Porous gate group scheduling method and system based on probability distribution

Abstract

The invention provides a multi-hole gate group scheduling method and system based on probability distribution, relates to the technical field of hydraulic engineering control, and is used for solving the technical problems that in the prior art, a scheduling scheme of a multi-hole gate group is low in decision-making efficiency, and it is difficult to guarantee that all engineering constraints are strictly met. The invention provides a multi-hole gate group scheduling method based on probability distribution, which comprises the following steps: enumerating a plurality of candidate scheduling schemes meeting engineering constraint conditions, solving probability distribution of discharge flow of each scheme and determining a distribution interval by combining target discharge flow and current water level forecast data, and constructing a discharge flow data set through multiple times of sampling, so as to realize multi-hole gate group scheduling. According to the method, the qualified scheduling schemes are screened by analyzing the data set, the final scheme is preferentially selected from the qualified scheduling schemes, and the scheduling instruction is generated and issued to the gate control system, so that the decision-making difficulty is reduced, the prediction uncertainty in the decision-making analysis process can be effectively coped with, and the safe and accurate scheduling of the porous gate group is realized.

Description

Technical Field

[0001] This invention relates to a method and system for scheduling multi-gate groups based on probability distribution, belonging to the field of control technology in water conservancy engineering. Background Technology

[0002] In the field of water conservancy engineering, large-scale gate systems are core infrastructure for watershed water resource regulation and flood control. Their scheduling process involves coordinating the opening combinations of multiple gates to precisely control the discharge flow of the hub, thereby dynamically balancing upstream and downstream water levels and ultimately achieving synergistic optimization of multiple engineering objectives, including flood control, water conservancy, and ecological benefits. However, some water conservancy projects typically include a considerable number of gates, such as 18 gates. Their scheduling scheme is an 18-dimensional decision vector, and high-dimensional vectors present significant challenges in decision-making. Furthermore, to ensure the safety of the hub structure, stable equipment operation, and controllable water flow, the multi-gate system must adhere to numerous engineering constraints. Some engineering scenarios even involve dozens of safety constraints. Traditional algorithms struggle to efficiently search within this high-dimensional, strongly constrained space, leading to low scheduling efficiency and difficulty in ensuring that the final scheduling scheme strictly meets all engineering constraints. This not only significantly increases the risk of scheme execution but may also trigger safety hazards such as torsional vibration of the gate structure and equipment overload damage.

[0003] For the scheduling and planning of large-scale gate groups, traditional methods often employ static machine learning models, such as static neural networks and support vector machines, to conduct decision analysis on scheduling schemes. These methods have the following drawbacks: Firstly, they treat water level and gate opening data at each moment as independent, identically distributed samples for training, completely ignoring the temporal dependence of water level. Water level changes have strong hydrodynamic inertia, and the discharge flow lags behind changes in gate opening. The discharge flow at the current moment depends not only on the current water level and opening but also on factors such as the previous moment's flow state and gate operation history. Static models ignore these dynamic characteristics, resulting in low prediction accuracy, inherent limitations, and large prediction residuals, leading to low accuracy in scheduling scheme decision analysis. Secondly, existing methods typically optimize the model's point prediction output as a fixed value, completely ignoring the prediction uncertainty caused by data noise, extrapolation, and insufficient modeling of dynamic characteristics. This means the final scheme may fall within a region of extremely low model prediction reliability, making it difficult to achieve the scheduling objective and posing a high execution risk, seriously affecting the operational safety and stability of water conservancy projects. Summary of the Invention

[0004] To address the aforementioned problems, or one of them, the present invention aims to provide a multi-gate group scheduling method based on probability distribution. This method enumerates multiple candidate scheduling schemes that meet engineering constraints, combines the target discharge flow rate with current water level forecast data, solves the probability distribution of the discharge flow rate for each scheme, determines the distribution interval, constructs a discharge flow rate dataset through multiple samplings, analyzes the dataset to screen qualified scheduling schemes, selects the best scheme from among them, generates a scheduling command, and sends it to the gate control system. This solves the problems of low decision-making efficiency and lack of uncertainty handling in decision analysis in existing technologies, thereby achieving safe and precise scheduling of multi-gate groups.

[0005] To address the aforementioned problems or one of them, the second objective of this invention is to provide a multi-gate group scheduling system based on probability distribution. By enumerating multiple candidate scheduling schemes that meet engineering constraints, and combining the target discharge flow rate with current water level forecast data, the system solves the probability distribution of the discharge flow rate for each scheme and determines the distribution interval. After multiple samplings, a discharge flow rate dataset is constructed. Qualified scheduling schemes are screened by analyzing the dataset, and the final scheme is selected from among them. A scheduling command is then generated and sent to the gate control system. This solves the problems of low decision-making efficiency and lack of uncertainty handling in decision analysis in the prior art, thereby achieving safe and precise scheduling of multi-gate groups.

[0006] To achieve one of the above objectives, the first technical solution of the present invention is as follows: A multi-gate group scheduling method based on probability distribution includes the following: Obtain the target discharge rate and the current water level forecast data; Based on the target discharge flow rate, multiple candidate scheduling schemes that meet the engineering constraints are enumerated, and the candidate scheduling schemes include combinations of gate opening degrees of the multi-hole gate group. Based on the water level forecast data at the current moment, the probability distribution data of the discharge flow under each candidate scheduling scheme is solved, and the distribution range of the discharge flow is determined. Multiple samples are taken in the distribution range to obtain multiple discharge flow sample values, which constitute the discharge flow dataset corresponding to the candidate scheduling scheme. By analyzing the leakage data set corresponding to each candidate scheduling scheme, qualified scheduling schemes are selected. Select the final scheduling scheme from all qualified scheduling schemes; The gate control system generates scheduling instructions based on the final scheduling scheme and sends them to the multi-gate group.

[0007] As a preferred technical measure: The engineering constraints include combined constraints and gate adjustment priority constraints. Based on the target discharge rate, the method for enumerating multiple candidate scheduling schemes that satisfy the engineering constraints is as follows: Based on the target discharge volume, determine the number of gates that need to be adjusted; Based on the required number of gates to be adjusted, and according to the gate adjustment priority constraint, select the candidate gates that need to be adjusted; For each candidate gate, based on the current opening degree, values ​​are taken at the minimum step size, and several discrete opening degree values ​​are sampled within the feasible range of the gate. The discrete opening values ​​of all candidate gates are combined to generate multiple partial opening combinations consisting of the openings of the candidate gates; In each partial opening combination, the original opening of other gates is added to form multiple complete opening combinations that include the openings of all gates; wherein, the other gates refer to the gates in the multi-hole gate group excluding the candidate gates; According to the combination constraints, all complete opening combinations are screened to obtain multiple gate opening combinations of multi-hole gate groups, thus forming multiple candidate scheduling schemes; The combined constraints include one or more of the following: single-hole stroke constraints, gate symmetry constraints, adjacent-hole opening difference constraints, and operational stability constraints.

[0008] As a preferred technical measure: The method for determining the number of gates that need to be adjusted based on the target discharge flow rate is as follows: Based on the difference between the target discharge flow rate and the current discharge capacity, calculate the total required opening of the gates to be increased or decreased, and determine the number of gates that need to be adjusted by combining the gate activation threshold constraints.

[0009] As a preferred technical measure: The method for determining the number of gates that need to be adjusted based on the target discharge flow rate is as follows: The number of gates that need to be adjusted is determined based on the target discharge flow, the water level in front of the dam, and the discharge capacity of the multi-gate group under different gate opening numbers.

