Multi-objective Decision-making Control Method for Staged Flood Season Water Levels of Reservoirs with Risk-Benefit Balance

Abstract

The present invention discloses a multi-objective decision-making control method for the stage flood season water level of a reservoir with a balance between risks and benefits, specifically relating to the technical field of data management. By introducing a genetic algorithm and a support vector machine for multi-generation optimization and non-linear regression model training, the water level change trend and the optimal water discharge are respectively predicted and used as input variables of a fuzzy control system. A weighted fusion algorithm is used to calculate the final continuous control quantity, and the matching degree between the control command and the equipment is evaluated according to the equipment response delay and the discretization accuracy. If the matching degree is low, the system will dynamically adjust the discretization rules and the water discharge strategy, and precisely dock with the equipment through the optimized control command, thereby improving the accuracy and efficiency of reservoir management, reducing the risk of dam break, and optimizing the water discharge operation of the reservoir.

Description

Technical Field

[0001] The present invention relates to the technical field of data management, and specifically to a multi-objective decision-making control method for the stage flood season water levels of a reservoir with a balance between risk and benefit. Background Art

[0002] The multi-objective decision-making control of the stage flood season water levels of a reservoir with a balance between risk and benefit refers to, in the process of reservoir management, by considering the water level control objectives in different flood season stages, weighing the risks (such as the risk of flood occurrence) and benefits (such as reservoir water storage benefits, power generation benefits, etc.), and conducting multi-objective decision-making control. Specifically, this control method aims to, in different flood season stages (such as the early, middle, and late flood seasons), according to factors such as the real-time water level of the reservoir, precipitation forecast, and flow change, formulate appropriate water level control strategies to ensure a balance between the safety and maximum benefit of the reservoir. Through the multi-objective decision-making model, the reservoir can make flexible adjustments when facing sudden risks and demands, not only reducing risks such as dam breakage but also making full use of the resources of the reservoir.

[0003] In the prior art, fuzzy logic is introduced to handle the uncertainties and ambiguities existing in water level control. It does not require an accurate mathematical model but sets fuzzy rules based on expert experience or historical data to adjust the reservoir water level in real time. For example, in the case of a high water level, fuzzy control can set rules to gradually increase the water discharge volume to reduce the risk of dam breakage while taking into account the water storage and power generation benefits. This technology can cope with the complexity and dynamic changes in reservoir management and has strong adaptability and flexibility.

[0004] The prior art has the following deficiencies:

[0005] When the reservoir fuzzy control system is connected to actual control devices (such as gates, pumps, etc.), there is an interface mismatch problem between the output of continuous control quantities and the discrete commands received by the devices. Specifically, the fuzzy control system usually generates refined control quantities (such as "50% of the water discharge volume"), but the devices may only be able to accept discrete control commands (such as opening or closing the gate), resulting in operational inconsistencies. In addition, the fuzzy control system may not consider the response delay or mechanical precision of the devices, causing deviations between the control decision and the actual execution, thereby affecting the accuracy and effectiveness of reservoir management. Summary of the Invention

[0006] The purpose of the present invention is to provide a multi-objective decision-making control method for the stage flood season water levels of a reservoir with a balance between risk and benefit to solve the problems in the background art.

[0007] To achieve the above purpose, the present invention provides the following technical solution: A multi-objective decision-making control method for the stage flood season water levels of a reservoir with a balance between risk and benefit, including:

[0008] Determine the input items affecting the reservoir water level control, including the current water level, precipitation, inflow rate, and reservoir water storage volume;

[0009] Input historical water level data, inflow rate, and precipitation data, and optimize the prediction of water level changes and the optimal water discharge volume prediction by combining the genetic algorithm and the support vector machine model;

[0010] Pass the prediction results as input items to the fuzzy control system, calculate the final continuous control quantities, including the water discharge volume and the water discharge speed, after using the weighted fusion algorithm, and convert 50% of the water discharge volume output by the fuzzy control system into the 50% opening state of the equipment;

[0011] Evaluate the interface matching degree between the continuous control quantity and the discrete command accepted by the equipment according to the discretization accuracy of the control command and the equipment response delay situation. Specifically, normalize the discrete command control accuracy index and the execution response delay anomaly index so that they are both in the range of [0, 1], and calculate the interface matching degree between the continuous control quantity and the discrete command accepted by the equipment according to the normalized discrete command control accuracy index and the execution response delay anomaly index;

[0012] If the matching degree is low, continuously adjust and optimize the control decision according to the evaluation results, including adjusting the discretization rule and the water discharge strategy, and send the adjusted discrete control command to the actual control equipment of the reservoir to complete the reservoir water discharge volume.

[0013] Preferably, use the genetic algorithm for multi-generation optimization to predict the future water level changes and output the predicted water level change trend value. Specifically, collect historical water level, inflow rate, and precipitation data, and construct a training data set. The independent variables of the data set include the water level, inflow rate, and precipitation data at several past moments, and the dependent variable is the water level at the predicted future moment; use the genetic algorithm for multi-generation optimization, and find the optimal solution by simulating the natural selection process. Specifically, initialize the population and randomly generate multiple candidate solutions, use the mean square error MSE as the fitness function, select the individuals with high fitness for crossover and mutation operations to generate a new generation of individuals, and stop the optimization process through the termination condition. Finally, obtain a set of optimal parameters; use the optimal parameters to predict the future water level change trend as the first input variable of the fuzzy control system.

