Real-time risk dispatching method and system for large-scale complex flood control system in river basin
Through the MA-parzen window method and gradient method driven scenario network reduction model, combined with Martin3-Zhukov's WBS dual-layer model and multi-task parallel technology, the problems of flood forecast uncertainty and high-dimensional complexity in complex flood control systems are solved. It realizes efficient and accurate decision-making in real-time multi-objective risk scheduling.
Patent Information
- Application Number
- CN202411140808.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-20
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-08-20
AI Technical Summary
In real-time flood control scheduling of complex flood control systems, due to the uncertainty of flood forecast and high-dimensional complexity, it is difficult to meet the real-time and accuracy requirements of flood control scheduling.
The probability density function of flood forecast error is estimated by MA-parzen window method, and a gradient-driven scenario network reduction model is constructed. The system decomposition and dimensionality reduction are performed through the Martin3-Zhukov's WBS two-layer model, combining MATLAB distributed clustering and CPU parallel technology. Using the perfect nested loop framework and Debye Waller factor optimization multi-objective mayfly algorithm, the clustering method of coupling system target decision-physical attributes-spatial geometric features is extracted, and a real-time multi-objective risk scheduling scheme was obtained.
It reduces the risks brought about by hydrological uncertainty, improves the timeliness and accuracy of flood control scheduling, meets the real-time risk scheduling needs of complex flood control systems, optimizes computing efficiency, and provides a more accurate flood control scheduling solution.
Smart Images

Figure CN118657356B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of flood control dispatching, and in particular to a real-time risk dispatching method and system for a large-scale complex flood control system in a river basin. Background Art
[0002] Real-time flood control dispatching of complex flood control systems is one of the important technical means of flood control and disaster reduction in river basins. It can improve the benefits of flood control projects with relatively small investments and significantly reduce flood disasters. Various uncertainties in the dispatching process bring risks to flood control dispatching, affecting real-time flood control dispatching decisions. Among them, flood forecasting is the main source of risk. There are many flood forecasting simulation methods, and the random simulation method is currently the most commonly used method. According to its probability distribution, sample points with full coverage of the interval are randomly selected, which significantly improves the error accuracy and is applied to the field of flood control by more and more people. Due to the different choices of random simulation methods, the accuracy of the error between the predicted flood and the measured flood is different, resulting in uncertainty in hydrological forecasting. It is crucial to find a high-precision random simulation method for random sampling in the flood forecasting process.
[0003] Real-time flood control dispatching of complex flood control systems is a complex multi-objective problem. The mutual constraints between multiple objectives make it difficult to quickly obtain the optimal solution for complex system dispatching, which brings inconvenience to flood control dispatching. The solution of multi-objective optimization model has been transformed from traditional optimization algorithm to intelligent optimization algorithm, which can search for global optimal solution flexibly and stably, and has strong adaptability and can be applied to multi-objective solution problems in various fields.
[0004] In the real-time risk scheduling of large-scale complex flood control systems in river basins, the coupling and nesting of random simulation methods and intelligent optimization algorithms brings high-dimensional and complex dimensionality curse problems to real-time scheduling operations, resulting in a huge amount of calculation in the scheduling process, which makes it difficult to meet the real-time nature of flood control scheduling. Summary of the invention
[0005] Purpose of the invention: To provide a real-time risk dispatching method for a large-scale complex flood control system in a river basin to solve the above-mentioned problems existing in the prior art.
[0006] Technical solution: A real-time risk dispatching method for a large-scale complex flood control system in a river basin, including the following steps:
[0007] Step S1, collecting flood data and engineering characteristic data of various flood control projects in the complex flood control system, determining that flood forecast error is a risk factor in flood control scheduling, using the MA-parzen window method to estimate the probability density function of the flood forecast error, obtaining distribution parameters, generating an initial error scenario set based on Monte Carlo random simulation of flood forecast error and inputting it into a pre-built scenario network reduction model, using the gradient method to drive the model, and obtaining a reduced error scenario set;
[0008] Step S2, constructing Martin3-Zhukov's WBS two-layer model and performing system decomposition and dimensionality reduction on the complex flood control system to obtain Z subsystems, constructing an objective function library and a constraint condition library based on the subsystems, using a weight adjustment method with dynamic criterion elimination to determine the weight of each objective of the subsystem, and converting the linear weighting into an overall objective function for evaluating the safety of the subsystem, and constructing a real-time multi-objective risk scheduling model for a large-scale complex flood control system in a river basin;
[0009] Step S3, using a perfect nested loop framework to construct a multi-task parallel module of forecast error scenario set, iteration number and population, and coupling MATLAB distributed cluster and CPU parallel technology to optimize the real-time multi-objective risk scheduling model of a large-scale complex flood control system in a basin, superimposing the simplified error scenario set on the forecast inflow of each flood control project and inputting the optimized real-time multi-objective risk scheduling model of a large-scale complex flood control system in a basin, using an improved multi-objective mayfly algorithm based on Debye Waller factor optimization to solve, and obtaining a real-time flood control scheduling solution set corresponding to the simplified error scenario set by the iteration number population;
[0010] Step S4, extract the typical scheduling scheme corresponding to the simplified error scenario set based on the clustering method coupled with system goal decision-making-physical attributes-spatial geometric characteristics, make decisions on the real-time flood control scheduling scheme set based on the typical scheduling scheme, and select the optimal scheme as the real-time risk scheduling scheme for the large-scale complex flood control system in the basin.
[0011] According to one aspect of the present application, step S1 further comprises:
[0012] Step S11, determine the study area, collect geographic spatial data and water system data of the study area, characteristic data of various flood control projects, hydrological data of flood control points and flood storage areas, and organize the historical flood process and forecast flood process of various flood control projects;
[0013] Step S12, determining flood forecast error as a risk factor for flood control scheduling, estimating the probability density function of flood forecast error of each flood control project according to the MA-parzen window method, determining the distribution characteristics of flood forecast error, and obtaining forecast error distribution parameters;
[0014] Step S13, based on the flood forecast error distribution characteristics and parameters of the flood control project, the Monte Carlo method is used to randomly simulate the forecast error to generate an initial forecast error scenario set;
[0015] Step S14: construct a scenario network reduction model driven by the gradient method, input the initial forecast error scenario set into the scenario network reduction model, and generate a reduced error scenario set.