[0010] As a preferred technical measure: The gate activation threshold constraint conditions include: single-gate operation is allowed when the total demand opening does not exceed the single-gate threshold, and multiple gates need to be activated when the single-gate threshold is exceeded. The priority constraints for gate adjustment include: even-numbered gates take precedence over odd-numbered gates; for gates with the same parity, the closer they are to the center of the spillway, the higher their priority. The single-hole stroke constraint includes: within the maximum stroke range of the gate, the gate opening is adjusted in integer multiples of the minimum step size; The gate symmetry constraint condition includes: the difference in opening degree between any gate and its symmetrical gate does not exceed the gate's set deviation threshold. The constraint condition for the difference in opening degree between adjacent gates includes: the difference in opening degree between adjacent gates does not exceed the maximum allowable drop. The operational stability constraint means that the maximum opening degree of each gate in a single scheduling does not exceed the maximum allowable change. Combining the discrete opening values ​​of all candidate gates means performing a Cartesian product combination on the discrete opening values ​​of all candidate gates.

[0011] As a preferred technical measure: The method for calculating the probability distribution of discharge flow under each candidate scheduling scheme based on the current water level forecast data is as follows: Acquire historical context data, which includes water level data, gate opening combinations, and actual discharge values ​​at several past moments; Based on historical context data, current water level forecast data, and the gate opening combinations included in the candidate scheduling schemes, the distribution of discharge flow is predicted according to the time-series dependence of water level, thus obtaining the probability distribution data of discharge flow.

[0012] As a preferred technical measure: The probability distribution data includes the mean and standard deviation of the leakage flow under a normal distribution; The method for determining the distribution range of the leakage flow is as follows: Based on the mean value of the leakage flow, three times the standard deviation are superimposed above and below it to obtain the interval of the leakage flow distribution. At the same time, the probability density of the distribution interval is calculated. The method for obtaining multiple leakage flow sample values ​​by performing multiple samplings within the distribution range is as follows: Multiple random samples are taken within the distribution interval, and the number of samples is allocated according to the probability density to obtain multiple leakage flow sample values; Or / and, the water level data includes upstream water level data and downstream water level data; the water level forecast data includes upstream forecast water level data and downstream forecast water level data; Or / and, predicting the distribution of leakage flow refers to using a trained dynamic Bayesian network model to predict the distribution of leakage flow.

[0013] As a preferred technical measure: The method for selecting qualified scheduling schemes by analyzing the leakage flow dataset corresponding to each candidate scheduling scheme is as follows: Obtain the allowable deviation of the target discharge flow rate, and generate the discharge flow rate evaluation range based on the target discharge flow rate; Obtain the pass / fail probability threshold used to determine whether a candidate scheduling scheme is qualified; For each candidate scheduling scheme, calculate the probability that the leakage flow sample value contained in the leakage flow dataset falls into the leakage flow evaluation interval, and use this probability as the evaluation probability. Candidate scheduling schemes with an evaluation probability greater than the pass probability threshold are classified as pass scheduling schemes.

[0014] As a preferred technical measure: The method for selecting the final scheduling scheme from all qualified scheduling schemes is as follows: Construct a comprehensive loss function consisting of the penalty target and its corresponding weights; For each qualified scheduling scheme, the comprehensive loss function is used to calculate the comprehensive score of each qualified scheduling scheme. The final gate opening scheduling scheme is selected based on the comprehensive score of each qualified scheduling scheme.

[0015] To achieve one of the above objectives, the second technical solution of the present invention is as follows: A multi-gate group scheduling system based on probability distribution includes the following modules: The data acquisition module is used to acquire the target discharge flow rate and the current water level forecast data; The scheme enumeration module is used to enumerate multiple candidate scheduling schemes that meet the engineering constraints based on the target discharge flow rate. The candidate scheduling schemes include combinations of gate opening degrees of a multi-hole gate group. The data generation module is used to solve the probability distribution data of the discharge flow under each candidate scheduling scheme based on the water level forecast data at the current time, and to determine the distribution range of the discharge flow. Multiple samples are taken in the distribution range to obtain multiple discharge flow sample values, thereby forming the discharge flow dataset corresponding to the candidate scheduling scheme. The qualified scheduling scheme screening module is used to screen out qualified scheduling schemes by analyzing the leakage flow dataset corresponding to each candidate scheduling scheme. The final scheme selection module is used to select the final scheduling scheme from all qualified scheduling schemes; The instruction generation and issuance module is used to generate scheduling instructions based on the final scheduling scheme and issue them to the gate control system of the multi-gate group.

[0016] Compared with existing technical solutions, the present invention has the following beneficial effects: This invention obtains the final scheduling scheme through enumeration and selection. The multiple candidate scheduling schemes based on the target discharge flow enumeration focus on feasibility coverage rather than precise optimization. All candidate scheduling schemes only need to meet the engineering constraints. There is no need to pursue the fine enumeration of gate opening values. A large number of preliminary combinations can be generated, which greatly reduces the difficulty of scheme generation. Furthermore, the engineering constraints are incorporated into the enumeration stage, which can strictly ensure that the final scheduling scheme can meet the engineering constraints and ensure execution safety.

[0017] After acquiring water level forecast data, this invention calculates the probability distribution of the discharge flow for each candidate scheduling scheme, thereby clarifying the distribution range of the discharge flow. Based on this, the invention performs multiple samplings within this distribution range, generating multiple discharge flow sample values. These sample values ​​collectively constitute the discharge flow dataset for the corresponding candidate scheme. These sample values ​​can simulate the discharge flow that may occur under various uncertain scenarios such as water level forecast uncertainty and fluctuations in water flow conditions, effectively quantifying the uncertainty in the discharge flow prediction process and providing data support for subsequent selection of the final scheduling scheme. By analyzing the dataset and selecting qualified schemes, the invention overcomes the shortcomings of low accuracy in enumerating schemes, eliminating schemes with low accuracy or those that cannot effectively meet control objectives, thus avoiding prediction error risks and providing a more reliable basis for subsequent optimal selection.

[0018] Finally, the optimal scheduling scheme is selected from all qualified scheduling schemes. This enumeration and selection method not only reduces the difficulty of decision-making, but also effectively copes with the predictive uncertainty in the decision analysis process, and achieves safe and precise scheduling of multi-gate groups. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the specific embodiments of this application or the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is a flowchart of Embodiment 1 of the present invention; Figure 2 This is a decision-making diagram of Embodiment 3 of the present invention; Figure 3 This is a schematic diagram of the modules in Embodiment 4 of the present invention. Detailed Implementation

[0021] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application. This invention covers any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of the invention as defined by the claims.

[0022] Example 1 like Figure 1 As shown, this embodiment describes a multi-gate group scheduling method based on probability distribution, including the following: Obtain the target discharge rate and the current water level forecast data; Based on the target discharge flow rate, multiple candidate scheduling schemes that meet the engineering constraints are enumerated, and the candidate scheduling schemes include combinations of gate opening degrees of the multi-hole gate group. Based on the water level forecast data at the current moment, the probability distribution data of the discharge flow under each candidate scheduling scheme is solved, and the distribution range of the discharge flow is determined. Multiple samples are taken in the distribution range to obtain multiple discharge flow sample values, which constitute the discharge flow dataset corresponding to the candidate scheduling scheme. By analyzing the leakage data set corresponding to each candidate scheduling scheme, qualified scheduling schemes are selected. Select the final scheduling scheme from all qualified scheduling schemes; The gate control system generates scheduling instructions based on the final scheduling scheme and sends them to the multi-gate group.