[0014] Preferably, an SVM regression model is used to predict the optimal water discharge under different conditions and output the predicted water discharge value. Specifically: the model is trained by the support vector regression method in the support vector machine SVM; during the training process, the current water level, inflow rate and precipitation prediction data are collected as independent variables, while the optimal water discharge of the reservoir is used as the dependent variable; by constructing a training data set, the input features include the water level, inflow rate and precipitation at the current moment, and the target output is the optimal water discharge; the Gaussian radial basis kernel function is used to handle non-linear problems, optimize the SVM regression model, minimize the loss function, and find the optimal decision boundary and regression hyperplane; after the training is completed, the SVM regression model predicts the optimal water discharge based on the current water level, inflow rate and precipitation prediction data, and uses the prediction result as the second input variable of the fuzzy control system.

[0015] Preferably, in the reservoir water level control, the first input variable and the second input variable of the fuzzy control system are combined and weighted to obtain the final continuous control amount. Specifically: the first input variable is the future water level change trend W(t + 1) predicted by the genetic algorithm, that is, the predicted value of the future water level; the second input variable is the optimal water discharge predicted by the support vector machine SVM regression model , that is, the amount of water to be discharged under the current conditions;

[0016] Let be the water level change trend W(t + 1) and the predicted water discharge Set the weight coefficients, be the weight of the water level change trend, be the weight of the optimal water discharge; the sum of the weight coefficients is 1; according to the set weight coefficients, the weighted average formula is used to fuse the two input variables into a comprehensive control amount: ; where: is the comprehensive control amount, which is a control signal calculated based on the future water level change trend and the predicted water discharge.

[0017] Preferably, the fuzzy control system calculates the final water discharge and water discharge speed according to the fused comprehensive control amount , including:

[0018] Convert the input into a fuzzy set and define a set of fuzzy rule bases;

[0019] According to the fuzzy rule base, combined with the input fuzzified data, infer a fuzzy output, which represents the fuzzy decision of the water discharge or water discharge speed;

[0020] Use the defuzzification method to convert the fuzzy output into an actual continuous control amount.

[0021] Preferably, after analyzing the accuracy when converting the continuous control quantity of the fuzzy control system into discrete commands executed by the device, a discrete command control accuracy index is generated. The method for obtaining the discrete command control accuracy index is as follows:

[0022] Express the continuous control quantity output by the fuzzy control system as ;

[0023] The actual discrete control command executed by the device is expressed as ;

[0024] Calculate the error of each discrete command. The error of the discrete command is calculated from the difference between the continuous control quantity and the command executed by the device. The expression is: ; Through the weighted error method, calculate the weighted error value , and the expression is: ; where: z is the number of discretized control commands, is the weight coefficient of the i-th control command, is the error of the i-th discrete command. Calculate the discrete command control accuracy index, and the calculation expression is: ; where: is the discrete command control accuracy index.

[0025] Preferably, after analyzing the time response delay situation required for the device to actually execute the command after the control command is issued, an execution response delay anomaly index is generated. The method for obtaining the execution response delay anomaly index is as follows:

[0026] Collect the historical response delay data of the device. Select k nearest neighbors from the historical response delay data, and then calculate the difference between the response delay at the neighboring moment and the response delay at the current moment. Use the Euclidean distance to calculate the difference between the current response delay and the historical delay data , and the expression is: ; where: is the historical moment, is the response delay at the current moment, is the historical moment 's response delay;

[0027] According to the calculated distance, select the k historical moment response delay values with the smallest distance as the nearest neighbors, and calculate the average response delay of the selected k nearest neighbors. The expression is: ; where: is the average value of the historical response delay at the current moment t, and k is the number of selected neighbors; Calculate the execution response delay anomaly index, and the expression is: ; where: is the standard deviation of the historical response delay data, and ERD is the execution response delay anomaly index.

[0028] Preferably, compare the interface matching degree between the obtained continuous control quantity and the discrete command accepted by the device with a preset matching degree reference threshold. If the interface matching degree between the continuous control quantity and the discrete command accepted by the device is greater than or equal to the preset matching degree reference threshold, it indicates that the interface matching degree between the continuous control quantity and the discrete command accepted by the device is high, and a normal matching signal is generated at this time; if the interface matching degree between the continuous control quantity and the discrete command accepted by the device is less than the preset matching degree reference threshold, it indicates that the interface matching degree between the continuous control quantity and the discrete command accepted by the device is low, and a matching abnormal signal is generated at this time, and the discretization rule and the water release strategy need to be adjusted.

[0029] In the above technical solution, the technical effects and advantages provided by the present invention are as follows:

[0030] The present invention provides a multi-objective decision-making control method for the reservoir stage flood season water level with a balance between risk and benefit. By combining the genetic algorithm and the support vector machine to optimize the prediction of water level changes and the prediction of the optimal water release volume, the intelligent regulation of the reservoir water level and the water release volume is realized. This method provides accurate control quantities for the fuzzy control system by weighted fusion of the predicted water level change trend and the optimal water release volume, and optimizes the interface matching degree between the continuous control quantity and the discrete command accepted by the device by evaluating the discrete command control accuracy and the execution response delay abnormal index, thereby reducing the deviation between the control decision and the device execution. By continuously adjusting the discretization rule and the water release strategy, the accurate water release and safety management of the reservoir are ensured, the efficiency of reservoir management and the accuracy of water level regulation are effectively improved, the management risk caused by device response delay or control interface mismatch is avoided, and scientific decision-making support is provided for the reservoir management during the flood season. Description of the Drawings

[0031] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings.