[0016] According to one aspect of the present application, step S14 is further:
[0017] Step S14a, using the forecast error of the Monte Carlo random simulation in step S13 as the initial node of the reduced scenario network;
[0018] Step S14b, evaluate the prior probability of the node according to the MA-parzen window method, and determine the position of the network node layer by layer, where 1-0.8 is a first-level node, 0.8-0.6 is a second-level node, ..., 0.2-0 is a fifth-level node; calculate the conditional probability between adjacent level nodes, select the node with a large conditional probability value as the node branch, gradually establish the network branch between nodes, and construct an optimized network structure;
[0019] Step S14c, using the gradient descent method to calculate the position of the current research node under a certain branch of the network, the congestion degree between the coupling node prior probability and the error, and the product of the three constitutes the evaluation parameter;
[0020] Step S14d, based on the Bayesian principle, the posterior scenario probability is calculated, the position of the current research node under a certain branch of the network is calculated using the gradient ascent method, and the diagnostic parameters between each node are determined by multiplying the three together in combination with the congestion degree;
[0021] Step S14e, respectively arranging the evaluation and diagnosis parameters in descending order, and eliminating the bottom a% of the rankings of the two, evaluating the representativeness and coverage of the retained nodes; if the evaluation passes, retain these nodes and their network branches, otherwise discard the nodes, determine the nodes that meet the conditions to build a simplified scenario network, and use MA-parzen to fit the scenario distribution to obtain the distribution parameters; a is the set value;
[0022] Step S14f, repeating steps S14b-S14e, so that the reduced scenario network continues to evolve until the absolute error between the evolved reduced scenario and the initial scenario distribution parameter is greater than a preset value, then stopping the evolution and obtaining a reduced error scenario set.
[0023] According to one aspect of the present application, step S2 further comprises:
[0024] Step S21, constructing a Martin3-Zhukov's WBS two-layer model and decomposing the complex flood control system by dimensionality reduction based on the Martin3-Zhukov's WBS two-layer model to obtain Z collaborative subsystems, where Z is a positive integer greater than 2;
[0025] Step S22: construct the objective function library of each subsystem respectively, use the weight adjustment method of dynamic criterion elimination to evaluate the weight of each objective, and convert the linear weighted method into the total objective function for evaluating the safety of the subsystem;
[0026] Step S23, respectively constructing a constraint library of the real-time flood control risk dispatch model of each subsystem, including: water balance constraint, reservoir discharge constraint, reservoir characteristic water level constraint, initial / final water level control constraint of the dispatch period, reservoir discharge amplitude constraint, flood storage and detention area water balance constraint, flood storage and detention area capacity constraint and flood control point safety constraint;
[0027] Step S24: Based on the overall objective function and constraint condition library, a real-time multi-objective risk scheduling model for a large-scale complex flood control system in a river basin is constructed.
[0028] According to one aspect of the present application, step S21 is further:
[0029] Step S21a, generalizing the watershed and distinguishing the geographical space of the complex flood control system, and determining the number of subsystem targets to be decomposed;
[0030] Step S21b, partition the complex flood control basin system according to the spatial geographical location. Due to the irregularity of the basin boundary, the smaller the partition, the better, but it should not be less than the number of flood control projects. Zhukov's method is used to gradually cluster adjacent partitions. According to the principle of minimizing the area difference between subsystems, the large-scale complex flood control system in the basin is decomposed into Z subsystems, where Z is a positive integer greater than 2;
[0031] Step S21c, continue to decompose the subsystem according to the attributes of the flood control project, and encode the flood control project to establish an index system;
[0032] Step S21d, using the Martini3 model to analyze the dependencies between flood control projects in the subsystem based on geographical location, project attributes, and target effectiveness. The closer the dependencies, the greater the probability that they are a subsystem.
[0033] Step S21e, check the rationality of the decomposition of the subsystem in the above step S21b, determine the members of each subsystem in the complex flood control basin, if the subsystem decomposition is incorrect, in order to coordinate and balance the flood control tasks and reasonably allocate flood control projects, use the shortest Euclidean distance method to re-assign the flood control projects to the subsystem, determine the final Z subsystem internal structures, and count the number of reservoirs, flood control points and flood storage areas in the subsystem.
[0034] According to one aspect of the present application, step S22 is further:
[0035] Step S22a, constructing the target function library of each subsystem of large-scale complex flood control in the basin, including: flood control safety of reservoir groups within the subsystem, safety of downstream flood protection areas and safety of flood storage and detention areas;
[0036] Step S22b: Determine the weight of each target using the weight adjustment method of dynamic criterion elimination, and construct the total objective function of the subsystem by linear weighting.
[0037] According to one aspect of the present application, step S3 further comprises:
[0038] Step S31, using a perfect nested loop framework to build a multi-task parallel module of forecast error scenario sets, iteration times and populations, decomposing and reducing the flood control scheduling process from multiple angles, and optimizing the real-time multi-objective risk scheduling model of a large-scale complex flood control system in a basin by combining MATLAB distributed clusters and CPU parallel technology;
[0039] Step S32, extract the simplified error scenario set and superimpose it on the forecast inflow of each flood control project, and input the obtained data into the optimized real-time multi-objective risk scheduling model of the large-scale complex flood control system in the basin, and use the improved multi-objective mayfly algorithm based on Debye Waller factor optimization to solve it, so as to obtain the real-time flood control scheduling plan set corresponding to the simplified error scenario set.
[0040] According to one aspect of the present application, step S31 is further:
[0041] Step S31a, building a MATLAB distributed cluster and determining the number of CPU cores of the nodes and cluster;
[0042] Step S31b, compiling a multi-task parallel module consisting of a forecast error scenario set, an iteration number, and a population nested loop;
[0043] Step S31c: To ensure load balancing of scheduling parallelism, the forecast error scenario set of the outermost loop is allocated to the cluster nodes;
[0044] Step S31d, based on the forecast error scenario set allocated in step S31c, allocating the number of MA iterations of the middle layer to different threads in each node;
[0045] Step S31e: Using the threads allocated in step S31d, parallelly calculate the MA population within the range of MA iteration times corresponding to each thread.
[0046] According to one aspect of the present application, step S32 is further:
[0047] Step S32a, taking the outflow of the reservoir as the initial population, introducing a sliding operator to perform sliding average processing on the initial population to ensure the smoothness of the outflow flow process;
[0048] Step S32b, calculate the system objective function, evaluate the fitness functions of all populations, and introduce external archives to screen non-inferior solution sets;
[0049] Step S32c, determining the iteration condition, if not satisfied, updating the speed and position of the female and male mayflies;
[0050] Step S32d, evaluate the solution, if it is a non-dominated solution, update the archive, otherwise there is a 50% probability of updating the archive;
[0051] Step S32e, introducing the Debye Waller factor to perform stable sorting on the mayflies, mating and evaluating the offspring, and determining whether the offspring is a non-dominated solution of its parent of the same sex, and if so, storing the non-dominated solution in an external archive;
[0052] Step S32f, satisfying the iteration condition, retaining the final non-dominated solution set as the real-time flood control scheduling solution set.
