Reservoir group saltwater pressure fresh water supplement random optimization scheduling method and system under multiple uncertainties

Abstract

The application provides a reservoir group saltwater compression fresh water random optimization scheduling method and system under multiple uncertainties, relates to the technical field of water resource scheduling, quantifies runoff prediction uncertainty through a Gaussian mixture model and Bayesian inference, quantifies salt tide upstream uncertainty through a deep graph neural network model, Bayesian inference and kernel density estimation, realizes quantification of double heterogenous uncertainty, takes multiple uncertainties as multi-source random input, constructs a random optimization scheduling model that propagates uncertainty to scheduling target probability distribution, improves adaptability of the scheduling scheme, obtains a Pareto solution set through a deep learning agent model and dimension reduction technology, reduces calculation complexity and avoids local optimum, forms a coordinated scheduling decision scheme through generation of a power characteristic curve that maps social benefit and economic benefit, and realizes multi-objective risk benefit balance under multiple uncertainties.

Description

Technical Field

[0001] This invention relates to the field of water resource scheduling technology, and more specifically, to a stochastic optimization scheduling method and system for reservoir group salinity suppression and freshwater replenishment under multiple uncertainties. Background Technology

[0002] With the development of water resource allocation technology, the joint operation of river basin reservoir groups has become an important means of ensuring water supply security in estuary areas. Especially in sections of rivers where saltwater intrusion is frequent and affected by ocean tides, the joint operation of upstream reservoir groups increases river runoff during the dry season to suppress saltwater intrusion and replenish freshwater. This "saltwater suppression and freshwater replenishment" operation is an effective way to deal with saltwater intrusion disasters.

[0003] In existing technologies, the salinity suppression and freshwater replenishment scheduling of reservoir groups is typically based on upstream runoff forecasts, using pre-set empirical rules or deterministic optimization models to formulate water replenishment plans for each reservoir. For example, by predicting upstream inflow over a future period, the required downstream discharge flow is calculated based on the chloride content control target at the downstream intake and allocated to each reservoir for execution. However, the above scheduling method has its limitations. The upstream runoff forecasts themselves are uncertain. Furthermore, the upstream salinity intrusion process in the downstream estuary is also influenced by a combination of factors, including ocean tides and meteorological factors, resulting in complex patterns and significant forecast uncertainties. Additionally, existing deterministic scheduling models struggle to effectively characterize and address the dual uncertainties of runoff and tides. This can lead to the generated scheduling plans failing in practical applications due to forecast biases, affecting the salinity suppression and freshwater replenishment effect and failing to meet the high-precision water supply security requirements of sensitive estuarine areas. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide a random optimization scheduling method and system for salinity suppression and freshwater replenishment of reservoir groups under multiple uncertainties. This method can increase the probability of freshwater intake at the downstream estuary while ensuring the ecological flow of the river channel. At the same time, it can reduce the impact of uncertainties in runoff and tidal forecasts on scheduling efficiency through effective scheduling methods, and avoid the risks brought about by uncertainties.

[0005] Firstly, this application provides a stochastic optimization scheduling method for salinity suppression and freshwater replenishment of reservoir groups under multiple uncertainties, including:

[0006] Acquire runoff forecast data, measured runoff data, and saltwater intrusion influencing factors for the target area;

[0007] A Gaussian mixture model was used to fit the joint distribution of runoff forecast data and measured runoff data, and the conditional probability density function of measured runoff was obtained through Bayesian inference. A deep graph neural network model was used to extract the spatiotemporal characteristics of the factors influencing the upstream intrusion of saltwater, and the probability density function of estuarine chloride content forecast was obtained through Bayesian inference and kernel density estimation.

[0008] A multi-objective stochastic optimization scheduling model is constructed based on the measured conditional probability density function of runoff and the predicted probability density function of estuarine chloride content.

[0009] Based on the measured runoff conditional probability density function and the predicted probability density function of estuary chloride content, a multi-objective stochastic optimization scheduling model is used to iteratively solve for the Pareto solution set;

[0010] Based on the Pareto solution set, the dynamic characteristic curve of the mapping relationship between social benefits and economic benefits is obtained, and a collaborative scheduling decision scheme under multiple uncertainties is generated according to the dynamic characteristic curve.

[0011] Optionally, a Gaussian mixture model is used to fit the joint distribution of the predicted runoff data and the measured runoff data, and the conditional probability density function of the measured runoff is obtained through Bayesian inference, including:

[0012] A joint distribution dataset is obtained based on runoff forecast data and measured runoff data, and the joint distribution dataset is split into forecast runoff sub-datasets and measured runoff sub-datasets.

[0013] The probability distribution of the joint distribution dataset is fitted using a Gaussian mixture model to obtain the mean vector and covariance matrix of each Gaussian component;

[0014] Based on the mean vector and covariance matrix of each Gaussian component, a joint probability density function of runoff forecast data and measured runoff data is constructed.

[0015] Based on full probability reasoning, the predicted runoff subset is processed to obtain the marginal probability density function;

[0016] Based on the joint probability density function and the marginal probability density function, Bayesian inference is used to process the measured runoff sub-data to obtain the measured runoff conditional probability density function.

[0017] Optionally, the deep graph neural network model includes a meteorological information extraction module and a time series analysis module; the meteorological information extraction module is used to extract the influence characteristics of tidal fluctuations, rainfall and wind speed factors on estuary chloride content through convolutional layers; the time series analysis module is used to establish the nonlinear time series relationship between the influence characteristics and hydrological characteristics of key stations and estuary chloride content through a gated cyclic network, and output the estuary chloride content forecast value.

[0018] Optionally, a deep graphical neural network model is used to extract the spatiotemporal characteristics of factors influencing saltwater intrusion, and the probability density function for predicting estuarine chloride content is obtained through Bayesian inference and kernel density estimation, including:

[0019] A variational inference strategy is adopted to approximate the true posterior probability distribution of the parameters of the deep graph neural network model with the variational distribution. The deep graph neural network model is trained by maximizing the lower bound of evidence or minimizing the relative entropy between the variational distribution and the posterior distribution.

[0020] The probability density function for predicting estuarine chloride content is obtained by processing the output of the depth map neural network model through kernel density estimation.

[0021] Optionally, based on the measured runoff conditional probability density function and the predicted probability density function of estuarine chloride content, a multi-objective stochastic optimization scheduling model is constructed, including:

[0022] Using the measured runoff conditional probability density function and the predicted estuarine chloride content probability density function as multi-source random inputs, a pre-set scheduling mode is used for simulation, and multiple scheduling objectives are used as multi-dimensional probabilistic objectives to construct a multi-objective stochastic optimization scheduling model. The pre-set scheduling mode includes at least one of the following: flow control mode, water level control mode, or rule-based scheduling mode. The multiple scheduling objectives include salinity suppression and freshwater replenishment objectives, power generation objectives, and ecological objectives.

[0023] Optionally, based on the measured runoff conditional probability density function and the predicted probability density function of estuarine chloride content, a multi-objective stochastic optimization scheduling model is constructed, which also includes:

[0024] The Markov chain Monte Carlo method is used to simulate multi-source random input and generate multiple random scenarios.