[0023] Furthermore, the engineering constraints include combined constraints and gate adjustment priority constraints; Based on the target discharge rate, the method for enumerating multiple candidate scheduling schemes that satisfy the engineering constraints is as follows: Based on the target discharge volume, determine the number of gates that need to be adjusted; Based on the required number of gates to be adjusted, and according to the gate adjustment priority constraint, select the candidate gates that need to be adjusted; For each candidate gate, based on the current opening degree, values ​​are taken at the minimum step size, and several discrete opening degree values ​​are sampled within the feasible range of the gate. The discrete opening values ​​of all candidate gates are combined to generate multiple partial opening combinations consisting of the openings of the candidate gates; In each partial opening combination, the original opening of other gates is added to form multiple complete opening combinations that include the openings of all gates; wherein, the other gates refer to the gates in the multi-hole gate group excluding the candidate gates; According to the combination constraints, all complete opening combinations are screened to obtain multiple gate opening combinations of multi-hole gate groups, thus forming multiple candidate scheduling schemes; The combined constraints include one or more of the following: single-hole stroke constraints, gate symmetry constraints, adjacent-hole opening difference constraints, and operational stability constraints.

[0024] The specific constraints can be as follows: The gate activation threshold constraint conditions include: single-gate operation is allowed when the total demand opening does not exceed the single-gate threshold, and multiple gates need to be activated when the single-gate threshold is exceeded. The priority constraints for gate adjustment include: even-numbered gates take precedence over odd-numbered gates; for gates with the same parity, the closer they are to the center of the spillway, the higher their priority. The single-hole stroke constraint includes: within the maximum stroke range of the gate, the gate opening is adjusted in integer multiples of the minimum step size; the minimum step size can be 0.5 meters. The gate symmetry constraint conditions include: the difference in opening degree between any gate and its symmetrical gate does not exceed the gate's set deviation threshold; for example, the gate's set deviation threshold is 0.3 meters; The constraint condition for the difference in opening degree between adjacent gates includes: the difference in opening degree between adjacent gates shall not exceed the maximum allowable drop, such as 0.5 meters; The operational stability constraint means that the maximum opening degree of each gate in a single scheduling does not exceed the maximum allowable change. Combining the discrete opening values ​​of all candidate gates means performing a Cartesian product combination on the discrete opening values ​​of all candidate gates.

[0025] Those skilled in the art can select the number of constraints or adjust the constraint thresholds accordingly based on the actual working conditions.

[0026] Furthermore, to better determine the number of gates that need adjustment, based on the target discharge flow rate, the method for determining the number of gates that need adjustment can be as follows: Based on the difference between the target discharge flow rate and the current discharge capacity, calculate the total required opening of the gates to be increased or decreased, and determine the number of gates that need to be adjusted by combining the gate activation threshold constraints.

[0027] Alternatively, based on the target discharge flow rate, another method for determining the number of gates that need to be adjusted is: The number of gates that need to be adjusted is determined based on the target discharge flow, the water level in front of the dam, and the discharge capacity of the multi-gate group under different gate opening numbers.

[0028] Furthermore, based on the current water level forecast data, the method for solving the probability distribution data of the discharge flow under each candidate scheduling scheme is as follows: Acquire historical context data, which includes water level data, gate opening combinations, and actual discharge values ​​at several past moments; the gate opening combination at a certain moment refers to the actions of all gates at that moment, and is a combination of the target opening values ​​of all gates at that moment. Based on historical context data, current water level forecast data, and the gate opening combinations included in the candidate scheduling schemes, the distribution of discharge flow is predicted according to the time-series dependence of water level, thus obtaining the probability distribution data of discharge flow.

[0029] By solving for the distribution of discharge flow based on the time-series dependence of water level and incorporating historical context data, the inertia of water flow and the lag effect of discharge flow can be captured, improving the accuracy of the distribution of discharge flow. If a dynamic Bayesian network model is used for prediction, the nonlinear discharge relationship can be accurately fitted, and the root mean square error of the model can be significantly reduced compared with the traditional static model, thus improving the prediction accuracy. By calculating and solving for the probability distribution of discharge flow under this scheme, the distribution range of discharge flow can be clarified. Based on this, multiple discharge flow sampling values ​​generated by multiple sampling can simulate the discharge flow that may occur under various uncertain scenarios such as uncertainty in water level forecast and fluctuation in water flow state. This effectively quantifies the uncertainty in the discharge flow prediction process and provides data support for the subsequent selection of the final scheduling scheme.

[0030] Furthermore, the probability distribution data includes the mean and standard deviation of the leakage flow rate under a normal distribution; The method for determining the distribution range of the leakage flow is as follows: Using the mean of the discharge flow rate as a benchmark, three times the standard deviation are superimposed above and below it to obtain the interval of the discharge flow rate distribution. At the same time, the probability density of this distribution interval is calculated. Such a distribution interval conforms to the characteristics of a normal distribution and can basically cover the distribution interval of the discharge flow rate.

[0031] The method for obtaining multiple leakage flow sample values ​​by performing multiple samplings within the distribution range is as follows: Multiple random samples are performed within the distribution interval, and the number of samples is allocated according to the probability density to obtain multiple leakage flow sample values. This sampling method can better cope with uncertainty. Obviously, in order to simplify the scheme, the sampling method in this embodiment is not limited to this. Other sampling methods can also be used, such as random sampling only within the interval, or sampling at fixed intervals.

[0032] Or / and, the water level data includes upstream water level data and downstream water level data; the water level forecast data includes upstream forecast water level data and downstream forecast water level data; wherein, the water level data may come from historical records, while the water level forecast data may come from forecast data from hydrological stations or meteorological stations.

[0033] Or / and, predicting the distribution of discharge flow refers to using a trained dynamic Bayesian network model to predict the distribution of discharge flow. The dynamic Bayesian network model can not only predict probabilities but also define state transition probabilities, treating water level changes as random disturbances superimposed on calculated values ​​based on the physical water balance formula. This reflects the uncertainties of exogenous variables such as inflow and demonstrates good water level dependence. In this invention, if a static Bayesian network model is used to predict the distribution of discharge flow, the accuracy is not good, but it is still superior to existing technologies.

[0034] Furthermore, the method for selecting qualified scheduling schemes by analyzing the leakage flow dataset corresponding to each candidate scheduling scheme is as follows: Obtain the allowable deviation of the target discharge flow rate, and generate the discharge flow rate evaluation range based on the target discharge flow rate; Obtain the pass / fail probability threshold used to determine whether a candidate scheduling scheme is qualified; For each candidate scheduling scheme, calculate the probability that the leakage flow sample value contained in the leakage flow dataset falls into the leakage flow evaluation interval, and use this probability as the evaluation probability. Candidate scheduling schemes with an evaluation probability greater than the pass probability threshold are classified as pass scheduling schemes.

[0035] The suitability of a plan is evaluated by probability, and a threshold for the probability of suitability is set. Only plans with a high probability of meeting the objective are retained, while plans with high uncertainty are eliminated, ensuring that the outflow after scheduling is stable within the expected range.

[0036] Furthermore, the method for selecting the final scheduling scheme from all qualified scheduling schemes is as follows: Construct a comprehensive loss function consisting of the penalty target and its corresponding weights; For each qualified scheduling scheme, the comprehensive loss function is used to calculate the comprehensive score of each qualified scheduling scheme. The final gate opening scheduling scheme is selected based on the comprehensive score of each qualified scheduling scheme.