[0053] According to one aspect of the present application, the clustering method based on the coupling of system target decision-physical attribute-spatial geometric feature in step S4 extracts the typical scheduling scheme corresponding to the reduced error scenario set, further comprising:
[0054] Step S41: based on the flood control objectives of the Z subsystems of large-scale complex flood control in the basin, a set of scheduling schemes is preferentially selected to obtain a set of decision solutions for each subsystem;
[0055] Step S42, dividing the physical properties of the solution set according to the fact that the intra-cluster spacing of the spatial point set is as small as possible and the inter-cluster spacing is as large as possible, and using Fuzzy·C-Means-Clustering to divide the decision solution set of each subsystem into M categories;
[0056] Step S43, using an adaptive curvature entropy point cloud simplification algorithm to calculate the curvature entropy in the M-type solution set, removing redundant points, keeping the spatial characteristics of the solution set unchanged, and extracting 0.1% of the representative point set from the M-type solution set;
[0057] Step S44, the point set finally extracted is a typical scheduling solution based on clustering decision-making coupled with system goal decision-physical attribute-spatial geometric feature.
[0058] According to another aspect of the present application, a real-time risk dispatching system for a large-scale complex flood control system in a river basin is provided, comprising:
[0059] at least one processor;
[0060] and a memory communicatively coupled to at least one of the processors;
[0061] Among them, the memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the real-time risk scheduling method for a large-scale complex flood control system in a river basin as described in any of the above technical solutions.
[0062] Beneficial effects: The scenario network simplification model driven by the gradient method proposed in the present invention efficiently and accurately reduces the forecast error scenario set, reducing the risk brought by hydrological uncertainty to the flood control system. The model is decomposed and reduced in dimension by constructing a multi-task parallel module of forecast error scenario set, number of iterations and population through a perfect nested loop framework, and the "dimensionality disaster" problem in the real-time risk scheduling system of large-scale and complex flood control in the basin is solved by coupling MATLAB distributed cluster and CPU parallel technology, meeting the timeliness requirements of flood control scheduling. The clustering method of system goal decision-physical property-spatial geometric feature coupling is adopted to obtain a more accurate and practical flood control scheduling plan, and provide a basis for flood control scheduling decision-making. BRIEF DESCRIPTION OF THE DRAWINGS
[0063] Figure 1 It is a flow chart of the present invention.
[0064] Figure 2 This is a flow chart of the scenario network simplification model driven by the gradient method of the present invention.
[0065] Figure 3 This is a flow chart of cluster parallel computing based on multi-task perfect nested loops of the present invention. DETAILED DESCRIPTION
[0066] The following is a detailed description of the technical solution of the present invention by taking the Huaihe River Basin complex flood control system as an example and combining with the accompanying drawings.
[0067] like Figure 1 As shown, the following technical solution is proposed.
[0068] According to one aspect of the present application, a real-time risk dispatching method for a large-scale complex flood control system in a river basin is provided, characterized in that it comprises the following steps:
[0069] Step S1, collecting flood data and engineering characteristic data of various flood control projects in the complex flood control system, determining that flood forecast error is a risk factor in flood control scheduling, using the MA-parzen window method to estimate the probability density function of the flood forecast error, obtaining distribution parameters, generating an initial error scenario set based on Monte Carlo random simulation of flood forecast error and inputting it into a pre-built scenario network reduction model, using the gradient method to drive the model, and obtaining a reduced error scenario set;
[0070] Step S2, construct Martin3-Zhukov's WBS two-layer model and perform system decomposition and dimensionality reduction on the complex flood control system to obtain Z subsystems, build an objective function library and a constraint condition library based on the subsystems, use the weight adjustment method of dynamic criterion elimination to determine the weight of each subsystem objective, and convert the linear weighted transformation into the total objective function for evaluating the safety of the subsystem, and build a real-time multi-objective risk scheduling model for a large-scale complex flood control system in a basin; wherein WBS is Work Breakdown Structure, which means work breakdown structure.
[0071] Step S3, optimizing the real-time multi-objective risk scheduling model of a large-scale complex flood control system in a basin based on cluster parallel technology of multi-task perfect nested loops, superimposing the simplified error scenario set on the forecast inflow of each flood control project and inputting the optimized real-time multi-objective risk scheduling model of a large-scale complex flood control system in a basin, solving it by using an improved multi-objective mayfly algorithm based on Debye Waller factor optimization, and obtaining a real-time flood control scheduling solution set corresponding to the simplified error scenario set by the number of iterations and the algorithm population;
[0072] Step S4, extract the typical scheduling scheme corresponding to the simplified error scenario set based on the clustering method coupled with system goal decision-making-physical attributes-spatial geometric characteristics, make decisions on the real-time flood control scheduling scheme set based on the typical scheduling scheme, and select the optimal scheme as the real-time risk scheduling scheme for the large-scale complex flood control system in the basin.
[0073] Traditional probability density function estimation mainly relies on specific distribution assumptions. The assumed distribution is easy to deviate from the actual data and cannot capture the distribution characteristics of new data in real time, resulting in inaccurate estimated probability density functions. The present invention introduces the Parzen window method to evaluate historical data, quantitatively simulates the distribution of errors, and uses the Mayfly optimization algorithm to optimize the bandwidth. It can adaptively find the optimal bandwidth that meets the historical data and avoid the problem of overfitting or underfitting of the bandwidth. It is particularly convenient for processing high-dimensional complex data, especially in the scheduling process where there are many uncertain factors and complex relationships. The MA-Parzen window method can accurately evaluate the distribution of hydrological uncertainty and provide a solid foundation for subsequent risk analysis.
[0074] There are many methods for random simulation at present, but most of them only consider the requirements of simulation accuracy and ignore the limitation of the number of models. Too many random simulation scenario sets will cause computational complexity and increase the operating load of the scheduling system. The present invention introduces a scenario simplification model to continuously reconstruct the scenario network by updating nodes, and proposes a scenario network simplification model driven by a gradient method, which achieves the effect of reducing the scenario set on the basis of meeting the simplification accuracy, and ensures the randomness and uniform distribution of the simplified scenario set. This simplification method not only reduces the complexity of uncertainty simulation calculations, but also dynamically changes and flexibly adjusts the network simplification scenarios, improves the operating efficiency and accuracy of the model, increases the practicality of random simulation of risk scheduling, and facilitates the optimization decision of risk scheduling.