[0025] For each random scenario, a preset scheduling mode is used for simulation to obtain random simulation results of the reservoir outflow process, reservoir water storage state, and river flow process;

[0026] The random simulation results are input into a deep graph neural network model to simulate the random process of chlorine content in estuaries;

[0027] Based on the stochastic simulation results and the stochastic process of estuary chlorine content, nonparametric kernel density estimation is used to quantify the probability density functions of reservoir water level, reservoir discharge, reservoir output, river flow, and estuary chlorine content as probabilistic feature data for scheduling response.

[0028] Optionally, based on the measured runoff conditional probability density function and the predicted probability density function of estuarine chloride content, a multi-objective stochastic optimization scheduling model is used for iterative solution to obtain the Pareto solution set, which includes:

[0029] A deep learning proxy model is used to approximate the mapping relationship between multi-source random inputs and the probability distribution of each scheduling objective in the multi-objective stochastic optimization scheduling model.

[0030] In the population update process of evolutionary algorithms, multi-scale dimensionality reduction techniques are used to select new populations;

[0031] The optimal Pareto solution set is obtained by non-dominated sorting and crowding distance calculation.

[0032] Secondly, this application provides a stochastic optimization scheduling system for salinity suppression and freshwater replenishment in a reservoir group under multiple uncertainties, including:

[0033] The data acquisition module is used to acquire runoff forecast data, measured runoff data, and saltwater intrusion influencing factors for the target area.

[0034] The data quantization module is used to fit the joint distribution of runoff forecast data and measured runoff data using a Gaussian mixture model, and obtain the measured runoff conditional probability density function through Bayesian inference; a deep graph neural network model is used to extract the spatiotemporal characteristics of the influencing factors of saltwater intrusion, and obtain the estuarine chloride content forecast probability density function through Bayesian inference and kernel density estimation.

[0035] A model building module is used to construct a multi-objective stochastic optimization scheduling model based on the measured runoff conditional probability density function and the predicted probability density function of estuary chloride content.

[0036] The data solving module is used to iteratively solve for the Pareto solution set using a multi-objective stochastic optimization scheduling model based on the measured runoff conditional probability density function and the predicted probability density function of estuary chloride content.

[0037] The strategy analysis module is used to obtain the dynamic characteristic curve of the mapping relationship between social benefits and economic benefits based on the Pareto solution set, and to generate a collaborative scheduling decision scheme under multiple uncertainties based on the dynamic characteristic curve.

[0038] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-mentioned stochastic optimization scheduling method for salinity suppression and freshwater replenishment of reservoir groups under multiple uncertainties.

[0039] Fourthly, this application provides a computer-readable storage medium storing computer instructions, which, when executed by a processor, implement the aforementioned stochastic optimization scheduling method for salinity control and freshwater replenishment in reservoir groups under multiple uncertainties.

[0040] This invention provides a stochastic optimization scheduling method and system for reservoir group salinity suppression and freshwater replenishment under multiple uncertainties. It employs a Gaussian mixture model to fit the joint distribution of predicted and measured runoff data, and combines this with Bayesian inference to obtain the measured runoff conditional probability density function. Simultaneously, a deep graph neural network model is used to extract the spatiotemporal characteristics of salinity intrusion influencing factors, and this is combined with Bayesian inference and kernel density estimation to obtain the estuary chloride content prediction probability density function. This achieves dual quantification of runoff forecast uncertainty and salinity intrusion uncertainty, overcoming the limitation of traditional methods in handling two heterogeneous uncertainties simultaneously, and providing more reliable input information for subsequent scheduling decisions. By using the measured runoff conditional probability density function and the estuary chloride content prediction probability density function as multi-source random inputs, and simulating using a preset scheduling mode with multiple scheduling objectives as multi-dimensional probability objectives, a multi-objective stochastic optimization scheduling model is constructed to propagate uncertainty to the probability distribution of scheduling objectives. This allows upstream runoff uncertainty and downstream salinity intrusion uncertainty to be chained together. The probability distributions passed to the objectives of salinity suppression and freshwater replenishment, power generation, and ecological protection solve the technical challenge of existing empirical rule-based salinity suppression and freshwater replenishment methods being unable to cope with short-term, sudden salinity intrusions. This significantly improves the adaptability of the scheduling scheme in complex hydrological environments and the accuracy of water supply security. By using a deep learning surrogate model to replace the multi-objective stochastic optimization scheduling model and employing dimensionality reduction techniques to reduce the dimensionality of the optimization population, Pareto solutions are obtained iteratively. This effectively reduces the computational complexity in high-dimensional decision spaces and overcomes the dimensionality curse problem of traditional stochastic optimization algorithms when dealing with reservoir group scheduling, avoiding the algorithm getting trapped in local optima. By generating a dynamic characteristic curve mapping the social and economic benefits based on the trade-offs between different objectives in the Pareto solution set, and generating a collaborative scheduling decision scheme under multiple uncertainties based on this dynamic characteristic curve, decision-makers are provided with an intuitive visualization of multi-objective trade-offs, achieving a risk-benefit balance between salinity suppression benefits, power generation benefits, and ecological benefits under multiple uncertainties of runoff and tides.

[0041] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0042] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1The flowchart of a stochastic optimization scheduling method for salinity suppression and freshwater replenishment of a reservoir group under multiple uncertainties provided by an embodiment of the present invention is shown.

[0044] Figure 2 A schematic diagram of the structure of the depth graph neural network model provided in an embodiment of the present invention is shown;

[0045] Figure 3 This diagram illustrates the structure of the runoff, tide, and meteorological chain transmission diagram provided in an embodiment of the present invention.

[0046] Figure 4 The diagram illustrates the process of constructing a multi-objective stochastic optimization scheduling model that propagates uncertainty to the probability distribution of scheduling objectives, as provided in an embodiment of the present invention.

[0047] Figure 5 The diagram illustrates the process of using dimensionality reduction techniques to reduce the dimensionality of an optimized population and iteratively solving for the Pareto solution set, as provided in an embodiment of the present invention.

[0048] Figure 6 This invention provides a schematic diagram of a stochastic optimization scheduling system for salinity suppression and freshwater replenishment in a reservoir group under multiple uncertainties, as shown in an embodiment of the present invention.

[0049] Figure 7 A schematic diagram of the structure of an electronic device provided in an embodiment of the present invention is shown. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0051] This application provides a stochastic optimization scheduling method for salinity suppression and freshwater replenishment in a reservoir group under multiple uncertainties. (See also...) Figure 1 As shown, the stochastic optimization scheduling method for salinity suppression and freshwater replenishment of reservoir groups under multiple uncertainties provided in this application includes at least the following steps:

[0052] Step 110: Obtain runoff forecast data, measured runoff data, and saltwater intrusion influencing factors for the target area.