[0032] Figure 1 It is the flowchart of the method of the present invention. Detailed Embodiments

[0033] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the protection scope of the present invention.

[0034] For the embodiments, please refer to Figure 1 As shown, the multi-objective decision-making control method for the stage flood season water levels of a reservoir with a balance between risks and benefits in this embodiment includes:

[0035] Determine the input items affecting the reservoir water level control, including the current water level, precipitation, inflow, and reservoir water storage;

[0036] Input historical water level data, inflow, and precipitation data, and optimize the water level change prediction and optimal water discharge prediction by combining the genetic algorithm and the support vector machine model;

[0037] Transfer the prediction results as input items to the fuzzy control system, calculate the final continuous control quantities, including the water discharge and the water discharge speed, after using the weighted fusion algorithm, and convert 50% of the water discharge output by the fuzzy control system into the 50% opening state of the equipment;

[0038] Evaluate the interface matching degree between the continuous control quantities and the discrete commands accepted by the equipment according to the discretization accuracy of the control commands and the equipment response delay;

[0039] If the matching degree is low, continuously adjust and optimize the control decision according to the evaluation results, including adjusting the discretization rules and the water discharge strategy, and send the adjusted discrete control commands to the actual control equipment of the reservoir to complete the reservoir water discharge.

[0040] Determining the input items affecting the reservoir water level control is the basis for effective reservoir management, and these input items provide key real-time information for the reservoir water level control decision-making. The following is a specific elaboration of each input item:

[0041] The current water level refers to the water surface height in the reservoir, usually measured in meters. It is the most direct indicator affecting the reservoir water discharge decision and the water storage strategy. The current water level determines whether the reservoir is approaching the flood overflow level or whether there is sufficient water for discharge. Too high a water level may lead to the risk of dam break, and too low a water level may affect the functions of the reservoir such as water supply and power generation. Therefore, accurate current water level data is the core of formulating the reservoir water level control strategy. It is monitored in real time through water level sensors and telemetry systems, and the data is updated regularly.

[0042] Precipitation refers to the total amount of precipitation within the reservoir's watershed area over a certain period of time, usually measured in millimeters (mm). Precipitation directly affects the water input to the reservoir's watershed. Heavy rain or continuous precipitation will cause the water volume in the rivers within the watershed to increase, thereby increasing the inflow rate of the reservoir. Conversely, during the dry season, when precipitation decreases, the water supply to the watershed also decreases correspondingly, and the reservoir needs to adjust its water level control strategy according to the changes in precipitation. Precipitation information is obtained through meteorological stations, satellite remote sensing technology, or meteorological forecast data. Reservoirs usually use short-term and long-term precipitation forecast data for prediction.

[0043] Inflow rate refers to the amount of water flowing into the reservoir per unit time, usually measured in cubic meters per second (m³ / s). This is a dynamic indicator affected by factors such as precipitation, upstream river flow, and snowmelt. The inflow rate directly determines the rising speed of the reservoir water level. An excessive inflow rate may cause the water level to rise rapidly, increasing the risk of dam failure; a too small inflow rate may lead to an overly low reservoir water level, unable to meet downstream demands. Accurate inflow rate data is crucial for reservoir management and helps to estimate the future water level change trend of the reservoir. The inflow volume is calculated through flow meters, river monitoring stations, remote sensing technology, or watershed hydrological models. The real-time monitoring system obtains the inflow rate data based on the hydrological characteristics of the reservoir's watershed.

[0044] The reservoir storage volume refers to the amount of water stored in the reservoir, usually measured in cubic meters (m³). It is affected not only by the current water level but also by factors such as the reservoir's capacity, design height, and the adjustable range of the water level. The reservoir storage volume determines the reservoir's water storage capacity and scheduling flexibility. When the reservoir storage volume is sufficient, the reservoir has more water available for downstream use (such as power generation, irrigation, water supply, etc.) and has the ability to regulate the water level. When the storage volume is low, the reservoir may need to implement more strict water level control to prevent the water level from dropping too quickly due to excessive or rapid water release. The reservoir storage volume is calculated through real-time water level data and the reservoir's storage capacity curve. The storage capacity curve is a graph showing the relationship between the water level and the storage volume given in the reservoir design, used to calculate the storage volume based on the current water level.

[0045] When using the Genetic Algorithm (GA) for reservoir water level prediction, the goal is to predict the future water level change trend value based on historical water level data, inflow rate, and precipitation through an optimization process, so as to serve as the first input variable of the fuzzy control system.

[0046] To predict the water level change trend, the following historical data needs to be collected first:

[0047] Historical water level data: including water level records over a certain period (such as the past few days or months). Inflow rate data: including the inflow rate data of the reservoir during this period. Precipitation data: the precipitation data during this period.