[0037] The penalty target can be selected from the following data: the deviation between the mean discharge flow and the target discharge flow, the total adjustment of the gate opening, the degree of deviation from the engineering constraint threshold, and the standard deviation of the discharge flow probability distribution; For the deviation between the average discharge flow rate of the qualified solution and the target discharge flow rate, the larger the deviation, the higher the penalty value, to ensure that it meets the core objective of discharge flow control; For single-hole or multi-hole gates, the larger the total adjustment of the gate opening, the higher the risk of equipment wear and water flow fluctuation, and the higher the penalty value. For cases where the scheme is close to the engineering constraint threshold (such as the difference in opening between adjacent holes being close to the maximum allowable drop, or the difference in opening between symmetrical gates being close to the set deviation threshold), the closer the scheme deviates from the threshold, the higher the penalty value. Regarding the standard deviation of the probability distribution of leakage flow, the larger the standard deviation, the higher the prediction uncertainty and the higher the penalty value. Therefore, the more stable option should be selected first. The penalty target can also include: the matching degree of gate adjustment priority, but as a negative penalty, it can be treated as a bonus item, and bonus points can be assigned according to the degree of matching priority. The closer the solution is to the preset priority of the project, the less unnecessary low-priority gate actions can be reduced, and the complexity of equipment scheduling and energy consumption can be reduced.

[0038] The enumeration-based selection method adopted in this embodiment not only reduces the difficulty of decision-making, but also effectively addresses the predictive uncertainty in the decision analysis process, enabling safe and precise scheduling of multi-gate groups.

[0039] Example 2 This embodiment describes a multi-gate group scheduling method based on probability distribution, including the following: Obtain the target discharge rate and the current water level forecast data; The water level forecast data at the current moment includes the water level forecast value and the error range, which together constitute a water level sampling interval; the water level forecast data at the current moment comes from a hydrological station. Based on the target discharge flow rate, multiple candidate scheduling schemes that meet the engineering constraints are enumerated, and the candidate scheduling schemes include combinations of gate opening degrees of the multi-hole gate group. Based on the current water level sampling interval, multiple water level sampling values ​​are sampled; Based on the water level sampling values, the probability distribution data of the discharge flow under each candidate scheduling scheme is solved, and the distribution range of the discharge flow is determined. Multiple samplings are performed in the distribution range to obtain multiple discharge flow sampling values. This step is repeated to obtain the discharge flow sampling values ​​of each candidate scheduling scheme under all water level sampling values, thereby constructing the discharge flow dataset corresponding to the candidate scheduling scheme. By analyzing the leakage data set corresponding to each candidate scheduling scheme, qualified scheduling schemes are selected. Select the final scheduling scheme from all qualified scheduling schemes; The gate control system generates scheduling instructions based on the final scheduling scheme and sends them to the multi-gate group.

[0040] This embodiment describes a multi-gate group scheduling method based on probability distribution. The water level forecast data is also sampled, and together with the discharge flow sampling value, it constitutes two samplings, which further reduces the impact of water level forecast error on prediction accuracy and can improve prediction accuracy. However, the scheme increases the amount of calculation and is suitable for occasions with higher accuracy requirements.

[0041] Example 3 like Figure 2 As shown, this embodiment describes a multi-gate group scheduling method based on probability distribution, including the following: Step 1: Obtain the target discharge rate and the current water level forecast data; Specifically, this includes receiving instructions and acquiring the target leakage flow. And the current water level forecast data, including the current water level forecast value. and its forecast error range .

[0042] This refers to the upstream water level forecast value. This refers to the uncertainty deviation of the upstream water level forecast; This refers to the predicted downstream water level. This refers to the uncertain deviation of the downstream water level forecast.

[0043] The water level forecast data for the current moment mentioned above comes from the hydrological station.

[0044] Step 2: Based on the target discharge flow rate, enumerate multiple candidate scheduling schemes that meet the engineering constraints. The candidate scheduling schemes include combinations of gate opening degrees of the multi-gate group. This embodiment transforms actual engineering constraints into strict mathematical constraints, thereby enabling the search for scheduling schemes within a feasible space that fully meets safety requirements. For a system containing 18 gates (opening from 0 to 16 meters), the feasible region of the decision space can be defined by the following types of constraints. In particular, all constraints are defined with a gate opening discretization accuracy of 0.5 meters, and these constraints can be combined and adjusted according to the actual situation of the specific hub.

[0045] Single-orifice stroke and discrete accuracy constraints: The opening degree of each gate must be within the mechanically permissible range and is limited by the accuracy of the actuator; the opening degree must be an integer multiple of 0.5 meters.

[0046] in, Refers to the gate The degree of opening.

[0047] Symmetry constraints (applicable to symmetrically arranged gate groups): For hubs with symmetrical arrangements, in order to ensure structural safety and load balance, assuming that the gate numbers are symmetrical about the center, the following formula can be used to apply symmetry constraints;

[0048] in, It is a symmetric mapping function; It is a gate The opening degree of the gate that is symmetrical to it; The symmetry deviation threshold is preferably in the range of 0.5 meters to 1.0 meters, and its value is usually an integer multiple of the minimum step size of 0.5 meters; if the project does not have this requirement, this constraint can be turned off.

[0049] The constraint on the difference in opening between adjacent gates is a limitation on the drop between gates. The difference in opening between any two adjacent gates must be strictly controlled to prevent torsional vibration and unbalanced stress on the structure.

[0050] in, This refers to the gate The opening degree of adjacent gates; The maximum allowable drop is usually set to 3 meters according to engineering requirements; this constraint also applies to discrete opening values.

[0051] Gate adjustment priority rule: To optimize flow, gate adjustments follow a spatial priority pattern expanding outwards from the center of the spillway. This rule is applied to each gate. Define a static priority coefficient To achieve this, the coefficient setting principle is as follows: (1) The central gate has a higher priority; (2) The even-numbered gates have a higher priority than the odd-numbered gates; (3) Within the same parity, the closer to the center, the higher the priority.

[0052] The exemplary static priority coefficients for the 18 gates are: 10, 8, 12, 6, 14, 4, 16, 2, 18, 9, 11, 7, 13, 5, 15, 3, 17, 1.

[0053] Gate activation threshold constraint: Used to determine whether to activate a single gate or multiple gates. The gate activation strategy is determined based on the total required opening degree, which is defined as the sum of the opening degrees of all gates. When the total demand opening is below the single-hole threshold (e.g., 3 meters) When the total required opening exceeds the single-gate threshold, operation is permitted only with a single gate; At that time, at least A gate, such as The value is 2; Operational stability constraint: Limiting the maximum opening variation of each gate in a single scheduling operation to protect the equipment and ensure smooth operation. This variation is also limited by the minimum step size.

[0054] in, The maximum allowable variation range for a single hole (e.g., 0.5 meters or 1.0 meters) must be an integer multiple of 0.5 meters; For the gate at the previous moment The degree of opening.

[0055] Among the above constraints, the minimum step size (discrete precision) requirement in the engineering is combined with the priority rule. It not only mathematizes the engineering experience by discretizing the priority coefficient and the opening value, but also ensures that any scheduling scheme generated is physically executable (because the opening value is an integer multiple of 0.5 meters), thereby achieving a seamless connection from "optimized solution" to "control command".

[0056] Based on the above engineering constraints, the enumeration process for candidate scheduling schemes is as follows: Determine the number of gates that need to be adjusted, based on the target discharge flow rate. With current discharge capacity The gap is used to estimate the total demand that needs to be increased (or decreased). By combining the gate activation threshold constraints for single-gate and multi-gate systems, the number of gates requiring opening adjustment in this scheduling is determined. (For example, if) ,but ).

[0057] Current discharge capacity It can be based on the actual gate opening vector after the end of the previous scheduling time. Calculations show that the gate opening corresponds to a certain discharge capacity, and the current discharge capacity can be calculated based on the existing gate opening conditions. Although the accuracy of the results is relatively low, they can still be used as a reference for the current discharge capacity.

[0058] The number of gates to be adjusted can also be determined in other ways, such as based on the target discharge flow, the water level upstream of the dam, and the discharge capacity of the multi-gate group under different gate opening numbers. Specific methods include: The water level in front of the dam is divided into several intervals, such as three intervals: less than 43 meters, 43 to 45 meters, and greater than 45 meters. For each water level range, the gates are opened one by one, the discharge capacity range under different numbers of gates opened is calculated, and a table is prepared in advance; Based on the current water level in front of the dam, select the appropriate water level range; Based on the pre-made table, the corresponding discharge capacity range is matched according to the target discharge flow, and the number of gates to be opened is found according to the discharge capacity range. The number of gates to be opened is determined as the number of gates that need to be adjusted.