[0053] In this embodiment, runoff forecast data refers to the predicted river runoff in the target area over a future period using meteorological and hydrological models or statistical methods; measured runoff data refers to historical runoff values ​​actually observed by monitoring equipment such as hydrological stations; and saltwater intrusion influencing factors include flow data and water level data from watershed control hydrological stations, tidal range data and tidal level data from estuary tidal stations, rainfall data from regional meteorological stations, and sea level data. By acquiring runoff forecast data, measured runoff data, and saltwater intrusion influencing factors for the target area, the characteristics of upstream runoff changes and the potential impact conditions of downstream saltwater intrusion can be comprehensively reflected. This provides data support for quantifying the uncertainty of runoff forecasts and the uncertainty of saltwater intrusion. Simultaneously, it effectively distinguishes the deviation patterns between forecast and measured runoff values ​​and identifies key environmental factors affecting saltwater intrusion, thereby improving the adaptability of the entire scheduling method to complex hydrological environments.

[0054] Step 120: Use a Gaussian mixture model to fit the joint distribution of runoff forecast data and measured runoff data, and obtain the measured runoff conditional probability density function through Bayesian inference; use a deep graph neural network model to extract the spatiotemporal characteristics of the influencing factors of saltwater intrusion, and obtain the estuary chloride content forecast probability density function through Bayesian inference and kernel density estimation.

[0055] In this embodiment, a Gaussian mixture model is used to fit the joint distribution of predicted runoff data and measured runoff data, and Bayesian inference is used to obtain the conditional probability density function of the measured runoff. This includes: obtaining a joint distribution dataset based on the predicted runoff data and measured runoff data, and splitting the joint distribution dataset into predicted runoff sub-datasets and measured runoff sub-datasets; fitting the probability distribution of the joint distribution dataset using a Gaussian mixture model to obtain the mean vector and covariance matrix of each Gaussian component; constructing a joint probability density function of the predicted runoff data and measured runoff data based on the mean vector and covariance matrix of each Gaussian component; processing the predicted runoff sub-dataset based on full probability inference to obtain a marginal probability density function; and processing the measured runoff sub-dataset using Bayesian inference based on the joint probability density function and the marginal probability density function to obtain the conditional probability density function of the measured runoff.

[0056] In this embodiment of the application, the Gaussian mixture model is used to fit the joint distribution of runoff forecast data and measured runoff data, and the specific process of obtaining the conditional probability density function of measured runoff through Bayesian inference is as follows:

[0057] First, based on the obtained runoff forecast data and measured runoff data, a joint distribution dataset is constructed:

[0058]

[0059] In the formula, For a jointly distributed dataset, These are the model parameters for the Gaussian mixture model. This represents the total number of Gaussian components in the Gaussian mixture model. This represents the index number of the Gaussian components in the Gaussian mixture model. The mixing coefficients of the Gaussian mixture model need to satisfy the following conditions: and ;

[0060] In the joint distribution dataset, each pair of data points consists of the predicted runoff value and the measured runoff value corresponding to the same time, in order to determine the intrinsic relationship between the predicted value and the measured value;

[0061] Secondly, the joint distribution dataset is split into predicted runoff subsets and measured runoff subsets to facilitate subsequent processing of the probabilistic features of the predicted and measured sides separately, i.e., the joint distribution dataset... Split into two subsets ,in, For forecasting runoff subsets, This is a subset of measured runoff data (measured runoff at the estuary).

[0062] Then, a Gaussian mixture model is used to fit the overall probability distribution of the joint distribution dataset. Specifically, the Gaussian mixture model approximates an arbitrarily complex probability distribution by weighted sum of multiple Gaussian components. Each Gaussian component has its own mean vector and covariance matrix, i.e.:

[0063]

[0064]

[0065] In the formula, For the first The mean vector of Gaussian components, For the first The first component in the mean vector of Gaussian components. For the first The second component in the mean vector of Gaussian components. For the first The covariance matrix of Gaussian components, For the first The elements in the first row and first column of the covariance matrix of the Gaussian components. For the first The elements in the first row and second column of the covariance matrix of the Gaussian components. For the first The elements in the second row and first column of the covariance matrix of the Gaussian components. For the first The elements in the second row and second column of the covariance matrix of the Gaussian components;

[0066] For example, based on the properties of the joint normal distribution, given the forecast factor, the conditional probability of the forecast value follows a joint normal distribution. The conditional mean vector and covariance matrix of the chlorine content are as follows:

[0067]

[0068]

[0069] Through the fitting process, the mean vector and covariance matrix of each Gaussian component can be obtained. The parameters of the mean vector and covariance matrix quantitatively describe the central location, dispersion and correlation of the predicted runoff and the measured runoff in the joint distribution dataset under different distribution patterns.

[0070] Secondly, based on the mean vector and covariance matrix of each Gaussian component, a joint probability density function of the runoff forecast data and the measured runoff data is constructed. That is, according to the multiplication formula of the joint Gaussian distribution, the joint probability density function of the Gaussian mixture model can be derived as follows:

[0071]

[0072] In the formula, The conditional mean parameter is the parameter at the th... Given runoff forecast data, the conditional expectation of measured runoff data under a given Gaussian component. Let be the conditional variance parameter, that is, the _th _ Conditional variance of measured runoff data given runoff forecast data under Gaussian components;

[0073] The joint probability density function is expressed in the form of a Gaussian mixture model, which can fully characterize the probability of predicted runoff and measured runoff under various combinations of values, thereby transforming the statistical regularity in the original data into a mathematical form that can be analyzed and calculated.

[0074] Then, based on full probability inference, the predicted runoff subset is processed to obtain the marginal probability density function of the predicted runoff data, namely:

[0075]

[0076] The marginal probability density function is obtained by integrating the measured runoff variables in the joint probability density function, so that the predicted runoff data itself follows a probability distribution without considering the specific values ​​of the measured runoff.

[0077] Finally, based on the obtained joint probability density function and marginal probability density function, Bayesian inference is used to process the measured runoff sub-data to obtain the measured runoff conditional probability density function, namely:

[0078]

[0079] Bayesian inference calculates the conditional probability distribution of measured runoff values ​​given a forecasted runoff value by dividing the joint probability density function by the marginal probability density function. This conditional probability density function of measured runoff quantifies the uncertainty of runoff forecasting, that is, the probability distribution that the actual runoff volume may deviate from the forecasted value when a certain runoff forecast value is obtained.

[0080] Through the above process, the statistical relationship between runoff forecast data and measured runoff data is transformed into a usable conditional probability density function, which provides a quantitative mathematical basis for introducing runoff uncertainty as one of the multi-source random inputs into the scheduling model, so that the scheduling decision can fully consider the risks brought about by runoff forecast errors.

[0081] In this embodiment, the deep graph neural network model includes a meteorological information extraction module and a time series analysis module. The meteorological information extraction module is used to extract the influence characteristics of tidal fluctuations, rainfall and wind speed factors on the estuary chloride content through convolutional layers. The time series analysis module is used to establish the nonlinear time series relationship between the influence characteristics and the hydrological characteristics of key stations and the estuary chloride content through a gated cyclic network, and output the estuary chloride content forecast value.