[0048] Construct a training dataset, and each record in the dataset will contain:

[0049] Independent variables (input features): Historical water level: water level data at several past moments. Inflow rate: inflow rate data at several past moments. Precipitation: precipitation data at several past moments.

[0050] Dependent variable (target output): Water level at a future moment (W(t+1)): predicted water level data, which is used as the prediction target of the model.

[0051] The basic idea of the genetic algorithm is to simulate the process of natural selection to find the optimal solution. GA usually includes the following steps:

[0052] Initialize the population: Randomly generate multiple candidate solutions (individuals), and the genes of each individual represent a set of parameters to be optimized. For example, the genes of a certain individual may include the water discharge of the reservoir, the influence coefficient of precipitation on the water level, the influence coefficient of inflow rate on the water level, etc.

[0053] Calculate the fitness function: Evaluate the fitness of each individual according to the prediction error, and set the mean squared error MSE as the fitness function: ; where, is the water level predicted by the current individual, is the actual historical water level data, and N is the number of data points.

[0054] Selection operation: Select individuals with higher fitness to enter the next generation according to the fitness value, usually using roulette wheel selection or tournament selection.

[0055] Crossover operation: Combine two parent individuals through the crossover operation to generate new offspring individuals. For example, exchange some gene segments of the parent individuals to generate new combinations.

[0056] Mutation operation: Perform mutation operations on some individuals, randomly change some gene values to increase the diversity of the population and avoid falling into local optimal solutions.

[0057] Termination condition: When the preset number of generations or the fitness meets a certain threshold, stop the optimization process.

[0058] Through multiple generations of optimization by the genetic algorithm, an optimal solution is finally obtained, and the prediction ability of the model is the best. According to this solution, use the optimal parameters obtained by the genetic algorithm to predict the future water level change trend. Set a set of optimal parameters optimized by the genetic algorithm as , and these parameters will be used to predict the water level change at future moments: ; where: W(t) is the water level at the current moment, is the inflow at the current moment; P(t) is the precipitation at the current moment; are other factors that may affect the water level, such as temperature, wind speed, etc.; The predicted future water level change trend W(t + 1) is output and used as the first input variable of the fuzzy control system.

[0059] The fuzzy control system will use this predicted water level change trend W(t + 1) to calculate the future water level and adjust the water release strategy based on this information. The output of the system (water release volume, water release speed, etc.) will be interfaced with the actual control equipment of the reservoir.

[0060] Using the Support Vector Machine (SVM) to predict the optimal water release volume of the reservoir is based on the current water level, inflow, and precipitation prediction data to train the model, and then provide the prediction results for the fuzzy control system to be used as the second input variable.

[0061] To train the SVM regression model, the following data needs to be collected:

[0062] Current water level: including the water level information of the reservoir at a certain moment. Inflow: the water volume data flowing into the reservoir per unit time. Precipitation prediction data: the precipitation prediction data for a future period. These data will be used as independent variables in the training dataset, and the optimal water release volume of the reservoir will be used as the dependent variable.

[0063] The training dataset will contain the following:

[0064] Independent variables (input features): the water level W(t) at the current moment, the inflow at the current moment ; the precipitation P(t) at the current moment;

[0065] Dependent variable: the optimal water release volume : the optimal water release volume calculated based on the current water level, inflow, and precipitation prediction data at this moment.

[0066] The structure of the training dataset can be expressed as: ; Each data record corresponds to the input features at a moment and the corresponding optimal water release volume.

[0067] Use the SVR (Support Vector Regression) method of SVM for regression training.

[0068] SVM regression models usually use different kernel functions (such as linear kernel, polynomial kernel, Gaussian radial basis kernel, etc.) to handle non - linear problems. The commonly used Gaussian radial basis kernel (RBF kernel) function is: ; where is the width parameter of the kernel function, which controls the measure of similarity between data points, respectively represent two different features in the training set.

[0069] Using the constructed training dataset, learn the pattern of the data through the optimization process of SVM, and find the support vectors and the regression hyperplane. The goal is to minimize the error of the regression model and find the optimal decision boundary.

[0070] The goal of the training process is to minimize the loss function: ; where: w is the weight of the model; b is the bias term; C is the regularization parameter, which is used to control the balance between the training error and the model complexity; is the error of each sample, and m is the total number of samples.

[0071] Use the support vectors (i.e., the data points that have a greater impact on the model) to train the regression model, so as to obtain the optimal prediction parameters.

[0072] Once the SVM regression model is trained, it is used to predict the optimal water release volume under the given current water level, inflow rate, and precipitation prediction data.

[0073] For each input feature through the SVM regression model make a prediction and output the corresponding optimal water release volume: ; where,

[0074] the optimal water release volume output by the SVM regression model will be used as the second input variable of the fuzzy control system. This variable, together with the predicted water level change trend value by the genetic algorithm, will be passed as inputs to the fuzzy control system.

[0075] In reservoir water level control, by combining the weighted fusion of the first input variable (future water level change trend) and the second input variable (predicted water release volume) of the fuzzy control system, the final continuous control quantities, such as the water release volume and the water release speed, can be obtained.

[0076] The inputs of the fuzzy control system include the following two main variables:

[0077] The first input variable: the predicted future water level change trend W(t + 1) obtained by the genetic algorithm, that is, the predicted value of the future water level.