[0059] Then, candidate gates are selected based on their static priority coefficients. Choose the one with the highest priority. These gates constitute the candidate gate set for this scheduling. ; Next, discretization and combination generation are performed on the set. For each candidate gate, based on the constraint of a minimum step size of 0.5 meters, at its current opening degree... Nearby, several discrete aperture values ​​are sampled within its feasible range [0,16] with a fixed step size of 0.5 meters, usually 3-5 candidate values ​​are selected; For example, based on the current opening degree, take... Meters, but truncated to the interval [0,16]. The Cartesian product of the sampled values ​​of all candidate gates is combined to generate alternative partial opening combinations; Then, the partial aperture combinations are completed into complete aperture combinations, and each partial aperture combination is mapped back to a complete 18-dimensional aperture vector. (Not in) (The gate in the middle remains at its original opening). Then, check each Do other hard constraints (stroke, adjacent hole difference, symmetry, single-step variation, etc.) satisfy the criteria? Only retain N fully feasible candidate solutions. , forming a set .

[0060] Step 3: Based on the water level forecast data at the current moment, solve the probability distribution data of the discharge flow under each candidate scheduling scheme, and determine the distribution range of the discharge flow. Multiple samples are taken in the distribution range to obtain multiple discharge flow sample values, which constitute the discharge flow dataset corresponding to the candidate scheduling scheme. Based on historical time-series data, a dynamic Bayesian network model is constructed and trained to obtain a model that can predict the probability distribution of leakage flow. Using a pre-trained dynamic Bayesian network model, based on the current water level forecast data and dataset... Each candidate scheduling scheme Probability distribution prediction of leakage flow is performed to obtain the result for each candidate scheduling scheme. The probability distribution data of the outflow; the method for predicting the probability distribution data using a trained dynamic Bayesian network model is as follows: The current water level forecast data (using the current upstream and downstream water level forecasts) and candidate scheduling schemes are used. The gate opening combinations and historical context data from the past 3-5 time points are input into the dynamic Bayesian network model. The historical context data includes water level data, gate opening combinations, and actual discharge values ​​from the past 3-5 time points. The dynamic Bayesian network model outputs probability distribution data that follows a normal distribution, including the mean discharge value. and standard deviation ; The distribution range of the leakage flow determined based on the probability distribution data is as follows:

[0061] Multiple samples were taken within the aforementioned interval to obtain multiple leakage flow sample values. This constitutes a data set for the outflow rate; the sampling method can be Monte Carlo sampling, and the number of samplings can be dozens, such as 50. Therefore, for each candidate scheduling scheme... Each corresponds to a sample value of multiple leakage flows. This constitutes the leakage flow dataset.

[0062] Step 4: Analyze the leakage flow dataset corresponding to each candidate scheduling scheme to select qualified scheduling schemes; Specifically, it is an evaluation set. Candidate scheduling schemes Does it meet the following opportunity constraints:

[0063] Among them, the above formula This refers to multiple leakage flow sampling values fall into The probability within; Allowable discharge deviation; An acceptable risk level (e.g., 0.05); This is the sampled value of the discharge flow rate; To discharge traffic to the target; The schemes that satisfy the above constraints are considered qualified scheduling schemes, meaning that schemes in which more than 95% of the discharge flow sampling values ​​fall within the discharge flow evaluation range are considered qualified schemes. Only schemes that satisfy the above constraints are retained, forming a robust feasible set consisting of qualified scheduling schemes. .

[0064] Step 5: Select the final scheduling scheme from all qualified scheduling schemes; This step involves multi-objective comprehensive selection. Constructing a comprehensive loss function Optimize:

[0065] in, and Candidate scheduling schemes The corresponding mean and standard deviation of the discharge volume; This item is used to penalize deviations between the expected leakage rate and the target. This item is used to proactively penalize schemes with high prediction uncertainty, guiding the selection of regions with high model confidence. This item is used to penalize the total range of motion and reduce equipment wear; among them, It refers to the L1 norm; The term is used to penalize symmetry deviation. For symmetrical penalty targets; (If symmetric constraints are enabled).

[0066] It refers to the first The opening degree of each gate; It is a gate The opening degree of the gate that is symmetrical to it; The item is used to reflect the priority rules for gate adjustment, where This is a static priority coefficient. It has a negative weight. The purpose of this item is to increase the priority of high-priority gates, provided that other objectives are met. The opening degree (with a large value) is thus mathematically equivalent to realizing the engineering experience of "middle first, then both sides, even numbers first, then odd numbers". to These are non-negative weighting coefficients, set through normalization or analytic hierarchy process.

[0067] Finally, the scheme with the minimum overall loss is selected as the final scheduling scheme based on the following formula. ;

[0068] in The meaning is: in the robust feasible set In the process, find the solution that minimizes the overall loss function L(a).

[0069] like If the result is empty, meaning there is no suitable scheduling solution, an alarm will be triggered. Alternatively, a conservative strategy can be implemented, such as removing the candidate solutions... The minimum solution is used as the final scheduling solution. .

[0070] Step 6, based on the final scheduling scheme The system generates scheduling instructions and sends them to the gate control system of the multi-gate group. This allows the gate control system to operate according to the final scheduling plan. To control the opening degree of the gate.

[0071] Furthermore, the multi-gate group scheduling method based on probability distribution described in this embodiment may further include: Step 7: Dynamic Bayesian Network Model Update Steps.

[0072] Will include the final scheduling scheme After the scheduling command is issued to the gate control system for execution, the corrected discharge flow rate at that scheduling moment can be further calculated once more accurate water level evolution information is obtained (e.g., through higher-precision delayed water level compensation, inversion algorithms, or recalibration of the entire process). And by combining the water level data at that moment (such as the water level forecast), new data samples are obtained. The data is stored in the historical data buffer and the historical observation dataset is updated. The system can perform online updates or incremental learning on the dynamic Bayesian network model according to the set strategy. For example, it can use sliding window data retraining or online gradient descent to enable the dynamic Bayesian network model to maintain high-precision prediction capability for the discharge process under new operating conditions, seasonal changes or equipment characteristic drift, thereby achieving continuous self-optimization of the prediction model.

[0073] In this embodiment, to determine the number of gates that need to be adjusted when enumerating schemes, it is necessary to first calculate the current discharge capacity. Current discharge capacity Alternatively, it can be predicted using a pre-trained dynamic Bayesian network model, as follows: The current water level forecast data, the gate opening combination from the previous moment, and historical context data from the past 3-5 moments are input into the dynamic Bayesian network model. The dynamic Bayesian network model outputs probability distribution data that follows a normal distribution, namely the mean and standard deviation of the discharge flow. The mean discharge flow is considered as the current discharge capacity. The gate opening combination from the previous moment has been adjusted after a certain period, and can be considered as the actual gate opening at the current moment; the current discharge capacity is thus obtained. The accuracy is relatively high.

[0074] In the above embodiments, the method for constructing and training a dynamic Bayesian network model based on historical time-series data may include the following steps: Collect water level data, gate opening combinations, and actual discharge values ​​at several historical moments to form time-series training samples, which are then divided into training set, validation set, and test set. Construct a dynamic Bayesian network model, define the state transition probability and observation probability, and parameterize the probability distribution function; Based on the training set, with maximum likelihood estimation as the objective, the model is trained and its parameters are updated using a loss function and an optimizer; during the training process, the performance is monitored using a validation set to prevent overfitting. The prediction accuracy was verified on the test set to obtain a qualified dynamic Bayesian network model.