[0082] Furthermore, a deep graph neural network model is used to extract the spatiotemporal characteristics of factors influencing saltwater intrusion, and the probability density function for predicting estuarine chloride content is obtained through Bayesian inference and kernel density estimation. This includes: using a variational inference strategy to approximate the true posterior probability distribution of the parameters of the deep graph neural network model with a variational distribution; training the deep graph neural network model by maximizing the lower bound of evidence or minimizing the relative entropy between the variational distribution and the posterior distribution; and processing the output of the deep graph neural network model through kernel density estimation to obtain the probability density function for predicting estuarine chloride content.

[0083] In this embodiment of the application, the spatiotemporal characteristics of the influencing factors of saltwater intrusion are extracted using a deep graph neural network model, and the specific process of obtaining the probability density function of estuary chloride content forecast through Bayesian inference and kernel density estimation is as follows:

[0084] First, construct a deep graph neural network model, such as Figure 2As shown, the deep graph neural network model includes a meteorological information extraction module and a time series analysis module. The meteorological information extraction module is used to extract the influence features of tidal fluctuations, rainfall, and wind speed factors on estuary chloride content through convolutional layers. Specifically, the convolutional layers extract local features from the input meteorological data using a sliding window approach, effectively capturing the periodic patterns of tidal fluctuations, the intensity variations of rainfall events, and the spatiotemporal distribution patterns of wind speed factors, transforming the high-dimensional raw meteorological data into a compact, physically meaningful influence feature vector. The time series analysis module is used to establish a nonlinear temporal relationship between the aforementioned influence features and key station hydrological features and estuary chloride content through a gated recurrent network. The gated recurrent network, through its internal update and reset gate mechanisms, can effectively learn long-term dependencies and avoid the gradient vanishing problem. The time series analysis module takes the influence features output by the meteorological information extraction module and the hydrological features of key stations as input, and through the sequence modeling capability of the gated recurrent network, establishes a mapping relationship from historical moments to future moments, ultimately outputting the estuary chloride content forecast value.

[0085] Then, a variational inference strategy is used to train the deep graph neural network model. Specifically, the network weight parameters of the deep graph neural network model are treated as random variables, and a variational distribution is used to approximate the true posterior probability distribution of the model parameters. The true posterior probability distribution represents the probability distribution of the model parameters after observing the training data. However, due to the high dimensionality of the parameter space of the deep graph neural network, the true posterior distribution is difficult to calculate directly analytically. Therefore, a variational distribution with a relatively simple form is introduced to approximate it. The deep graph neural network model is trained by maximizing the lower bound of evidence or minimizing the relative entropy between the variational distribution and the posterior distribution. The lower bound of evidence is a lower bound function of the logarithmic marginal likelihood of the observed data. Maximizing the lower bound of evidence is equivalent to making the variational distribution as close as possible to the true posterior distribution while fitting the training data. The relative entropy, also known as KL divergence, is used to measure the degree of difference between two probability distributions. Minimizing the relative entropy makes the variational distribution approximate the true posterior distribution. After training, when given the influencing factor of saltwater intrusion as input, the deep graph neural network model outputs a predicted estuarine chloride content value that follows a distribution induced by parameter uncertainty, rather than a single deterministic value.

[0086] Finally, the output of the deep graph neural network model is processed by kernel density estimation to obtain the estuarine chloride prediction probability density function. Specifically, the deep graph neural network model is run multiple times within a Bayesian framework. Each time, a set of model parameters is sampled from the posterior distribution of the parameters to obtain the corresponding estuarine chloride prediction values, thus obtaining a set of prediction samples. Kernel density estimation places a kernel function centered on each prediction sample point, and all kernel functions are superimposed and normalized to obtain a continuous probability density curve that does not require assumptions about a specific distribution pattern. This probability density curve is the estuarine chloride prediction probability density function, used to describe the probability of the estuarine chloride taking each possible value at a future time.

[0087] further, Figure 3 The diagram illustrates the structure of the runoff, tide, and meteorological chain transmission graph, namely the superposition effect and transmission path of runoff pulsation, tidal phase, and meteorological factors on the upstream intrusion of saltwater, providing a theoretical basis for the spatial topology construction of the deep graph neural network model.

[0088] Through the above process, the spatiotemporal feature extraction capability of the deep graph neural network model, the parameter uncertainty quantification capability of Bayesian inference, and the nonparametric probability density estimation capability of kernel density estimation are organically combined to finally output the estuary chloride content prediction probability density function. The estuary chloride content prediction probability density function is used to fully characterize the uncertainty of saltwater intrusion, providing probabilistic downstream boundary conditions for subsequent scheduling models, so that the saltwater suppression and freshwater replenishment scheduling decision can be scientifically quantified and the risk of saltwater intrusion can be addressed.

[0089] Step 130: Based on the measured runoff conditional probability density function and the predicted estuary chloride content probability density function, construct a multi-objective stochastic optimization scheduling model.

[0090] In this embodiment of the application, a multi-objective stochastic optimization scheduling model is constructed based on the measured runoff conditional probability density function and the predicted estuary chloride content probability density function. This includes: using the measured runoff conditional probability density function and the predicted estuary chloride content probability density function as multi-source random inputs, simulating the process using a preset scheduling mode, and constructing a multi-objective stochastic optimization scheduling model using multiple scheduling objectives as multi-dimensional probabilistic objectives. The preset scheduling mode includes at least one of a flow control mode, a water level control mode, or a rule-based scheduling mode; the multiple scheduling objectives include salinity control and freshwater replenishment objectives, power generation objectives, and ecological objectives.

[0091] Furthermore, the construction of a multi-objective stochastic optimization scheduling model includes: using the Markov chain Monte Carlo method to simulate multi-source random inputs and generate multiple random scenarios; for each random scenario, simulating using a preset scheduling mode to obtain stochastic simulation results of the reservoir outflow process, reservoir water storage state, and river flow process; inputting the stochastic simulation results into a deep graph neural network model to simulate the stochastic process of estuary chloride content; and based on the stochastic simulation results and the stochastic process of estuary chloride content, using nonparametric kernel density estimation to quantify the probability density functions of reservoir water level, reservoir outflow, reservoir output, river flow, and estuary chloride content as scheduling response probability feature data.

[0092] In this embodiment, the multi-source random input includes two aspects: upstream runoff uncertainty and downstream salinity intrusion uncertainty. Upstream runoff uncertainty is quantitatively determined by the measured runoff conditional probability density function, reflecting the probability distribution of possible values ​​of actual inflow runoff under a given runoff forecast. Downstream salinity intrusion uncertainty is quantitatively determined by the estuary chloride content forecast probability density function, reflecting the probability distribution of possible values ​​of estuary chloride content under a given salinity intrusion upstream influence factor. The preset scheduling mode includes at least one of a flow control mode, a water level control mode, or a rule-based scheduling mode. The flow control mode uses the reservoir discharge flow as the scheduling control variable, the water level control mode uses the reservoir water level as the scheduling control variable, and the rule-based scheduling mode uses a pre-set scheduling diagram or scheduling function as the decision-making basis. Multiple scheduling objectives include salinity reduction and freshwater replenishment objectives, power generation objectives, and ecological objectives. The salinity reduction and freshwater replenishment objective aims to increase the expected value of freshwater access time at the downstream estuary, i.e., based on the estuary chloride content probability density function. And take the critical value of dilute chlorine content. Calculate the expected value of the available light time. The goal is to maximize the expected value of the available fresh water time as the target for saltwater suppression and replenishment.

