ability to capture long-term dependencies in time series. The dropout rate of the LSTM layer is set to 0.2, and the return sequence is set to True; the hidden state output by the LSTM neural network is used as the query, key, and value, and the attention score is obtained by calculating the dot product of the query and the key and dividing it by the scaling factor; The attention scores are normalized using the Softmax function to obtain the attention weights of each input element. Finally, the values ​​are weighted and summed according to the attention weights. The Attention mechanism automatically learns the importance between water supply demand and each input variable, resulting in a CNN-LSTM-Attention demand prediction model. A water transmission demand prediction module is used to input the initial multi-source real-time data into the CNN-LSTM-Attention demand prediction model to perform prediction and obtain a water transmission demand curve; The water delivery control and adjustment module is used to set the objective function as minimizing energy consumption and maximizing supply-demand matching according to the water delivery demand curve through the PPO reinforcement learning algorithm, obtain water delivery control instructions by simulating different control actions, and adjust the gate opening and the start and stop status of the pump station according to the water delivery control instructions.

6. The artificial intelligence-based water conservancy project water delivery regulation system according to claim 5, characterized in that: The water delivery control and regulation module includes the following submodules: Definition submodule, used to define the state space and state space, and establish the state transfer equation according to the physical model of the water conservancy project and the dynamic characteristics of the water delivery system; The judgment submodule is used to design the objective function and provide positive rewards when energy consumption is reduced and the supply-demand matching is improved; Negative rewards are given when energy consumption increases or the matching between supply and demand decreases.

7. The artificial intelligence-based water conservancy project water delivery regulation system according to claim 5, characterized in that: The water delivery control and regulation module includes the following submodules: The simulation submodule is used to simulate different control actions in the state space through the PPO reinforcement learning algorithm and calculate the reward value of each action according to the state transition equation and reward function; The selection submodule is used to select the control action that minimizes the objective function as the optimal water delivery control instruction.

Images