[0075] In this embodiment, the specific method for constructing and training the dynamic Bayesian network model can be as follows: First, data was collected, and a historical observation dataset was established based on the hub's records over the past three years. The historical observation dataset contains 5,000 data points, from which training, validation, and test sets can be allocated.

[0076] in:

[0077] This represents the water level data at time t, including upstream water level data. and downstream water level data .

[0078]

[0079] This represents the gate opening combination at time t, which is the action of the gate group at time t, and includes the target opening values ​​of 18 gates at time t. .

[0080] This represents the actual discharge rate at time t. Historical discharge rate values ​​are usually obtained after the fact through hydraulic formula inversion or extrapolation and can be used as a label for the model's "near-true value".

[0081] Secondly, a dynamic Bayesian network model is constructed and trained. This model assumes that the system follows a hidden Markov process over a short time scale. The core of the model is to define the following conditional probability distribution: First, define the state transition probability. It is used to describe the evolution of the system water level under historical conditions and actions, and expresses the time-series dependence of water level. Within the short-term scheduling scale, water level changes are regarded as random disturbances superimposed on the calculated value based on the physical water balance formula, so as to reflect the uncertainty of exogenous variables such as inflow. Next, we define the observation probability. This probability distribution represents the probability distribution in Real-time water level data, gate opening combinations, and historical context data Under the conditions, The flow rate at any moment The probability distribution of ; where Typically, it is a vector composed of water level data, gate opening combinations, and actual discharge values ​​from the most recent 3 to 5 time points, used to model temporal inertia; Parameterize this probability distribution as a normal distribution ;

[0082] in, and It is a function parameterized by a neural network (such as a fully connected network), which outputs the mean and standard deviation of the predicted leakage flow.

[0083] Historical data analysis reveals that within the monotonic interval of the discharge flow, the prediction error distribution is nearly symmetrical, indicating that a normal approximation can meet engineering accuracy requirements. (Historical context data) The introduction of this feature enables the model to learn time-series characteristics such as the dynamic inertia of water flow and the hysteresis effect of gate action, thereby providing more accurate mean predictions than static models. At the same time, the output standard deviation directly quantifies the prediction uncertainty under dynamic conditions.

[0084] Furthermore, model training is performed; The model training employs maximum likelihood estimation and a loss function; the training objective is to learn the optimal parameters by maximizing the likelihood function of historical observation data. .

[0085] For each sequence sample, its negative log-likelihood loss function Defined as:

[0086] Substituting the probability density function of the normal distribution, the loss function expands to:

[0087] To predict the average discharge volume; For the predicted standard deviation; This loss function has two key roles, the first of which... As a weighted mean square error, it drives the predicted mean. Approaching the true value, and in situations with high prediction uncertainty (i.e. The penalty is smaller on larger samples, which enhances the robustness of the model; the second term As a regularization term, it prevents the network from predicting excessive uncertainty in order to minimize the loss, and prompts the model to learn a well-calibrated uncertainty estimate that matches the input conditions.

[0088] The model is trained using the Adam optimizer, and the network parameters are iteratively updated using mini-batch data based on the training set. During training, early stopping is used to monitor performance on an independent validation set to prevent overfitting.

[0089] Finally, model validation and performance evaluation: After training, the model is evaluated on the reserved test set. Key performance metrics include: Prediction accuracy: Calculate the average predicted discharge volume. Compared to the actual leakage volume The root mean square error (RMSE) and / or mean absolute error (MAE) between them enable it to meet the industry's accuracy requirements.

[0090] Uncertainty calibration degree: evaluating the prediction The confidence interval (e.g., 95%) should ideally cover the proportion of the true values ​​actually found on the test set. .

[0091] The trained dynamic Bayesian network model becomes a powerful probability predictor.

[0092] For any given Water level data at any time Gate opening combination By combining historical context data, the model can output the leakage flow. The complete probability distribution, including the mean and standard deviation of the leakage flow, rather than a single deterministic point prediction.

[0093] The probability distribution-based multi-gate group scheduling method described in the above embodiments was applied to a water conservancy project. At this project, the system collected complete operational data from the past four years, including hourly upstream and downstream water level data, the gate opening combinations of the 18 floodgates, and the actual discharge volume calculated through hydraulic model inversion. After cleaning and standardization, the data formed a dataset containing approximately 5,000 valid samples.

[0094] The typical water level operating range for this hub is: upstream water level 36-46 meters, downstream water level 32-42 meters. A dynamic Bayesian network model containing a 3-layer fully connected neural network was constructed, with historical context data set to include data from the previous 4 hours. After 30 training epochs, the model achieved the following performance metrics on the validation set: Root mean square error of prediction (RMSE): reduced from 45 m³ / s in the traditional static model to 28 m³ / s; Prediction uncertainty calibration (the proportion of predicted 95% confidence intervals that contain the true value): 92%; Time-series prediction lag effect capture: The model successfully learned the 2-3 hour lag response law of the discharge flow to the change of gate opening; Nonlinear relationship modeling capability: The model accurately captures the parabolic relationship between discharge flow and gate opening. Specifically, when the gate opening increases from 0.5 meters to 1 meter, the increase in discharge flow per orifice is approximately 140-150 m³ / s; when the opening increases from 1 meter to 2 meters, the increase in discharge flow increases significantly; after the opening exceeds a certain threshold, the increase in discharge efficiency slows down. This nonlinear characteristic cannot be accurately expressed in traditional linear models.

[0095] At a certain point in time, the hub received a flood control order, requiring it to increase the discharge flow from the current 800 m³ / s to 1500 m³ / s within the next 3 hours to cope with heavy rainfall in the upstream basin. Simultaneously, the meteorological department forecast that the upstream water level might rise by 0.3-0.5 meters in the next 3 hours, with an uncertainty of ±0.1 meters. The execution process is as follows: Obtain the target discharge rate and the current water level forecast data; Target leakage flow: m³ / s; Forecast value of upstream water level m, downstream water level forecast m; Uncertainty in water level forecasts: m, m; Current gate status: (Currently, gates 5 and 6 are opened manually. Normally, the manual opening of gates on site does not strictly follow the scheduling rules.) Historical context data: Extract water level data, gate opening combinations, and actual discharge values ​​for the previous 4 hours.

[0096] Based on the target discharge flow rate, multiple candidate scheduling schemes that meet the engineering constraints are enumerated, and the candidate scheduling schemes include combinations of gate opening degrees of the multi-hole gate group. According to the actual operational requirements of the hub, the minimum opening of all gates must be ≥0.5 meters.

[0097] Demand Analysis and Gate Quantity Determination: The current discharge capacity is approximately 150 m³ / s (from two 0.5-meter opening gates), requiring an increase of 1350 m³ / s. Based on historical model patterns and nonlinear discharge characteristics, under typical water level conditions at this hub (42.5 meters upstream, 37.2 meters downstream), the discharge capacity of a single gate is approximately 140-150 m³ / s at an opening of 1 meter, and approximately 450-500 m³ / s at an opening of 2.5 meters. Based on the total required increase of 1350 m³ / s and the discharge capacity of the single gates, it is determined that approximately four additional medium-opening gates are needed.