[0093]

[0094]

[0095] In the formula, This refers to the number of time periods for saltwater intrusion forecasts; This refers to the length of the saltwater intrusion forecast period; The expected value of the freshwater time within the t-th time period is calculated by taking the product of the definite integral of the estuary chloride content probability density function and the time period length.

[0096] The power generation objective is to maximize the power generation benefits of the watershed reservoir group, while the ecological objective aims to minimize the probability that the river flow falls below the ecological minimum flow, i.e., the ecological damage rate, which is determined by the reservoir output probability density function. , probability density function of reservoir's final capacity The goal is to calculate the sum of hydropower generation during the entire scheduling period and the energy storage at the end of the period, maximizing power generation efficiency.

[0097]

[0098]

[0099]

[0100] In the formula, R represents the number of reservoirs; The expected power output of the r-th reservoir during the t-th time period is obtained from the first moment estimate of the reservoir power output. Let be the expected reservoir capacity of the r-th reservoir at the end of the scheduling period, which is obtained from the first moment estimate of the reservoir capacity at the end of the period;

[0101] Ecological target quantification is based on the river flow probability density function. Calculate the flow rate below the ecological minimum. The probability value is used as the ecological damage rate, and the scheduling objective is to minimize the ecological damage rate.

[0102]

[0103] In the formula, This represents the probability that the river flow will be lower than the minimum flow during time period t, and its range is between 0 and 1. The ecological damage rate is obtained by calculating the average value over the entire scheduling period T.

[0104] Furthermore, such as Figure 4 As shown, the specific process of constructing a multi-objective stochastic optimization scheduling model is as follows:

[0105] First, the Markov chain Monte Carlo method is used to simulate multi-source random inputs and generate multiple random scenarios. Specifically, a Markov chain is constructed such that its stationary distribution is the target probability distribution, thereby generating a sample sequence that follows this distribution. The measured runoff conditional probability density function and the estuary chloride prediction probability density function are used as the target distributions. A large number of samples are extracted from these distributions using the Markov chain Monte Carlo method, each representing a possible runoff occurrence process or a salinity intrusion occurrence process. These samples are combined and paired to generate multiple random scenarios. Each random scenario contains a complete runoff process curve and a complete chloride process curve, together forming a set of possible input conditions for the scheduling model.

[0106] Secondly, for each random scenario, a preset scheduling mode is used for simulation to obtain the random simulation results of the reservoir outflow process, reservoir storage state, and river flow process. Specifically, the runoff process line in each random scenario is used as the reservoir inflow sequence. Combined with the reservoir's operating rules and constraints, the outflow that the reservoir should release is calculated time-by-time according to the preset scheduling mode, thus obtaining the reservoir outflow process for the entire scheduling period. At the same time, based on the water balance principle, the change process of the reservoir storage state is deduced from the inflow, outflow, and reservoir capacity curves. Furthermore, the reservoir outflow process is superimposed with the inflow of water in the interval to obtain the flow process of each section of the river. Since the input consists of multiple sets of random scenarios, each set of scenarios produces corresponding simulation results. Therefore, the above output results all present as a set of random processes, rather than a single deterministic trajectory.

[0107] Then, the random simulation results are input into a deep graph neural network model to simulate the stochastic process of estuarine chloride content. The deep graph neural network model, trained in previous steps, is capable of predicting changes in estuarine chloride content based on input information such as river flow, tidal level, and meteorological factors. Using the simulated river flow process under each random scenario as the key input to the deep graph neural network model, combined with corresponding tidal boundary conditions and meteorological conditions, the stochastic process of estuarine chloride content evolution over time under that scenario can be obtained. By calculating each random scenario, multiple possible trajectories of estuarine chloride content are obtained, fully reflecting the uncertainty transmission relationship from upstream scheduling decisions to downstream saline intrusion response.

[0108] Finally, based on the stochastic simulation results and the stochastic process of estuary chloride content, nonparametric kernel density estimation was used to quantify the probability density functions of reservoir water level, reservoir discharge, reservoir output, river flow, and estuary chloride content as dispatch response probability characteristic data. Specifically, the reservoir water level sequence, reservoir discharge sequence, reservoir output sequence, river flow sequence, and estuary chloride content sequence calculated under each stochastic scenario were summarized to form a set of sample data at each time point. The kernel density estimation method was used to process the sample data, which can estimate the probability density function of each variable at each time point without assuming that the data follows a specific distribution form. This data describes the probability distribution characteristics of the key response variables of the dispatch system under the condition of considering the dual uncertainties of runoff and tides, and constitutes the dispatch response probability characteristic data.

[0109] Through the above process, the multi-objective stochastic optimization scheduling model propagates the uncertainties of upstream runoff and downstream saltwater intrusion from the model input layer to the reservoir scheduling state layer and the estuary chloride response layer, ultimately manifesting as the uncertainty of the probability distribution of each scheduling objective. This allows scheduling decisions to no longer rely on single runoff and saltwater intrusion forecasts, but to assess the impact of different scheduling schemes on the risk of achieving each objective within the complete probability space, laying the model foundation for subsequent solutions to the Pareto set and the generation of collaborative scheduling decision schemes.

[0110] Step 140: Based on the measured runoff conditional probability density function and the predicted estuary chloride content probability density function, a multi-objective stochastic optimization scheduling model is used to iteratively solve for the Pareto solution set.

[0111] In this embodiment, based on the measured runoff conditional probability density function and the predicted estuarine chloride content probability density function, the Pareto solution set is obtained by iteratively solving a multi-objective stochastic optimization scheduling model. This includes: using a deep learning surrogate model to approximate the mapping relationship between the multi-source random input and the probability distribution of each scheduling objective in the multi-objective stochastic optimization scheduling model; using multi-scale dimensionality reduction technology to select a new population during the population update process of the evolutionary algorithm; and obtaining the optimal Pareto solution set through non-dominated sorting and crowding distance calculation.

[0112] In the embodiments of this application, such as Figure 5 As shown, a deep learning surrogate model is used to replace the multi-objective stochastic optimization scheduling model, and dimensionality reduction techniques are used to reduce the dimensionality of the optimization population. The specific process of iteratively solving to obtain the Pareto solution set is as follows:

[0113] First, a deep learning surrogate model is used to approximate the mapping relationship between multi-source random inputs and the probability distributions of each scheduling objective in the multi-objective stochastic optimization scheduling model. The multi-objective stochastic optimization scheduling model involves multiple complex computational stages, including runoff uncertainty quantification, saltwater intrusion uncertainty quantification, hydrodynamic simulation, and multi-objective evaluation. Directly embedding it into the optimization iteration process would result in extremely high computational costs, making it difficult to meet the timeliness requirements of practical engineering applications. To address this issue, a surrogate model is pre-constructed using deep learning technology. This surrogate model takes multi-source random inputs as input variables and the probability distribution characteristics of each scheduling objective as output variables, learning the complex nonlinear mapping relationship between them through extensive sample training. After training, in subsequent optimization iterations, the original multi-objective stochastic optimization scheduling model is no longer invoked; instead, the deep learning surrogate model is directly used for objective function evaluation, significantly reducing the computational cost of each evaluation and enabling large-scale iterative optimization.