[0078] The second input variable: the optimal water release volume predicted by the support vector machine SVM regression model i.e., the amount of water that needs to be released under the current conditions.

[0079] These two input variables respectively represent two key decision-making factors in water level control: the water level change trend reflects the dynamic change of the reservoir water level and provides necessary reference for future water discharge. The optimal water discharge prediction gives specific water discharge values based on the current water level, inflow rate, and precipitation prediction data.

[0080] The weighted fusion algorithm combines these two information sources by weighting the two input variables to obtain a comprehensive control quantity. The core idea of weighted fusion is to calculate the final control decision by setting appropriate weights and combining the influences of the two variables. The specific steps are as follows:

[0081] First, it is necessary to set weight coefficients for the water level change trend W(t + 1) and the predicted water discharge The weight coefficients can be optimized through experiments or historical data according to the actual situation, or set according to the rules of the fuzzy control system. Set the weights for the two inputs: is the weight of the water level change trend, is the weight of the optimal water discharge; it is required that the sum of the weight coefficients is 1.

[0082] According to the set weight coefficients, use the weighted average formula to fuse the two input variables into a comprehensive control quantity: ; where: is the comprehensive control quantity, which is a control signal calculated based on the future water level change trend and predicted water discharge. The comprehensive control quantity after weighted fusion will be used as the input of the fuzzy control system to calculate the final continuous control quantities, such as water discharge and water discharge rate.

[0083] The fuzzy control system calculates the final water discharge and water discharge rate according to the fused comprehensive control quantity The basic steps of fuzzy control include:

[0084] Fuzzification: Convert the input into a fuzzy set and define a set of fuzzy rule bases. For example: If is relatively high, the water discharge should be increased. If is relatively low, the water discharge should be decreased.

[0085] Based on the fuzzy rule base, combined with the fuzzified input data, infer a fuzzy output, representing the fuzzy decision of water discharge or water discharge rate.

[0086] Use a defuzzification method (such as the centroid method) to convert the fuzzy output into actual continuous control quantities (water discharge and water discharge rate).

[0087] After the fuzzy control system calculates the final continuous control quantity, if this control quantity (such as the water discharge of 50%) cannot be directly matched with the control equipment of the reservoir (such as gates, pumps) (for example, the equipment can only accept discrete switch commands), then further dynamic adjustment is required: evaluate the matching degree between the continuous control quantity and the discrete commands that the equipment can accept. If the matching degree is low, the control decision needs to be optimized. According to the adjustment rules of the control strategy, change the discretization accuracy of the water discharge and adjust the compensation for the equipment response delay. For example, if the equipment response delay is large, adjust the water discharge command in advance to ensure that the equipment can respond accurately at the appropriate time. Send the adjusted discrete control command to the actual control equipment of the reservoir, such as gates or pumps.

[0088] Here it should be noted that through the real-time feedback mechanism, monitor the actual water level of the reservoir and the control effect. If it is found that the change in the actual water level does not match the expectation, the system will automatically adjust the fuzzy control rules and weight coefficients to optimize the water level management strategy of the reservoir.

[0089] In the reservoir control system, the fuzzy control system usually outputs a continuous control quantity (such as the water discharge of 50%), while the actual control equipment (such as gates, pumps, etc.) can only accept discrete control commands (such as "open 50%"). Therefore, the continuous control quantity must be converted into discrete control commands that the equipment can understand and be matched according to the response delay of the equipment.

[0090] First, convert the continuous control quantity (such as the water discharge of 50%) output by the fuzzy control system into discrete control commands that the equipment can execute. Since the equipment can only accept discrete states (such as "open 50%" or "closed"), this requires discretizing the continuous output of the fuzzy control. The specific steps are as follows: Define the range of discrete control commands: Assume that the control commands that the equipment can accept have multiple discrete states, such as: the gate is fully closed (0%); the gate is partially open (such as 20%, 40%, 60%, 80%); the gate is fully open (100%).

[0091] Discretize the control quantity: Match the continuous control quantity (such as 50%) output by the fuzzy control system with the discrete commands of the equipment. Commonly used discretization methods are the rounding method or the interval division method:

[0092] If the continuous control quantity is 50%, round it to the nearest discrete command, that is, 50%. If the control quantity is other values (such as 45%), it may be rounded to 40% or 50%.

[0093] For example: control quantity 50% → discrete command: 50%; control quantity 65% → discrete command: 60%; control quantity 37% → discrete command: 40%.

[0094] The interface matching degree refers to the matching degree between the continuous output of the fuzzy control system and the discrete commands that the device can accept. To ensure that the device accurately executes the control commands, it is necessary to evaluate the matching degree between the continuous control quantity and the discrete commands.

[0095] The discretization precision of the control command refers to the precision when converting the continuous control quantity of the fuzzy control system into discrete commands that the device can execute. A higher discretization precision means that the control state of the device (such as the water discharge amount, the opening degree of the gate) can be adjusted meticulously. For example, the control quantity is discretized with a smaller step size (such as adjusting every 5%). However, the higher the precision, the frequent changes of the control commands caused by the fine adjustments executed by the device may increase the response burden of the device and even cause unnecessary operation fluctuations. Therefore, the discretization precision needs to be balanced according to the actual response ability and operation stability of the device to ensure that the control commands are precise enough and can avoid the adverse effects brought by frequent switching.