[0098] When enumerating solutions, the following constraints should be enabled: Symmetry constraints ( rice); Neighbor-to-hole difference constraint ( rice); Single-step change constraints ( rice); Minimum gate opening constraint: When the gate participates in discharge, rice; Gate selection: According to the specific "middle first, even number first" project priority of this hub, the system selects the four gates with the highest priority: No. 10 (center), No. 8, No. 12, and No. 9 as the set to be adjusted. ; Discretization sampling: To fully utilize the nonlinear discharge efficiency, for these four gates, based on the minimum opening constraint, three discrete values ​​(e.g., 1.0, 2.0, 3.0 meters) are sampled within their possible opening ranges to generate... A preliminary combination; Feasibility filtering: After constraint checks, combinations that violate the rules of adjacent gate drop > 3 meters, symmetry deviation > 0.8 meters, or opening less than 0.5 meters are excluded, leaving 32 feasible candidate scheduling schemes. ; Using a pre-trained dynamic Bayesian network model, based on set The probability distribution of the outflow is predicted for each candidate scheduling scheme to obtain the probability distribution data of the outflow under each candidate scheduling scheme. The method for predicting probability distribution data using a trained dynamic Bayesian network model is as follows: The current upstream water level forecast value Downstream water level forecast The gate opening combinations from the candidate scheduling schemes, along with historical context data from the past four time points, are input into the dynamic Bayesian network model. The historical context data includes water level data, gate opening combinations, and actual discharge rates from the past four time points. The dynamic Bayesian network model outputs probability distribution data that follows a normal distribution, including the mean discharge rate. and standard deviation The distribution range of the leakage flow determined based on the probability distribution data is as follows:

[0099] Fifty Monte Carlo samplings were performed within the aforementioned interval, resulting in 50 leakage flow sampling values ​​for each candidate scheduling scheme. This constitutes the leakage flow dataset; For each candidate scheduling scheme, the corresponding leakage flow dataset is subjected to chance constraint filtering: set up m³ / s, The evaluation found that only 14 candidate scheduling schemes met the requirements. ,form ; Multi-objective optimization selects the final scheduling scheme from all qualified scheduling schemes; The weights of the loss function are set as follows: (Drainage deviation), (Uncertainty) (range of motion) (symmetry), (Priority-driven); Calculations show the optimal solution is:

[0100] The openings of gates 8, 10, and 12 are 2.0 meters, 3.0 meters, and 2.0 meters respectively, while gate 9 remains at 0 meters.

[0101] Key feature verification: All gates involved in the discharge (Nos. 5, 6, 8, 10, and 12) have an opening of 0.5 meters or more, satisfying the minimum opening constraint; Nonlinear discharge characteristics are as follows: Under typical water level conditions (42.5 meters upstream, 37.2 meters downstream), gate No. 10 with an opening of 3.0 meters provides a discharge flow of approximately 700 m³ / s, gates No. 8 and No. 12 with openings of 2.0 meters each provide approximately 350 m³ / s, and adding the original two 0.5-meter gates, the total predicted discharge flow is: m³ / s, m³ / s; satisfies all hard constraints; Total range of motion: Meters (within a safe range); The final scheduling scheme conforms to the engineering experience of center priority and even number priority; therefore, the final scheduling scheme satisfies the engineering constraints and is a good executable scheme.

[0102] The gate control system generates scheduling instructions based on the final scheduling scheme and sends them to the multi-gate group.

[0103] The dispatching instruction was issued to the gate control system at 08:05. Gates 8, 10, and 12 were opened to 2.0 meters, 3.0 meters, and 2.0 meters respectively as planned, while gates 5 and 6 were kept at 0.5 meters.

[0104] At 11:00 a.m., the actual monitored upstream water level was 42.6 meters (within the forecast range), still within the typical operating range of 36-46 meters. The actual average discharge during this period was calculated to be 1525 m³ / s using a more accurate hydraulic inversion model.

[0105] The system will provide new data samples Stored in the buffer, where During the low-load period at night, the system initiated incremental learning, using sliding window data from the past 7 days to fine-tune the parameters of the dynamic Bayesian network model, making the model more adaptable to the current flood season conditions and further optimizing the fitting accuracy of the nonlinear discharge parabolic relationship.

[0106] Upon testing, the method described in this invention has significant advantages in the following aspects; The precise handling of the minimum opening constraint is achieved by converting the engineering rule (minimum 0.5 meters) into a hard constraint, which eliminates all non-compliant schemes during the scheme generation stage, thus avoiding the problem of invalid fine-tuning that may occur in traditional methods. By integrating engineering experience with mathematical optimization, and through priority coefficient weighting, the engineering experience of "opening No. 10 first, then No. 8 and No. 12" is naturally realized in the mathematical optimization framework, rather than a simple external rule judgment. Active management of uncertainty: When faced with an uncertainty of ±0.1 meters in the upstream water level forecast, the opportunity constraint ensures that the discharge flow can be controlled within the range of 1450-1550 m³ / s in 95% of cases, thereby effectively dealing with the uncertainty of the water level forecast. In terms of computational efficiency, it can quickly converge to dozens of feasible solutions from a massive number of possible solutions and complete robust evaluation within 4 minutes, meeting the requirements of real-time scheduling.

[0107] This embodiment fully demonstrates the effectiveness and reliability of the present invention in practical applications of water conservancy projects, especially its ability to accurately model the nonlinear characteristics of discharge, providing a mature and feasible technical solution for the intelligent scheduling of large-scale water conservancy projects.

[0108] Compared with traditional manual experience-based scheduling, the present invention demonstrates better performance in practical applications. Regarding model prediction capabilities, traditional methods rely on static relationships or empirical formulas, failing to characterize the dynamic process of water flow and assess prediction reliability. In contrast, the dynamic Bayesian network model employed in this invention explicitly models temporal inertia and outputs a complete probability distribution, achieving a fundamental shift from "deterministic point prediction" to "probabilistic interval prediction."

[0109] In terms of decision optimization objectives, traditional scheduling often takes approximating the target discharge flow rate as the sole guideline. However, the multi-objective comprehensive loss function constructed in this invention can systematically balance discharge accuracy, model confidence, equipment action cost, and engineering priority, realizing an upgrade from "single objective satisfaction" to "multi-objective collaborative optimization".

[0110] Most importantly, in terms of dealing with uncertainty, traditional methods lack a quantitative management mechanism for model errors and external water level forecast errors. However, this invention actively utilizes the prediction distribution, performs chance constraint screening, and penalizes high uncertainty schemes in the loss function, so that the scheduling instructions can still maintain high reliability when facing disturbances, completing a paradigm shift from passive response to proactive robust defense.

[0111] These three core improvements together constitute a complete, reliable, and adaptive intelligent scheduling solution.

[0112] Example 4 like Figure 3 As shown, this embodiment describes a multi-gate group scheduling system based on probability distribution, including the following modules: The data acquisition module is used to acquire the target discharge flow rate and the current water level forecast data; The scheme enumeration module is used to enumerate multiple candidate scheduling schemes that meet the engineering constraints based on the target discharge flow rate. The candidate scheduling schemes include combinations of gate opening degrees of a multi-hole gate group. The data generation module is used to solve the probability distribution data of the discharge flow under each candidate scheduling scheme based on the water level forecast data at the current time, and to determine the distribution range of the discharge flow. Multiple samples are taken in the distribution range to obtain multiple discharge flow sample values, thereby forming the discharge flow dataset corresponding to the candidate scheduling scheme. The qualified scheduling scheme screening module is used to screen out qualified scheduling schemes by analyzing the leakage flow dataset corresponding to each candidate scheduling scheme. The final scheme selection module is used to select the final scheduling scheme from all qualified scheduling schemes; The instruction generation and issuance module is used to generate scheduling instructions based on the final scheduling scheme and issue them to the gate control system of the multi-gate group.

[0113] The module in this application is an object that uses physical or virtual representation to form an objective description of form and structure. The object is not the same as a physical object, and is not limited to physical or virtual. It can be a data processing function, software program, processing mode, usage method, operation mode, workflow, application process, electronic hardware, circuit module, processing system, system imitation or simulation object.