[0114] Secondly, during the population update process of the evolutionary algorithm, a multi-scale dimensionality reduction technique is employed to select a new population. Evolutionary algorithms search the decision variable space by simulating natural selection and genetic mutation mechanisms to find the optimal solution set. When the dimensionality of the decision variables in the reservoir group scheduling problem is high, the search space grows exponentially, leading to a significant decrease in the algorithm's convergence speed—the so-called curse of dimensionality. The multi-scale dimensionality reduction technique projects the high-dimensional decision variable space onto a low-dimensional feature space, performs comparisons and selections of individuals in the low-dimensional space, and then maps the selected individuals back to the original high-dimensional space. This effectively compresses the search space dimensionality while preserving key information about the decision variables, accelerating the population convergence process and helping to escape local optima, thus improving global search capabilities.

[0115] Finally, the optimal Pareto solution set is obtained through non-dominated ranking and crowding distance calculation. In multi-objective optimization problems, there is usually no single solution that can simultaneously optimize all objectives. Therefore, it is necessary to find a set of non-dominated solutions that balance the various objectives, i.e., the Pareto solution set. Non-dominated ranking is used to divide individuals in the population into different levels according to their dominance relationships. Individuals not dominated by any other individuals form the first non-dominated front, individuals dominated only by individuals in the first front form the second non-dominated front, and so on. Crowding distance is used to measure the density of individuals within the same non-dominated front. Individuals with a larger crowding distance indicate that their surroundings are relatively sparse, and these individuals are preferentially retained during population selection to maintain the diversity of the solution set. By alternately performing non-dominated ranking and crowding distance calculation, the optimal individuals are selected from the merged set of the current population and the offspring population to form a new generation of population. After multiple generations of iterative evolution, a Pareto solution set with uniform distribution and good convergence is finally obtained.

[0116] Through the above process, the deep learning proxy model effectively replaces the computationally complex original scheduling model, the multi-scale dimensionality reduction technique overcomes the difficulty of searching in high-dimensional decision spaces, and the non-dominated sorting and crowding distance calculation ensure the convergence and diversity of multi-objective optimization. The synergistic effect of the three technologies enables efficient solution of multi-objective stochastic optimization scheduling models, providing a reliable Pareto solution set for the subsequent generation of collaborative scheduling decision schemes.

[0117] Step 150: Based on the trade-off relationship between different objectives in the Pareto solution set, generate the dynamic characteristic curve of the mapping relationship between social benefits and economic benefits, and generate a collaborative scheduling decision scheme under multiple uncertainties based on the dynamic characteristic curve.

[0118] In this embodiment, the Pareto solution set includes multiple non-dominated scheduling schemes, each achieving different trade-offs between the objectives of salinity suppression and freshwater replenishment, power generation, and ecological benefits. In this application, social benefits include the objectives of salinity suppression and freshwater replenishment, and ecological benefits include the objective of power generation. Since different objectives often compete with each other—for example, increasing the discharge flow from the upstream reservoir helps suppress salinity and increase the expected freshwater extraction time, it may also lead to a rapid drop in reservoir water level and a decrease in power generation head, thereby reducing power generation benefits and potentially impacting the river's ecology due to flow fluctuations—the various schemes in the Pareto solution set actually represent a mapping relationship between different degrees of trade-off between social and economic benefits.

[0119] To visually represent this mapping relationship, a dynamic characteristic curve depicting the mapping relationship between social and economic benefits is generated based on the trade-offs between different objectives in the Pareto solution set. Specifically, with economic benefit objectives such as power generation benefits as the horizontal axis and social benefit objectives such as salinity control and freshwater replenishment objectives or ecological objectives as the vertical axis, each scheme in the Pareto solution set is plotted as a point in a two-dimensional coordinate system. By connecting these points or fitting their changing trends, a curve sloping upwards to the right or exhibiting a specific shape is obtained. This reveals that as economic benefits increase, social benefits may increase or decrease, and whether there is a clear inflection point or critical region between the two. The slope of the curve reflects the marginal rate of change of social benefits relative to economic benefits, that is, the change in social benefits caused by each unit increase or decrease in economic benefits. By analyzing the shape and key features of this curve, the synergistic enhancement range, the competitive conflict range, and the optimal balance range between social and economic benefits can be identified.

[0120] Finally, different decision-makers or different practical application scenarios may place varying degrees of emphasis on social and economic benefits. For example, during periods of severe saltwater intrusion, the social benefits of ensuring water supply security take precedence over power generation benefits, while during non-saltwater intrusion periods, economic benefits may be of greater concern. Based on the dynamic characteristic curve, decision-makers can select corresponding points on the curve according to current actual needs and their tolerance for risk, thereby determining the corresponding dispatching scheme parameters. Furthermore, due to the multiple uncertainties in runoff and tides, the collaborative dispatching decision-making scheme includes not only deterministic dispatching rules or quantities but also a prior assessment of uncertain risks, i.e., the probability range of achieving each dispatching objective after the scheme is implemented and the risk level under extreme conditions. Through the above methods, a collaborative dispatching decision-making scheme that can achieve a dynamic balance between social and economic benefits under multiple uncertainties is ultimately generated, providing a scientific decision-making basis for the operation of reservoir groups for saltwater suppression and freshwater replenishment.

[0121] In this embodiment, the reservoir group in watershed A is taken as the example object. The reservoirs involved in the optimization include reservoirs B1, B2, B3, B4, and B5, and the freshwater extraction probability of the C estuary and pumping station D is considered. Taking a typical process from December 1 to December 15, 2021 as an example, the reservoir group salinity suppression and freshwater replenishment scheduling based on runoff tidal uncertainty provided in this application is adopted. The optimization of the upstream backbone reservoir group in watershed A is aimed at maximizing the freshwater extraction probability of the C estuary and pumping station D, maximizing the power generation benefit of the watershed reservoir group, and maximizing the ecological guarantee rate. Table 1 shows the Pareto optimal front of the algorithm optimization. It can be seen that the stronger the intensity of the upstream reservoir group salinity suppression and freshwater replenishment scheduling to replenish the downstream, the greater the freshwater extraction probability of pumping station D. However, due to the rapid drop in water level, the power generation head benefit of the reservoir group is damaged, the residual power is reduced, and the ecological flow guarantee rate gradually increases with the intensity of salinity suppression and freshwater replenishment.