[0075] The complex flood control system has a complex engineering structure and close hydraulic connections. The complexity of the system seriously affects the management and operation of flood control scheduling in the basin. In order to better solve this problem, the Martin3-Zhukov's-WBS two-layer model is proposed to roughly divide the subsystems of the complex flood control basin system from the outside to the inside and the detailed attribution of flood control projects. Martin3 is used to evaluate and test the initially divided subsystems, and the subsystems are coordinated and scheduled with the overall system safety as the goal, so as to rationalize the decomposition process of the complex basin flood control system. This system decomposition method divides the attributes of flood control projects in a forward and progressive manner, and then reversely recursively tests the rationality of the subsystem division. Finally, the complex flood control problem is divided into smaller and easier to handle parts, which doubles the orderliness of the complex system scheduling tasks and lays the foundation for the subsequent clear construction of a real-time multi-objective risk scheduling model for a large-scale complex flood control system in the basin and the improvement of the overall scheduling efficiency of the flood control system.
[0076] Large-scale and complex flood control dispatch in the basin is facing an increasingly serious "dimensionality disaster" problem, which causes high-dimensional complexity calculation problems and has always been a difficulty in the flood control dispatch process, seriously affecting the timeliness of flood control dispatch and not conducive to real-time and accurate decision-making of dispatch plans. In response to this problem, this application constructs an error scenario set-intelligent algorithm iteration number-intelligent algorithm population multi-task layered nested loop calculation model to clarify the risk dispatch model process and facilitate the technical dimensionality reduction processing of the dispatch model.
[0077] In order to further optimize and solve the high-dimensional computing problems caused by the coupling of hydrological uncertainty random simulation and intelligent algorithm populations, MATLAB distributed cluster parallel computing scheduling tasks are introduced. In order to optimize the parallel structure, the population in the multi-task perfect nested loop framework is processed in parallel. From a technical perspective, structural dimensionality reduction is performed on flood control scheduling in complex flood control basins to improve the scheduling efficiency, which is of great significance to meeting the real-time nature of flood control scheduling decisions.
[0078] Due to the layered coupling of uncertain random simulation, iteration number and intelligent algorithm population, many flood control scheduling solution sets are generated, which brings inconvenience to real-time flood control scheduling decision-making and risk analysis. How to quickly screen the optimal solution from the numerous solution sets generated and ensure the practical significance and high coverage of the optimal solution is crucial for the decision-making of the scheduling scheme. To this end, the present invention first performs preference selection based on the goals of each subsystem of the complex flood control system decomposition, so that the obtained solution set is more in line with the flood control requirements of each subsystem, and then physically clusters the solution set based on the preference of each subsystem, quickly extracts the scheduling solution set under different scenarios, and finally physically extracts the spatial solution set to ensure the spatial distribution characteristics of the extracted solution set. The scheduling solution set extracted based on this method not only represents the actual preferences of the subsystem, but also covers the spatial characteristics of the overall solution set, ensuring the uniformity of the distribution of the extracted solution set.
[0079] In summary, forecast error is determined as the main risk factor in the flood control scheduling process, and the distribution of flood forecast error is obtained through probability density function estimation, which improves the accuracy of distribution fitting; random simulation generates a simplified forecast error scenario set, and the physical reduction of the scenario set improves the efficiency of flood control scheduling operation, and the high precision of the simplification provides support for the accurate decision-making of flood control scheduling schemes; the complex basin flood control system is decomposed into cooperatively operated flood control subsystems through the Martin3-Zhukov's WBS two-layer model, which clarifies the internal structure of the complex system and increases the orderliness and stability of system scheduling; the multi-task perfect nested loop framework based on the simplified network scenario set-iteration number-algorithm population construction structures the flood control scheduling operation process, providing a reference method for the construction of a real-time risk scheduling model for a large-scale complex flood control system in a basin; the cluster parallel technology of the multi-task perfect nested loop reduces the dimension of the scheduling, and obtains an optimized real-time multi-objective risk scheduling model for a large-scale complex flood control system in a basin, which improves the efficiency of model solution; the clustering method based on the coupling of system goal decision-physical properties-spatial geometric characteristics facilitates real-time flood control scheduling decisions and improves scheduling accuracy. It can solve the following existing problems: inaccuracy of error distribution estimation; the need to balance the simplification of quantity and simulation accuracy in uncertain random simulation; engineering structure decomposition and coordinated scheduling of complex flood control systems; the orderliness of the operating structure of the scheduling model; inefficiency of complex large-scale calculations; and optimal extraction and decision-making of numerous uncertain scheduling schemes.
[0080] According to one aspect of the present application, step S1 further comprises:
[0081] Step S11, determine the study area, collect geographic spatial data and water system data of the study area, characteristic data of various flood control projects, hydrological data of flood control points and flood storage areas, and organize the historical flood process and forecast flood process of various flood control projects;
[0082] These data can be obtained from public data or from field surveys and measurements. The purpose is to understand the basic situation and hydrological characteristics of the study area and provide data support for subsequent flood control system decomposition, dimensionality reduction and scheduling research.
[0083] Step S12, determining flood forecast error as a risk factor for flood control scheduling, estimating the probability density function of flood forecast error of each flood control project according to the MA-parzen window method, determining the distribution characteristics of flood forecast error, and obtaining forecast error distribution parameters;
[0084] Step S13, using the Monte Carlo method to randomly simulate the forecast error according to the flood forecast error distribution and probability density function of each flood control project, and generate an initial forecast error scenario set;
[0085] Step S14: using a gradient method driven scenario network reduction model to dynamically reduce the initial forecast error scenario set to generate a reduced forecast scenario set.
[0086] According to one aspect of the present application, step S12 further comprises:
[0087] Step S12a: Since hydrological uncertainty brings risks to the flood control reservoir itself and the downstream, the forecast error is determined as a risk factor in the flood control process based on experience and expert knowledge;
[0088] Step S12b, using the absolute deviation index to calculate the difference between the predicted flood and the measured flood, and obtaining the absolute error series, which is the flood forecast error;
[0089] Step S12c, using the error series {x1, x2, ..., xn} as samples for probability density estimation using the MA-parzen window method, and establishing an error probability density estimation expression:
[0090] P*(x)=(1 / nh n )∑ i=1 n K((xx i ) / h n ), i=1, 2, …, n;
[0091] Where x is the center point of the error distribution, P*(x) is a kernel estimate of the overall probability density function P(x), and h n is the bandwidth and K is the kernel function.