[0114] The embodiments described above are merely examples and not limitations. The scope of protection of this invention is not limited thereto, and should be determined by the description in the claims. Modifications and substitutions made by those skilled in the art based on the spirit and essence of this invention should all fall within the scope of protection of this invention.

Claims

1. A multi-gate group scheduling method based on probability distribution, characterized in that... Includes the following: Obtain the target discharge rate and the current water level forecast data; Based on the target discharge flow rate, multiple candidate scheduling schemes that meet the engineering constraints are enumerated, and the candidate scheduling schemes include combinations of gate opening degrees of the multi-hole gate group. Based on the water level forecast data at the current moment, the probability distribution data of the discharge flow under each candidate scheduling scheme is solved, and the distribution range of the discharge flow is determined. Multiple samples are taken in the distribution range to obtain multiple discharge flow sample values, which constitute the discharge flow dataset corresponding to the candidate scheduling scheme. By analyzing the leakage data set corresponding to each candidate scheduling scheme, qualified scheduling schemes are selected. Select the final scheduling scheme from all qualified scheduling schemes; The gate control system generates scheduling instructions based on the final scheduling scheme and sends them to the multi-gate group.

2. The multi-gate group scheduling method based on probability distribution as described in claim 1, characterized in that, The engineering constraints include combined constraints and gate adjustment priority constraints. Based on the target discharge rate, the method for enumerating multiple candidate scheduling schemes that satisfy the engineering constraints is as follows: Based on the target discharge volume, determine the number of gates that need to be adjusted; Based on the required number of gates to be adjusted, and according to the gate adjustment priority constraint, select the candidate gates that need to be adjusted; For each candidate gate, based on the current opening degree, values ​​are taken at the minimum step size, and several discrete opening degree values ​​are sampled within the feasible range of the gate. The discrete opening values ​​of all candidate gates are combined to generate multiple partial opening combinations consisting of the openings of the candidate gates; In each partial opening combination, the original opening of other gates is added to form multiple complete opening combinations that include the openings of all gates; wherein, the other gates refer to the gates in the multi-hole gate group excluding the candidate gates; According to the combination constraints, all complete opening combinations are screened to obtain multiple gate opening combinations of multi-hole gate groups, thus forming multiple candidate scheduling schemes; The combined constraints include one or more of the following: single-hole stroke constraints, gate symmetry constraints, adjacent-hole opening difference constraints, and operational stability constraints.

3. The multi-gate group scheduling method based on probability distribution as described in claim 2, characterized in that, The method for determining the number of gates that need to be adjusted based on the target discharge flow rate is as follows: Based on the difference between the target discharge flow rate and the current discharge capacity, calculate the total required opening of the gates to be increased or decreased, and determine the number of gates that need to be adjusted by combining the gate activation threshold constraints.

4. The multi-gate group scheduling method based on probability distribution as described in claim 2, characterized in that, The method for determining the number of gates that need to be adjusted based on the target discharge flow rate is as follows: The number of gates that need to be adjusted is determined based on the target discharge flow, the water level in front of the dam, and the discharge capacity of the multi-gate group under different gate opening numbers.

5. The multi-gate group scheduling method based on probability distribution as described in claim 3, characterized in that, The gate activation threshold constraint conditions include: single-gate operation is allowed when the total demand opening does not exceed the single-gate threshold, and multiple gates need to be activated when the single-gate threshold is exceeded. The priority constraints for gate adjustment include: even-numbered gates take precedence over odd-numbered gates; for gates with the same parity, the closer they are to the center of the spillway, the higher their priority. The single-hole stroke constraint includes: within the maximum stroke range of the gate, the gate opening is adjusted in integer multiples of the minimum step size; The gate symmetry constraint condition includes: the difference in opening degree between any gate and its symmetrical gate does not exceed the gate's set deviation threshold. The constraint condition for the difference in opening degree between adjacent gates includes: the difference in opening degree between adjacent gates does not exceed the maximum allowable drop. The operational stability constraint means that the maximum opening degree of each gate in a single scheduling does not exceed the maximum allowable change. Combining the discrete opening values ​​of all candidate gates means performing a Cartesian product combination on the discrete opening values ​​of all candidate gates.

6. The multi-gate group scheduling method based on probability distribution as described in claim 1, characterized in that, The method for calculating the probability distribution of discharge flow under each candidate scheduling scheme based on the current water level forecast data is as follows: Acquire historical context data, which includes water level data, gate opening combinations, and actual discharge values ​​at several past moments; Based on historical context data, current water level forecast data, and the gate opening combinations included in the candidate scheduling schemes, the distribution of discharge flow is predicted according to the time-series dependence of water level, thus obtaining the probability distribution data of discharge flow.

7. The multi-gate group scheduling method based on probability distribution as described in claim 6, characterized in that, The probability distribution data includes the mean and standard deviation of the leakage flow under a normal distribution; The method for determining the distribution range of the leakage flow is as follows: Based on the mean value of the leakage flow, three times the standard deviation are superimposed above and below it to obtain the interval of the leakage flow distribution. At the same time, the probability density of the distribution interval is calculated. The method for obtaining multiple leakage flow sample values ​​by performing multiple samplings within the distribution range is as follows: Multiple random samples are taken within the distribution interval, and the number of samples is allocated according to the probability density to obtain multiple leakage flow sample values; Or / and, the water level data includes upstream water level data and downstream water level data; the water level forecast data includes upstream forecast water level data and downstream forecast water level data; Or / and, predicting the distribution of leakage flow refers to using a trained dynamic Bayesian network model to predict the distribution of leakage flow.

8. The multi-gate group scheduling method based on probability distribution as described in claim 1, characterized in that, The method for selecting qualified scheduling schemes by analyzing the leakage flow dataset corresponding to each candidate scheduling scheme is as follows: Obtain the allowable deviation of the target discharge flow rate, and generate the discharge flow rate evaluation range based on the target discharge flow rate; Obtain the pass / fail probability threshold used to determine whether a candidate scheduling scheme is qualified; For each candidate scheduling scheme, calculate the probability that the leakage flow sample value contained in the leakage flow dataset falls into the leakage flow evaluation interval, and use this probability as the evaluation probability. Candidate scheduling schemes with an evaluation probability greater than the pass probability threshold are classified as pass scheduling schemes.

9. The multi-gate group scheduling method based on probability distribution as described in claim 1, characterized in that, The method for selecting the final scheduling scheme from all qualified scheduling schemes is as follows: Construct a comprehensive loss function consisting of the penalty target and its corresponding weights; For each qualified scheduling scheme, the comprehensive loss function is used to calculate the comprehensive score of each qualified scheduling scheme. The final gate opening scheduling scheme is selected based on the comprehensive score of each qualified scheduling scheme.

10. A multi-gate group scheduling system based on probability distribution, characterized in that... Includes the following modules: The data acquisition module is used to acquire the target discharge flow rate and the current water level forecast data; The scheme enumeration module is used to enumerate multiple candidate scheduling schemes that meet the engineering constraints based on the target discharge flow rate. The candidate scheduling schemes include combinations of gate opening degrees of a multi-hole gate group. The data generation module is used to solve the probability distribution data of the discharge flow under each candidate scheduling scheme based on the water level forecast data at the current time, and to determine the distribution range of the discharge flow. Multiple samples are taken in the distribution range to obtain multiple discharge flow sample values, thereby forming the discharge flow dataset corresponding to the candidate scheduling scheme. The qualified scheduling scheme screening module is used to screen out qualified scheduling schemes by analyzing the leakage flow dataset corresponding to each candidate scheduling scheme. The final scheme selection module is used to select the final scheduling scheme from all qualified scheduling schemes; The instruction generation and issuance module is used to generate scheduling instructions based on the final scheduling scheme and issue them to the gate control system of the multi-gate group.

Images