[0122] Table 1 Objective values ​​of the Pareto optimal solution set

[0123]

[0124] Table 2 shows the optimized scheme, the upper and lower limits of the flow forecast, the interval flow, and the upper and lower limits of the flow after optimized scheduling of the upstream reservoir group. The data in the table shows that, firstly, the flow uncertainty after scheduling, i.e., the interval range, is reduced from 327-569 m³ / s to 93-239 m³ / s, indicating that a reliable scheduling process can effectively reduce the uncertainty of the flow at key downstream sections, thereby reducing the uncertainty of the freshwater extraction probability at the estuary pumping station; secondly, the minimum flow lower limit after scheduling increases from 1473 m³ / s to 1956 m³ / s, thus increasing the freshwater extraction probability at the downstream estuary. In summary, the results indicate that the multi-objective stochastic optimization scheduling technology of this invention can improve the cascade power generation benefits of the basin and the freshwater extraction probability at the downstream estuary while ensuring the ecological flow of the river channel. Simultaneously, it can reduce the impact of runoff and tidal forecast uncertainties on scheduling benefits through effective scheduling methods, avoiding the risks brought by uncertainty.

[0125] Table 2. Traffic scheduling before and after the scheme and its uncertainties

[0126]

[0127] This application provides a stochastic optimization scheduling system for salinity control and freshwater replenishment in a reservoir group under multiple uncertainties. (See also...) Figure 6 As shown, the stochastic optimization scheduling system for salinity control and freshwater replenishment of a reservoir group under multiple uncertainties provided in this application includes:

[0128] Data acquisition module 610 is used to acquire runoff forecast data, measured runoff data and saltwater intrusion influencing factors for the target area;

[0129] The data quantization module 620 is used to fit the joint distribution of runoff forecast data and measured runoff data using a Gaussian mixture model, and obtain the measured runoff conditional probability density function through Bayesian inference; and to extract the spatiotemporal characteristics of the influencing factors of saltwater intrusion using a deep graph neural network model, and obtain the estuary chloride content forecast probability density function through Bayesian inference and kernel density estimation.

[0130] Model module 630 is constructed to build a multi-objective stochastic optimization scheduling model based on the measured runoff conditional probability density function and the predicted probability density function of estuary chloride content.

[0131] The data solving module 640 is used to iteratively solve for the Pareto solution set using a multi-objective stochastic optimization scheduling model based on the measured runoff conditional probability density function and the predicted probability density function of estuary chloride content.

[0132] The strategy analysis module 650 is used to obtain the dynamic characteristic curve of the mapping relationship between social benefits and economic benefits based on the Pareto solution set, and to generate a collaborative scheduling decision scheme under multiple uncertainties based on the dynamic characteristic curve.

[0133] It should be noted that the principle of the random optimization scheduling system for salinity control and freshwater replenishment of reservoir groups under multiple uncertainties provided in this application embodiment to solve the technical problem is similar to the random optimization scheduling method for salinity control and freshwater replenishment of reservoir groups under multiple uncertainties provided in this application embodiment. Therefore, the implementation of the random optimization scheduling system for salinity control and freshwater replenishment of reservoir groups under multiple uncertainties provided in this application embodiment can refer to the implementation of the random optimization scheduling method for salinity control and freshwater replenishment of reservoir groups under multiple uncertainties provided in this application embodiment, and the repeated parts will not be described again.

[0134] After introducing the stochastic optimization scheduling method and apparatus for salinity suppression and freshwater replenishment of reservoir groups under multiple uncertainties provided in the embodiments of this application, the electronic equipment provided in the embodiments of this application will be briefly introduced next.

[0135] See Figure 7 As shown, the electronic device 500 provided in this application embodiment includes at least a processor 501, a memory 502, and a computer program stored in the memory 502 and capable of running on the processor 501. When the processor 501 executes the computer program, it implements the stochastic optimization scheduling method for salinity suppression and freshwater replenishment of reservoir groups under multiple uncertainties provided in this application embodiment.

[0136] The electronic device 500 provided in this application embodiment may further include a bus 503 connecting different components (including processor 501 and memory 502). The bus 503 represents one or more types of bus structures, including memory bus, peripheral bus, local area bus, etc.

[0137] Memory 502 may include a readable storage medium in the form of volatile memory, such as random access memory (RAM) 5021 and / or cache memory 5022, and may further include read-only memory (ROM) 5023. Memory 502 may also include a program tool 5025 having a set (at least one) of program modules 5024, including but not limited to an operating subsystem, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0138] Processor 501 can be a single processing element or a collective term for multiple processing elements. For example, processor 501 can be a central processing unit (CPU), or one or more integrated circuits configured to implement the stochastic optimization scheduling method for salinity control and freshwater replenishment of reservoir groups under multiple uncertainties provided in the embodiments of this application. Specifically, processor 501 can be a general-purpose processor, including but not limited to CPUs, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0139] Electronic device 500 can communicate with one or more external devices 504 (e.g., keyboard, remote control, etc.), and also with one or more devices that enable a user to interact with electronic device 500 (e.g., mobile phone, computer, etc.), and / or with devices that enable electronic device 500 to communicate with one or more other electronic devices 500 (e.g., router, modem, etc.). This communication can be performed through input / output (I / O) interface 505. Furthermore, electronic device 500 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) through network adapter 506. Figure 7 As shown, network adapter 506 communicates with other modules of electronic device 500 via bus 503. It should be understood that, although... Figure 7As not shown, other hardware and / or software modules may be used in conjunction with the electronic device 500, including but not limited to microcode, device drivers, redundant processors, external disk drive arrays, Redundant Arrays of Independent Disks (RAID) subsystems, tape drives, and data backup storage subsystems.

[0140] It should be noted that, Figure 7 The electronic device 500 shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0141] The following describes the computer-readable storage medium provided in the embodiments of this application. The computer-readable storage medium provided in the embodiments of this application stores computer instructions, which, when executed by a processor, implement the stochastic optimization scheduling method for salinity control and freshwater replenishment of reservoir groups under multiple uncertainties provided in the embodiments of this application. Specifically, the computer instructions can be built into or installed in the processor, so that the processor can implement the stochastic optimization scheduling method for salinity control and freshwater replenishment of reservoir groups under multiple uncertainties provided in the embodiments of this application by executing the built-in or installed computer instructions.

[0142] In addition, the stochastic optimization scheduling method for reservoir group salinity suppression and freshwater replenishment under multiple uncertainties provided in the embodiments of this application can also be implemented as a computer program product. The computer program product includes program code, which implements the stochastic optimization scheduling method for reservoir group salinity suppression and freshwater replenishment under multiple uncertainties provided in the embodiments of this application when running on a processor.

[0143] The computer program product provided in this application embodiment may employ one or more computer-readable storage media, which may be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination thereof. Specifically, more specific examples (a non-exhaustive list) of computer-readable storage media include electrical connections with one or more wires, portable disks, hard disks, RAM, ROM, erasable programmable read-only memory (EPROM), optical fibers, portable compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0144] The computer program product provided in this application embodiment can be a CD-ROM and include program code, and can also run on electronic devices such as computers. However, the computer program product provided in this application embodiment is not limited thereto. In this application embodiment, the computer-readable storage medium can be any tangible medium that contains or stores program code, which can be used by or in conjunction with an instruction execution system, device, or apparatus.

[0145] It should be noted that although several units or sub-units of the device have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of this application, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of one unit described above can be further divided and embodied by multiple units.