[0092] Step S12d, quickly optimizing the adaptability of bandwidth and probability density function based on the mayfly optimization algorithm, and adaptively searching for the optimal bandwidth;
[0093] Step S12e, determining the parameters of each variable in the probability density function, obtaining the forecast error probability density function fitted by the MA-parzen window method, further determining the forecast error distribution, and obtaining the forecast error distribution parameters.
[0094] In this embodiment, the probability distribution of flood forecast errors of reservoirs, flood control points, and flood storage areas is fitted by the MA-parzen window method according to the actual flood forecast error data, and the mayfly optimization algorithm is used to quickly converge and adaptively optimize the bandwidth. , we can get the forecast error probability density function of each flood control project obeys N( ) normal distribution, where the parameters μ=0, σ=0.138. Compared with the traditional probability estimation method that directly assumes a specific distribution to evaluate the probability density function, the MA-parzen window method captures the distribution characteristics of the forecast error at any time. The error distribution of dynamic fitting is more in line with the actual situation, the fitting distribution is faster, and the accuracy is higher. This provides a feasible and practical method for fitting the probability distribution function.
[0095] like Figure 2 As shown, according to one aspect of the present application, the step S14 is further:
[0096] Step S14a, using the forecast error of the Monte Carlo random simulation in step S13 as the initial node of the reduced scenario network;
[0097] Step S14b, evaluate the prior probability of the node according to the MA-parzen window method, determine the position of the network node layer by layer: 1-0.8 for the first-level node, 0.8-0.6 for the second-level node, ... 0.2-0 for the fifth-level node, calculate the conditional probability between adjacent level nodes, give priority to selecting the node branch with a large conditional probability value, gradually establish the network branch between nodes, and build a reasonable and optimized network structure;
[0098] Step S14c, using the gradient descent method to calculate the position of the current research node under a certain branch of the network, the congestion degree between the coupling node prior probability and the error, and the product of the three constitutes the evaluation parameter;
[0099] Step S14d, based on the Bayesian principle, the posterior scenario probability is calculated, the position of the current research node under a certain branch of the network is calculated using the gradient ascent method, and the diagnostic parameters between each node are determined by multiplying the three together in combination with the congestion degree;
[0100] Step S14e, sort the evaluation and diagnosis parameters from high to low, eliminate the bottom 0.1% of the rankings, and evaluate the representativeness and coverage of the retained nodes. If the evaluation passes, retain these nodes and their network branches, otherwise discard the nodes, determine the nodes that meet the conditions to build a simplified scenario network, and use MA-parzen to fit the scenario distribution to obtain the distribution parameters;
[0101] Step S14f: With the repetition of the above steps S14b-S14e, the simplified scenario network continues to evolve until the absolute error between the evolved simplified scenario and the initial scenario distribution parameter is greater than 0.01, then the evolution is stopped to obtain a simplified forecast error scenario set.
[0102] In this embodiment, the number of flood control project forecast error scenario sets generated by Monte Carlo random simulation is 1,000. Taking the Niuyushan Reservoir as an example, the specific implementation steps of the simplified scenario are as follows:
[0103] The 1000 randomly simulated Niuyushan reservoir inflow error scenarios were used as the initial nodes of the simplified scenario network.
[0104] The forecast error scenario follows a normal distribution N (0, 0.138 2 ), the MA-parzen window method can further determine the prior probability of 1000 error scenarios of random simulation. This prior probability is independent and not affected by other error sets, which ensures the stability of the initial network nodes;
[0105] According to the prior probability of the nodes, the nodes are divided into 5 levels in descending order of prior probability: 1-0.8 is the first-level node, 0.8-0.6 is the second-level node, and so on;
[0106] Calculate the conditional probabilities between adjacent level nodes, prioritize the nodes with larger conditional probability values to establish the network relationship between the two nodes, optimize the network structure, and construct a preliminary scenario network;
[0107] Calculate the congestion between nodes and the position of the current node (the value range is 0~1), the evaluation parameter = node position driven by gradient descent method × node prior probability × congestion between nodes;
[0108] Reverse diagnosis is performed based on the network structure branch. The gradient ascent method is used to calculate the position of the current research node under a certain branch of the network (the value range is 0~1). The posterior scenario probability is calculated according to the Bayesian principle. The diagnostic parameter = node position × posterior scenario probability × the degree of crowding between nodes.
[0109] Rank the assessment and diagnostic parameters from high to low, and eliminate the bottom 0.1% of both rankings;
[0110] The representativeness and coverage of the retained nodes are evaluated based on the global spatial autocorrelation. If the correlation between the retained nodes and the global is between 0.5 and 1, the evaluation is passed and these nodes and their network branches are retained. Otherwise, the nodes are discarded and the nodes that meet the conditions are determined to construct a simplified scenario network. MA-parzen is used to fit the scenario distribution to obtain the distribution parameters.
[0111] The simplified scenario network evolves continuously with the repetition of the above steps until the absolute error between the evolved simplified scenario and the initial scenario distribution parameter is greater than 0.01, then the evolution is stopped, and the final set of 380 simplified forecast error scenarios for the inflow of the Niuyushan Reservoir is obtained.
[0112] This embodiment proposes a scenario network simplification model driven by the gradient method, which divides the node levels by conditional probability to construct an optimized network structure, introduces the congestion between nodes to judge the distance between scenarios and the uniformity of scenario distribution, and better balances the representativeness of samples. At the same time, the gradient method is used to update the evolving network nodes in real time to improve the randomness and diversity of scenario selection. This simplification model can not only dynamically reduce the error scenario set but also ensure the accuracy of random simulation, avoiding the problem that traditional random simulation ignores the large amount of data generated by simulation and high redundancy in order to ensure simulation accuracy, ensuring that the generated simplified error samples have high representativeness and high coverage, improving the accuracy of flood forecasting, reducing the computational load of flood control scheduling, and well meeting the requirements of uncertain random simulation to achieve high accuracy when there is less simulation data.
[0113] In this embodiment, considering the hydrological uncertainty, the simplified forecast error scenario set is superimposed on the forecast inflow process of each flood control project, specifically:
[0114] Q i (t) = Q i mean (t)[1+ξ i (t)], i=1, 2, …, I;
[0115] In the formula, Q i (t) is the predicted inflow scenario of flood control project i at time t, Q i mean (t) is the mean value of the predicted inflow of flood control project i at time t, ξ i (t) is the simplified forecast error scenario corresponding to the inflow of flood control project i at time t.