[0096] After analyzing the precision when converting the continuous control quantity of the fuzzy control system into discrete commands executed by the device, a discrete command control precision index is generated. The method for obtaining the discrete command control precision index is as follows:

[0097] Express the continuous control quantity (such as 50%) output by the fuzzy control system as .

[0098] Express the actual discrete control command executed by the device (such as "open 50%") as , which is usually obtained by discretizing the continuous control quantity.

[0099] Calculate the error of each discrete command: The error of the discrete command is calculated from the difference between the continuous control quantity and the command executed by the device, and the expression is: ; if the error is 0, it means that the continuous control quantity and the discrete command are completely matched. If is greater than 0, it means that there is an error and the system fails to match precisely.

[0100] Since different discretized commands may have different degrees of influence on the system, a weight coefficient needs to be assigned to each discretization error to reflect the importance of the command. Usually, the weight coefficient can be set according to the response sensitivity of the device, the change range of the control quantity, or the operation priority.

[0101] Weighted error calculation: By the weighted error method, calculate the weighted error value , and the expression is: ; where: z is the number of discretized control commands, is the weight coefficient of the i-th control command, is the error of the i-th discrete command. Calculate the discrete command control accuracy index, and the calculation expression is: ; where: is the discrete command control accuracy index, which reflects the system control accuracy. The larger the value, the higher the accuracy.

[0102] The device response delay refers to the time required for the device to actually execute a control command after the command is issued. When the response delay is large, there may be a deviation between the immediate control quantity of the fuzzy control system and the actual effect during device execution. For example, if the water discharge amount output by the system is 50%, but the response delay of the device is 1 minute, the actual water discharge amount during operation may not be consistent with the system expectation, thus affecting the accuracy of water level control. To compensate for the impact of response delay on the control effect, the system can adjust the control decision in advance according to the delay time of the device to ensure that the device adjusts its state as expected. In addition, a long response delay may require the system to design more flexible and robust control strategies to reduce the negative impact of the delay on the overall control effect.

[0103] After analyzing the response delay situation of the time required for the device to actually execute the command after the control command is issued, generate the execution response delay anomaly index. The acquisition method of the execution response delay anomaly index is:

[0104] Collect the historical response delay data of the device, and evaluate the anomaly situation by calculating the similarity between the current response delay and the historical moment. First, select k nearest neighbors (k similar moments) from the historical response delay data, and then calculate the difference between the response delay of the neighboring moments and the current response delay, and use the Euclidean distance to calculate the difference between the current response delay and the historical delay data , and the expression is: ; where: is the historical moment (i = 1, 2,..., k). is the response delay at the current moment, is the historical moment 's response delay.

[0105] According to the calculated distance, select the k historical moment response delay values with the smallest distance as the nearest neighbors. The value of k can be determined by the following method:

[0106] If the historical data is relatively stable, a smaller value of k (such as 3 or 5) can be selected. If the historical data changes greatly, a larger value of k (such as 10 or 15) can be selected to smooth out abnormal fluctuations.

[0107] Calculate the average response delay of the selected k nearest neighbors, and the expression is: ; where: is the average value of the historical response delay at the current moment t, and k is the number of selected neighbors; calculate the execution response delay anomaly index, and the expression is: ; where: is the standard deviation of the historical response delay data (measuring the volatility of the device response delay), and ERD is the execution response delay anomaly index.

[0108] According to the calculated anomaly index ERD, determine whether there is an execution response delay anomaly. Common judgment methods include: if ERD > 2, it means that the response delay at this moment is abnormal, which may lead to a deviation between the control strategy and the device execution. If ERD ≤ 1, it is considered that the response delay of the current device is within the normal range.

[0109] Normalize the discrete command control accuracy index and the execution response delay anomaly index so that they are both within [0, 1], and calculate the interface matching degree between the continuous control quantity and the discrete command accepted by the device according to the normalized discrete command control accuracy index and the execution response delay anomaly index.

[0110] For example, the present invention can use the following formula to calculate the interface matching degree between the continuous control quantity and the discrete command accepted by the device, and the calculation expression is: ; In the formula, is the interface matching degree between the continuous control quantity and the discrete command accepted by the device, is the discrete command control accuracy index, ERD is the execution response delay anomaly index, are the weight coefficients of the discrete command control accuracy index and the execution response delay anomaly index (which can be optimized according to experimental experience or machine learning), and are all greater than 0.

[0111] Compare the obtained interface matching degree between the continuous control quantity and the discrete command accepted by the device with the preset matching degree reference threshold. If the interface matching degree between the continuous control quantity and the discrete command accepted by the device is greater than or equal to the preset matching degree reference threshold, it means that the interface matching degree between the continuous control quantity and the discrete command accepted by the device is high, and at this time, a matching normal signal is generated; if the interface matching degree between the continuous control quantity and the discrete command accepted by the device is less than the preset matching degree reference threshold, it means that the interface matching degree between the continuous control quantity and the discrete command accepted by the device is low, and at this time, a matching abnormal signal is generated.