[0146] Furthermore, although the operations of the method of this application are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0147] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0148] Obviously, those skilled in the art can make various modifications and variations to the embodiments of this application without departing from the spirit and scope of the embodiments of this application. Therefore, if these modifications and variations to the embodiments of this application fall within the scope of the claims of this application and their equivalents, this application also intends to include these modifications and variations.

Claims

1. A stochastic optimization scheduling method for salinity suppression and freshwater replenishment in a reservoir group under multiple uncertainties, characterized in that, include: Acquire runoff forecast data, measured runoff data, and saltwater intrusion influencing factors for the target area; A Gaussian mixture model is used to fit the joint distribution of the predicted runoff data and the measured runoff data, and the conditional probability density function of the measured runoff is obtained through Bayesian inference. A deep graphical neural network model is used to extract the spatiotemporal characteristics of the influencing factors of saline intrusion, and the predicted probability density function of estuarine chloride content is obtained through Bayesian inference and kernel density estimation. Specifically, the process of extracting the spatiotemporal characteristics of the influencing factors of saline intrusion using a deep graphical neural network model and obtaining the predicted probability density function of estuarine chloride content through Bayesian inference and kernel density estimation includes: using a variational inference strategy to approximate the true posterior probability distribution of the parameters of the deep graphical neural network model with a variational distribution; training the deep graphical neural network model by maximizing the lower bound of evidence or minimizing the relative entropy between the variational distribution and the true posterior probability distribution; and processing the output of the deep graphical neural network model through kernel density estimation to obtain the predicted probability density function of estuarine chloride content. A multi-objective stochastic optimization scheduling model is constructed based on the measured runoff conditional probability density function and the predicted estuary chloride concentration. The construction of the multi-objective stochastic optimization scheduling model includes: using the measured runoff conditional probability density function and the predicted estuary chloride concentration as multi-source random inputs, simulating the process using a preset scheduling mode, and using multiple scheduling objectives as multi-dimensional probabilistic objectives to construct the multi-objective stochastic optimization scheduling model. The preset scheduling mode includes at least one of a flow control mode, a water level control mode, or other rule-based scheduling modes. The scheduling objectives include salinity reduction and freshwater replenishment, power generation, and ecological objectives. A Markov chain Monte Carlo method is used to simulate the multi-source random input, generating multiple random scenarios. For each random scenario, a preset scheduling mode is used for simulation to obtain random simulation results of the reservoir outflow process, reservoir water storage state, and river flow process. The random simulation results are input into the deep graph neural network model to simulate the random process of estuary chloride content. Based on the random simulation results and the random process of estuary chloride content, nonparametric kernel density estimation is used to quantify the probability density functions of reservoir water level, reservoir outflow, reservoir output, river flow, and estuary chloride content as scheduling response probability feature data. Based on the measured runoff conditional probability density function and the predicted estuary chloride content probability density function, the Pareto solution set is obtained by iteratively solving using the multi-objective stochastic optimization scheduling model. Based on the Pareto solution set, the dynamic characteristic curve of the mapping relationship between social benefits and economic benefits is obtained, and a collaborative scheduling decision scheme under multiple uncertainties is generated according to the dynamic characteristic curve.

2. The stochastic optimization scheduling method for salinity suppression and freshwater replenishment of reservoir groups under multiple uncertainties as described in claim 1, characterized in that, A Gaussian mixture model is used to fit the joint distribution of the predicted runoff data and the measured runoff data, and the conditional probability density function of the measured runoff is obtained through Bayesian inference, including: A joint distribution dataset is obtained based on the runoff forecast data and the measured runoff data, and the joint distribution dataset is split into a forecast runoff sub-dataset and a measured runoff sub-dataset. The probability distribution of the joint distribution dataset is fitted using a Gaussian mixture model to obtain the mean vector and covariance matrix of each Gaussian component; Based on the mean vector and covariance matrix of each Gaussian component, a joint probability density function of the runoff forecast data and the measured runoff data is constructed. Based on full probability reasoning, the predicted runoff subset is processed to obtain the marginal probability density function; Based on the joint probability density function and the marginal probability density function, the measured runoff subset is processed using the Bayesian inference to obtain the measured runoff conditional probability density function.

3. The stochastic optimization scheduling method for salinity suppression and freshwater replenishment of reservoir groups under multiple uncertainties as described in claim 1, characterized in that, The deep graph neural network model includes a meteorological information extraction module and a time series analysis module. The meteorological information extraction module is used to extract the influence characteristics of tidal fluctuations, rainfall and wind speed factors on the estuary chloride content through convolutional layers. The time series analysis module is used to establish the nonlinear time series relationship between the influence characteristics and the hydrological characteristics of key stations and the estuary chloride content through a gated cyclic network, and output the estuary chloride content forecast value.

4. The stochastic optimization scheduling method for salinity suppression and freshwater replenishment of reservoir groups under multiple uncertainties as described in claim 1, characterized in that, Based on the measured runoff conditional probability density function and the predicted estuary chloride content probability density function, the Pareto solution set obtained by iteratively solving using the multi-objective stochastic optimization scheduling model includes: A deep learning proxy model is used to replace the mapping relationship between the multi-source random input and the probability distribution of each scheduling objective in the multi-objective stochastic optimization scheduling model; In the population update process of evolutionary algorithms, multi-scale dimensionality reduction techniques are used to select new populations; The optimal Pareto solution set is obtained by non-dominated sorting and crowding distance calculation.

5. A stochastic optimization scheduling system for salinity suppression and freshwater replenishment of a reservoir group under multiple uncertainties, applicable to the stochastic optimization scheduling method for salinity suppression and freshwater replenishment of a reservoir group under multiple uncertainties as described in any one of claims 1 to 4, characterized in that, include: The data acquisition module is used to acquire runoff forecast data, measured runoff data, and saltwater intrusion influencing factors for the target area. The data quantization module is used to fit the joint distribution of the runoff forecast data and the measured runoff data using a Gaussian mixture model, and to obtain the measured runoff conditional probability density function through Bayesian inference; it uses a deep graph neural network model to extract the spatiotemporal characteristics of the saltwater intrusion influencing factors, and to obtain the estuary chloride content forecast probability density function through Bayesian inference and kernel density estimation. A model building module is used to construct a multi-objective stochastic optimization scheduling model based on the measured runoff conditional probability density function and the predicted probability density function of estuary chloride content. The data solving module is used to iteratively solve for the Pareto solution set using the multi-objective stochastic optimization scheduling model based on the measured runoff conditional probability density function and the predicted probability density function of estuary chloride content. The strategy analysis module is used to obtain the dynamic characteristic curve of the mapping relationship between social benefits and economic benefits based on the Pareto solution set, and generate a collaborative scheduling decision scheme under multiple uncertainties based on the dynamic characteristic curve.

6. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the stochastic optimization scheduling method for salinity suppression and freshwater replenishment of reservoir groups under multiple uncertainties as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions, which, when executed by a processor, implement the stochastic optimization scheduling method for salinity suppression and freshwater replenishment of reservoir groups under multiple uncertainties as described in any one of claims 1 to 4.

Images