[0116] Since the distribution of the simplified error scenario is known, the distribution of the coupled hydrological uncertainty is consistent with the simplified scenario distribution, both of which are normal distributions, which avoids repeated evaluation of the hydrological uncertainty distribution, optimizes the operation of flood control scheduling tasks, and improves work efficiency.
[0117] According to one aspect of the present application, step S2 further comprises:
[0118] Step S21, constructing a Martin3-Zhukov's WBS two-layer model and decomposing the complex flood control system by dimensionality reduction based on the Martin3-Zhukov's WBS two-layer model to obtain Z collaborative subsystems, where Z is a positive integer greater than 2;
[0119] Step S22: construct the objective function library of each subsystem respectively, use the weight adjustment method of dynamic criterion elimination to evaluate the weight of each objective, and convert the linear weighted method into the total objective function for evaluating the safety of the subsystem;
[0120] Step S23, respectively constructing a constraint library of the real-time flood control risk dispatch model of each subsystem, including: water balance constraint, reservoir discharge constraint, reservoir characteristic water level constraint, initial / final water level control constraint of the dispatch period, reservoir discharge amplitude constraint, flood storage and detention area water balance constraint, flood storage and detention area capacity constraint and flood control point safety constraint;
[0121] Step S24: Based on the overall objective function and constraint condition library, a real-time multi-objective risk scheduling model for a large-scale complex flood control system in a river basin is constructed.
[0122] In this embodiment, the large-scale and complex flood control system in the Huaihe River Basin is large in scale, and the hydraulic connections between the various projects are complex, which increases the difficulty of constructing the flood control dispatch model in the basin and the complexity of the flood control system dispatch. Therefore, the Martin3-Zhukov's WBS two-layer model is proposed to decompose the complex flood control system in the Huaihe River Basin into subsystems that operate in a coordinated manner, decompose it into coordinated subsystems with the overall goal of system flood control safety, determine the composition of flood control projects within each subsystem, clearly generalize the structure of the flood control system, and establish an objective function with subsystems as units to construct a large-scale and complex flood control system dispatch model in the basin, which helps to balance the flood control tasks between subsystems and steadily improve the flood control capacity of the basin.
[0123] According to one aspect of the present application, step S21 is further:
[0124] Step S21a, generalizing the watershed and distinguishing the geographical space of the complex flood control system, and determining the number of subsystem targets to be decomposed;
[0125] Step S21b, partitioning the complex flood control basin system based on spatial geographic location, using Zhukov's method to gradually cluster adjacent partitions, and decomposing the large-scale complex flood control system in the basin into Z subsystems according to the principle of minimizing the area difference between subsystems, where Z is a positive integer greater than 2;
[0126] Step S21c, continue to decompose the subsystem according to the attributes of the flood control project, and encode the flood control project to establish an index system;
[0127] Step S21d, using the Martini3 model to analyze the dependencies between flood control projects in the subsystem based on geographical location, project attributes, and target effectiveness. The closer the dependencies, the greater the probability that they are a subsystem.
[0128] Step S21e, check the rationality of the decomposition of the subsystem in the above step S21b, determine the members of each subsystem in the complex flood control basin, if the subsystem decomposition is incorrect, in order to coordinate and balance the flood control tasks and reasonably allocate flood control projects, use the shortest Euclidean distance method to re-assign the flood control projects to the subsystem, determine the final Z subsystem internal structures, and count the number of reservoirs, flood control points and flood storage areas in the subsystem.
[0129] In this embodiment, the large-scale complex flood control system in the Huaihe River Basin is composed of flood control projects such as reservoirs, river embankments, and flood storage areas. There are complex hydraulic connections and synergies between the flood control projects, and it is difficult to directly perform system flood control scheduling. Therefore, the Martin3-Zhukov's WBS two-layer model is used to decompose and reduce the dimensionality of the high-dimensional complex system. The specific implementation process is as follows:
[0130] The complex flood control system in the Huaihe River Basin was generalized and a generalized map was drawn. The number of subsystem targets to be decomposed was initially determined to be 5 through geographic spatial discrimination.
[0131] Due to the irregularity of the boundary of the Huaihe River Basin, the smaller the partition is, the better, but it should not be less than the total number of flood control projects, which is 51. Based on this principle, the complex flood control system of the Huaihe River Basin is divided into regions according to the spatial geographical location, and the adjacent partitions are gradually clustered using Zhukov's method. According to the principle of minimizing the area difference between subsystems, the large-scale complex flood control system of the basin is decomposed into five preliminary subsystems: A, B, C, D, and E;
[0132] The five subsystems were further decomposed according to the attributes of flood control projects in the Huaihe River Basin, and the flood control projects in each subsystem were coded to establish an overall flood control index system.
[0133] The Martini3 model was used to analyze the interdependence between flood control projects within the subsystem based on geographical location, project attributes, and target effectiveness, and to test the rationality of the structural division of the five subsystems: (1) It was found that the project attributes of subsystem A were incomplete and there was a lack of flood storage areas; (2) The boundary divisions between subsystems C and D were not obvious. Therefore, it was necessary to reclassify each flood control project into a system. According to the principle that "there must be reservoirs, flood control points, and flood storage areas in the same area to better coordinate and balance flood control tasks", the shortest Euclidean distance method was used to reclassify the flood control projects into subsystems;
[0134] Update the internal structure of the A, B, C, and D systems, adjust the four systems into two systems, A and C, and change the original E system to the B system.
[0135] The large-scale and complex flood control system in the Huaihe River Basin is decomposed into three subsystems, A, B, and C. The internal structure of each subsystem is determined, and the number of reservoirs, flood control points, and flood storage and detention areas in the subsystem is counted: System A has 11 reservoirs, 9 flood control points, and 3 flood storage and detention areas; System B has 9 reservoirs, 3 flood control points, and 4 flood storage and detention areas; System C has 5 reservoirs, 3 flood control points, and 4 flood storage and detention areas.
[0136] Martin3-Zhukov's WBS two-layer model decomposes and reduces the high-dimensional complex system of the Huaihe River Basin into three subsystems, clearly divides the internal structure of each flood control subsystem, and tests the rationality of the flood control engineering structure of the three subsystems based on geographical location-engineering attributes-target effectiveness, promotes the rational and orderly scheduling of flood control risk in the Huaihe River Basin, ensures the efficiency and safety of the overall flood control system, and provides a basis for maximizing the engineering benefits of the Huaihe River Basin flood control system. It solves the problem of reasonable decomposition and coordinated scheduling of the complex flood control system engineering structure, and provides a feasible method for the decomposition and coordinated scheduling of other complex basins.