[0112] When the evaluation result shows that the interface matching degree between the continuous control quantity and the discrete command accepted by the device is low, that is, the matching degree does not reach the preset reference threshold, it is necessary to continuously adjust and optimize the control decision. This mainly includes adjusting the discretization rule and the water release strategy to ensure that the device can better respond to the commands of the control system.

[0113] When the matching degree is low, it is first necessary to analyze the reasons. The possible reasons include but are not limited to the following points:

[0114] The control precision of discrete commands is not high, resulting in a large error between refined control and the actual execution of the device. The response delay of the device is relatively long and fails to keep up with the instruction changes of the fuzzy control system in a timely manner. The discretization rules do not match the control capabilities of the device. For example, the device cannot accurately respond to overly refined control commands (e.g., "discharge 50% of the water" cannot be directly translated into the actions of the device).

[0115] The discretization rule is the standard for converting the continuous control quantity of the fuzzy control system into discrete commands that the actual device can accept. The discretization rule can be adjusted in the following ways:

[0116] If the response ability of the device is limited, it may be necessary to adjust the discretization rule so that the refinement degree of the control command adapts to the execution ability of the device. For example: Adjust "discharge 50% of the water" to an interval of "the discharge volume is 50% or 40%", thereby reducing the complexity of the device control command. Adopt a more relaxed water discharge strategy, such as increasing the discretization accuracy from 10% to 20% or 30%, and reducing the frequency of each adjustment of the device.

[0117] According to the actual execution ability of the device, a dynamic adjustment strategy can be adopted. For example:

[0118] When the water level changes rapidly, adopt a lower discretization accuracy to make the device respond more quickly; when the water level is relatively stable, the discretization accuracy can be increased to achieve more detailed control. Use a real-time feedback mechanism to dynamically adjust the discretization rule so that the system can better match the device under different conditions.

[0119] Adjusting the water discharge strategy can make the water discharge of the reservoir more efficient and reduce the difference from the device control ability. The water discharge strategy can be optimized in the following ways:

[0120] Based on historical data and real-time prediction, redefine the water discharge range of the reservoir to ensure that the device can perform precise control within this range. For example: If the current water level of the reservoir is high and strong precipitation is predicted in the next period of time, the upper limit of the water discharge can be increased so that the device can effectively lower the water level. In the case of a low water level in the reservoir, appropriately reduce the water discharge to avoid waste of resources caused by excessive water discharge.

[0121] Based on real-time data (such as the current water level, precipitation, inflow, etc.), adopt a dynamic water discharge strategy to cope with different flood seasons and drought periods. The dynamic strategy includes: automatically adjusting the water discharge frequency and magnitude so that the device can adapt according to the actual situation. Introduce a multi-level water discharge strategy (such as small, medium, and large water discharges) to cope with different situation changes.

[0122] After implementing the adjusted control decision, the system needs to further optimize and adjust the control strategy based on real-time feedback information: The current state of the reservoir, including key data such as water level and flow rate, is obtained in real time through sensors and monitoring systems. The system can compare this data with the prediction model to adjust the water discharge in real time. By comparing with the actual execution results (equipment responses), the discretization rules and water discharge strategies are continuously optimized: If it is found that the equipment cannot accurately respond at a certain stage, the discretization accuracy can be further optimized or the interval of the water discharge strategy can be adjusted. According to the actual performance of the equipment response delay, the time window of the control decision is adjusted to ensure an improved matching degree between the continuous control quantity and the actual operation.

[0123] After the discretization rules and water discharge strategies are adjusted and optimized, the finally generated control commands need to be sent to the actual control equipment of the reservoir (such as gates, pumps, etc.) for execution. At this time, the control commands have been optimized according to the equipment response capabilities and interface matching degrees, ensuring that the equipment can execute accurately and achieve the expected effects.

[0124] As described above, it is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed by this application can easily think of changes or substitutions, which should all be covered within the protection scope of this application.

Claims

1. A multi-objective decision-making control method for the stage flood season water level of a reservoir with a balance between risks and benefits, characterized in that: Including: Determine the input items affecting the reservoir water level control, including the current water level, precipitation, inflow rate, and reservoir water storage volume; Input historical water level data, inflow rate, and precipitation data, and optimize the water level change prediction and optimal water discharge prediction by combining the genetic algorithm and the support vector machine model; Transfer the prediction results as input items to the fuzzy control system, calculate the final continuous control quantities, including the water discharge volume and the water discharge speed, after using the weighted fusion algorithm, and convert 50% of the water discharge volume output by the fuzzy control system into the 50% opening state of the device; Evaluate the interface matching degree between the continuous control quantity and the discrete command accepted by the device according to the discretization accuracy of the control command and the device response delay situation. Specifically, it includes: normalizing the discrete command control accuracy index and the execution response delay anomaly index so that they are both within [0,1], and calculating the interface matching degree between the continuous control quantity and the discrete command accepted by the device according to the normalized discrete command control accuracy index and the execution response delay anomaly index; The method for obtaining the discrete command control accuracy index is: Express the continuous control quantity output by the fuzzy control system as ; The actual discrete control commands executed by the device are represented as ; Calculate the error of each discrete command, the error of the discrete command is calculated from the difference between the continuous control quantity and the device execution command, and the expression is: ; Calculate the weighted error value by the weighted error method , and the expression is: ; where: z is the number of discretized control commands, is the weight coefficient of the i-th control command, is the error of the i-th discrete command, calculate the discrete command control precision index, and the calculation expression is: ; where: is the discrete command control precision index; The method for obtaining the execution response delay anomaly index is: Collect the historical data of the response delay of the collection device, select k nearest neighbors from the historical response delay data, and then calculate the difference between the response delay at the neighboring moment and the response delay at the current moment, and use the Euclidean distance to calculate the difference between the current response delay and the historical delay data , the expression is: ; where: is the historical moment, is the response delay at the current moment, is the historical moment of the response delay; Based on the calculated distance, select the k historical moment response delay values with the smallest distance as the nearest neighbors, and calculate the average response delay of the selected k nearest neighbors. The expression is: ; where: is the average value of the historical response delays at the current moment t, and k is the number of selected neighbors; calculate the execution response delay anomaly index. The expression is: ; where: is the standard deviation of the historical response delay data, is the execution response delay anomaly index; If the matching degree is low, continuously adjust and optimize the control decision according to the evaluation results, including adjusting the discretization rule and the water discharge strategy, and sending the adjusted discrete control command to the actual control device of the reservoir to complete the reservoir water discharge volume.

2. The multi-objective decision-making control method for the stage flood season water levels of a reservoir with a balance between risks and benefits according to claim 1, characterized in that: Use the genetic algorithm for multi-generation optimization to predict the future water level change and output the predicted water level change trend value. Specifically, collect historical water level, inflow rate, and precipitation data, and construct a training data set. The independent variables of the data set include the water level, inflow rate, and precipitation data at several past moments, and the dependent variable is the predicted water level at the future moment; Adopt the genetic algorithm for multi-generation optimization, and find the optimal solution by simulating the natural selection process. Specifically, it includes: initializing the population and randomly generating multiple candidate solutions, using the mean square error MSE as the fitness function, selecting individuals with high fitness for crossover and mutation operations to generate a new generation of individuals, and stopping the optimization process through the termination condition. Finally, obtain a set of optimal parameters; use the optimal parameters to predict the future water level change trend as the first input variable of the fuzzy control system.

3. The multi-objective decision-making control method for the reservoir's staged flood-season water levels with a balance between risks and benefits according to claim 2, characterized in that: Use the SVM regression model to predict the optimal water discharge volume under different conditions and output the predicted water discharge volume value. Specifically, train the model through the support vector regression method in the support vector machine SVM; during the training process, collect the current water level, inflow rate, and precipitation prediction data as independent variables, and the optimal water discharge volume of the reservoir as the dependent variable; by constructing a training data set, the input features include the water level, inflow rate, and precipitation at the current moment, and the target output is the optimal water discharge volume; use the Gaussian radial basis kernel function to process nonlinear problems, optimize the SVM regression model, minimize the loss function, and find the optimal decision boundary and regression hyperplane; after the training is completed, the SVM regression model predicts the optimal water discharge volume based on the current water level, inflow rate, and precipitation prediction data, and uses the prediction result as the second input variable of the fuzzy control system.

4. The multi-objective decision-making control method for the reservoir's staged flood-season water levels with risk-benefit balance according to claim 3, characterized in that: In the reservoir water level control, the final continuous control quantity is obtained by weighted fusion of the first input variable and the second input variable of the fuzzy control system. Specifically, the first input variable is the future water level change trend predicted by the genetic algorithm , that is, the predicted value of the future water level; the second input variable is the optimal water discharge predicted by the support vector machine (SVM) regression model , that is, the amount of water that needs to be discharged under the current conditions; For the water level change trend and the predicted water discharge set the weight coefficients, where is the weight of the water level change trend, and The sum of the weight coefficients is 1; according to the set weight coefficients, two input variables are fused into a comprehensive control quantity using the weighted average formula: ; where: is the comprehensive control quantity, which is a control signal calculated based on the future water level change trend and the predicted water discharge volume.

5. The multi-objective decision-making control method for the stage flood season water levels of a reservoir with a balance between risks and benefits according to claim 4, characterized in that: The fuzzy control system calculates the final water discharge volume and water discharge speed according to the fused comprehensive control quantity including: Convert the input into a fuzzy set and define a set of fuzzy rule bases; According to the fuzzy rule base, combined with the fuzzified data of the input, infer a fuzzy output, representing the fuzzy decision of the water discharge amount or the water discharge speed; Use the defuzzification method to convert the fuzzy output into an actual continuous control quantity.

6. The multi-objective decision-making control method for the reservoir's staged flood-season water levels with risk-benefit balance according to claim 1, characterized in that: Compare the interface matching degree between the obtained continuous control quantity and the discrete command accepted by the device with the preset matching degree reference threshold. If the interface matching degree between the continuous control quantity and the discrete command accepted by the device is greater than or equal to the preset matching degree reference threshold, it indicates that the interface matching degree between the continuous control quantity and the discrete command accepted by the device is high. At this time, generate a normal matching signal; If the interface matching degree between the continuous control quantity and the discrete command accepted by the device is less than the preset matching degree reference threshold, it indicates that the interface matching degree between the continuous control quantity and the discrete command accepted by the device is low. At this time, generate an abnormal matching signal, and it is necessary to adjust the discretization rule and the water discharge strategy.

Images