[0137] In this embodiment, the step S22 is specifically as follows:
[0138] Construct the target function library of each subsystem of large-scale and complex flood control in the Huaihe River Basin, including flood control safety of reservoir groups in subsystems A, B, and C, safety of downstream flood protection areas, and safety of flood storage areas;
[0139] Objective 1: Taking the flood control safety of the reservoir group in the subsystem as the goal, an objective function for maximizing the total flood control safety of the reservoir group is established based on the storage capacity-load weight, which can be expressed as:
[0140] ;
[0141] In the formula, z=1, 2, 3 correspond to subsystems A, B, C, R respectively. z is the total flood control safety of the reservoir group in subsystem z, V z,j * is the flood control capacity of the jth reservoir in subsystem z, max (V z,j (t)) is the maximum flood control storage capacity used by the jth reservoir in subsystem z during the dispatch period, α z,j is the weight coefficient of the jth reservoir in subsystem z.
[0142] α z,j It is determined by the storage capacity of each reservoir and the load of the basin where it is located, and is expressed as:
[0143] ;
[0144] Where P z,j is the rainfall in the local basin where the jth reservoir in subsystem z is located, Fz,j is the area of the local basin where the jth reservoir in subsystem z is located.
[0145] Objective 2: Taking the safety of the downstream flood protection area as the goal, establish an objective function to minimize the maximum flow rate at the downstream flood control point, expressed as:
[0146] ;
[0147] In the formula, QC z,l (t) is the water flow at the downstream flood control point l in subsystem z during period t.
[0148] Objective 3: Taking the safety of flood storage and detention areas as the goal, establish an objective function to minimize the maximum flood diversion volume of flood storage and detention areas in complex flood control basins, expressed as:
[0149] ;
[0150] Where t0, t d are the initial and final time of scheduling, respectively, q z,k (t) is the actual flow of the river where the flood storage area k in subsystem z is located, q z,k,an is the safe flow allowed by the river channel where the flood storage area k in subsystem z is located. Since flood control in the subsystem is multi-objective, it is difficult to quickly optimize and obtain the overall goal of the flood control system. Therefore, taking the subsystem as a unit, the weight of each goal in the subsystem is determined based on the criterion elimination method, and the subsystem goal is obtained by weighting. The specific process is as follows:
[0151] The relevant criteria for determining the target evaluation are: geographic space-engineering attributes-target effectiveness-flood storage capacity-risk resistance capacity;
[0152] Target F within the subsystem z,1 , F z,2 , F z,3 According to the above-mentioned relevant criteria, the target decision matrix is constructed and normalized;
[0153] Based on the Huaihe River's geographic space - reservoirs, control sections, flood storage areas and other engineering attributes - F z,1 , F z,2 , F z,3 Standard weights of each target in the target effectiveness-flood storage capacity-flood risk resistance capacity calculation subsystem;
[0154] The criterion elimination method is used to calculate the target F after eliminating a certain criterion. z,1 , F z,2 , F z,3 The weight value of
[0155] The statistical criteria remove the absolute deviation of each target before and after and add them up. Each target F z,1 , Fz,2 , F z,3 The absolute deviation / total deviation is the weight of each target;
[0156] The total goal of each subsystem is obtained by linearly weighting the weight combination and each subsystem goal, which can be expressed as:
[0157] F z =β1F z,1 +β2F z,2 +β3F z,3 ;
[0158] In the formula, β1, β2, β3 are the weight coefficients of each target in subsystem z. The specific determination steps are:
[0159] According to the number of 380 error scenarios, the weight adjustment method of dynamic criterion elimination is used to generate the target weight pool of the corresponding subsystems under different error scenarios. The dimension of the target weight pool is 380*3, 380 is the scenario set, and 3 is the corresponding β1, β2, and β3 values under different scenario sets. Each time the scenario is run, the weight combination of the reliability of the corresponding error scenario number is selected, and the weight coefficient of the subsystem corresponding to the target weight pool is found according to the simplified error scenario (for example, the total target weight coefficients of subsystem C under simplified error scenario 1 are 0.09454, 0.268062, and 0.637399, respectively).
[0160] In this embodiment, each subsystem F is constructed z,1 , F z,2 , F z,3 The constraint library of the real-time flood control risk dispatch model is as follows:
[0161] Constraint 1: Water balance constraint
[0162] ;
[0163] Constraint 2: Reservoir discharge constraint
[0164] ;
[0165] Constraint 3: Reservoir characteristic water level constraint ;
[0166] Constraint 4: Water level control constraint at the beginning and end of the dispatch period
[0167] ; ;
[0168] Constraint 5: Reservoir discharge amplitude constraint ;
[0169] Constraint 6: Water balance constraint in flood storage area
[0170] ;
[0171] Constraint 7: Flood storage and detention area capacity constraint ;
[0172] Constraint 8: Flood control point safety constraints ;
[0173] Where V z,j (t) and V z,j (t-1) are the storage capacity of reservoir j in subsystem z at time t and t-1 respectively; I z,j (t) and Q z,j (t) are the inflow and outflow of reservoir j in subsystem z at time t; Q z,j,e is the ecological flow requirement of the river downstream of reservoir j in subsystem z; Q z,j,s is the minimum discharge flow of reservoir j in subsystem z to meet the downstream navigation requirements; f ZQ (·) is the discharge capacity curve of reservoir j; Q z,j,down max is the safe discharge of the river downstream of reservoir j in subsystem z; z,j (t) is the water level of reservoir j in subsystem z at time t, Z z,j,min and Z z,j,max is the minimum and maximum water level allowed for reservoir j in subsystem z; z,j,s is the starting water level of reservoir j in subsystem z at the beginning of the dispatch period; z,j,e is the expected water level of reservoir j in subsystem z at the end of the dispatch period; ∆Q z,j,con V is the discharge variation allowed at adjacent moments of reservoir j in subsystem z; z,k (t-1) and V z,k (t) is the initial and final water storage capacity of flood storage area k in subsystem z during period t, QX z,k (t) and QX z,k (t-1) is the diversion flow of flood storage area k in subsystem z at the beginning and end of period t; V z,k * is the designed water storage capacity of flood storage area k in subsystem z; Qh z,l,an is the maximum flow rate allowed at flood control point l in subsystem z.
[0174] In summary, the complex flood control system in the Huaihe River Basin is divided into three subsystems, A, B, and C. The structural hierarchy is clearer, which is convenient for the dispatching and management of complex systems, optimizes the dispatching and operation structure, and improves the efficiency of the dispatching and operation of complex systems; the objective function is constructed with each subsystem as a unit, and the total goal of the subsystem is obtained by weighting based on the criterion elimination method, and the dynamic weight assignment increases the diversity of model solutions; the flood control safety of the subsystem is fully considered, and constraints are established to constrain the safety of flood control, and a risk dispatching model for the complex flood control system in the Huaihe River Basin is constructed. This method solves the problem of engineering decomposition and coordinated dispatching of the complex flood control system in the Huaihe River Basin, and promotes the orderly and stable operation of the risk dispatching of the complex flood control system in the Huaihe River Basin.
[0175] According to one aspect of the present application, step S3 further comprises:
[0176] Step S31, using a perfect nested loop framework to build a multi-task parallel module of forecast error scenario sets, iteration times and populations, decomposing and reducing the flood control scheduling process from multiple angles, and optimizing the real-time multi-objective risk scheduling model of a large-scale complex flood control system in a basin by combining MATLAB distributed clusters and CPU parallel technology;
[0177] Step S32, extract the simplified error scenario set and superimpose it on the forecast inflow of each flood control project, and input the obtained data into the optimized real-time multi-objective risk scheduling model of the large-scale complex flood control system in the basin, and use the improved multi-objective mayfly algorithm based on Debye Waller factor optimization to solve it, so as to obtain the real-time flood control scheduling plan set corresponding to the simplified error scenario set.
[0178] like Figure 3 As shown, according to one aspect of the present application, the step S31 is further:
[0179] Step S31a, building a MATLAB distributed cluster and determining the number of CPU cores of the nodes and cluster;
[0180] Step S31b, compiling a multi-task parallel module consisting of a forecast error scenario set, an iteration number, and a population nested loop;
[0181] Step S31c: To ensure load balancing of scheduling parallelism, the forecast error scenario set of the outermost loop is allocated to the cluster nodes;
[0182] Step S31d, cooperating with step S31c to allocate the number of iterations of the middle layer to different threads of the node;
[0183] Step S31e, in conjunction with step S31d, calculates the innermost loop population in parallel.
[0184] In this embodiment, step S31 is further as follows:
[0185] Determine the environment for building the cluster: computing hardware: HP ZHAN 99 Pro G4 Microtower PC, CPU: 11th Gen Intel(R) Core(TM) i7-11700 @ 2.50GHz, 2.50 GHz, 12 cores, 16 logical processors, RAM: 16GB, disk drive: SAMSUNG MZVLQ256HBJD-00BH1; software configuration: to ensure consistent MATLAB operating performance and avoid compatibility issues between different versions, MATLAB R2023b version and MATLAB Parallel Server tool are uniformly installed on each client; cluster communication environment: connect all three computers to the same router so that the computers in the cluster are in one gateway.
[0186] Determine the configuration of nodes and clusters: configure MATLAB Job Scheduler on each computer (computing node) and start the service; use the command line to start Admin Center on the master node computer, add Hosts in the unified gateway, and specify the master node and computing nodes; start MATLAB Job Scheduler, and select the maximum number of cores for each child node in cluster construction, Cores=12; test the connectivity between the nodes, and finally determine the total number of cores in the cluster, Cores=12*3=36.
[0187] Constructing a multi-task perfect nested loop framework: The flood control system in the Huaihe River Basin is complex, and the hydraulic connections between various projects are close. The simplified scenario set in the case is 380, the number of nested iterations is 1000, and the number of intelligent algorithm populations is 200. The coupled computational complexity is O=380@1000@200, which increases the difficulty of solving the real-time flood control multi-objective risk scheduling model. Based on the perfect nested loop framework, a multi-task perfect nested loop framework with a forecast error scenario set of 380, iterations of 1000, and populations of 200 is constructed. The nested loops are layered, and the multi-task flood control scheduling work structure is organized, which improves the efficiency of solving the risk scheduling model of the complex flood control system in the Huaihe River Basin.
[0188] Population parallel computing: Open the MATLAB distributed cluster and set the maximum number of parallel pools (workers) to 36 through the parpool(n) command line; change the outer loop of the distributed module to the parfor function of the MATLAB Parallel Computing Toolbox, and click the main function to run the program to start multi-task parallelism; in the main program, the client applies for the core processor (Central Processing Unit, CPU) and memory from the master node; to ensure the load balance of cluster task parallelism, the forecast error scenario set of the outermost loop is assigned to the cluster nodes, and the number of IMA iterations in the middle layer is assigned to different threads of the node, and the innermost loop population is calculated in parallel. In this way, the algorithm population parallel tasks are successfully distributed to different computing nodes. Each node has a certain amount of CPU and memory and completes the computing tasks independently; when all nodes complete the computing tasks, the results are summarized and returned to the master node and then fed back to the client.
[0189] In summary, by building a MATLAB distributed cluster with 36 cores of workers, the parallel computing time of the perfect nested population is 120s, which is significantly shorter than the model serial computing time of 3214s, and the model operation efficiency is greatly improved; based on the perfect nested loop framework, the scheduling task is decomposed layer by layer, and compared with the traditional parallel computing, the technical dimensionality reduction processing is directly performed in the task processing module, and the MATLAB distributed cluster is coupled to perform CPU technology parallel optimization of the scheduling running time, which solves the problem of low efficiency of large-scale flood control scheduling models in complex watersheds, and provides a feasible method for optimizing the high-dimensional complex calculation of real-time flood risk scheduling of large-scale complex flood control systems in watersheds.
[0190] In another embodiment of the present application, S31d is specifically:
[0191] The forecast error scenario set assigned to each node is further subdivided, and the MA iterations are assigned to different threads within the node.
[0192] Get the number of node threads: First, use MATLAB's Parallel Computing Toolbox to get the number of threads available on each node. Calculate the number of iterations per thread: Divide the total number of MA iterations assigned to each node by the number of threads on the node to get the approximate number of iterations that each thread needs to process.
[0193] Processing remainders: If the total number of iterations is not divisible by the number of threads, the remainder of iterations is distributed to the first few threads to ensure that all iterations are processed.
[0194] Create tasks: Create a parallel task for each thread. The task contains the range of iterations that the thread needs to process and the corresponding forecast error scenario.
[0195] S31e is specifically: each thread will parallelly process the MA population calculation within the range of MA iterations assigned to it.
[0196] Loop within each thread: In each thread's task, use a parfor loop to parallelize the number of MA iterations assigned to that thread.
[0197] Calculate the MA population: In each iteration of the parfor loop, calculate the fitness value and other relevant indicators of the corresponding MA population.
[0198] Collect results: After the parfor loop ends, collect the results calculated by each thread and merge them into a complete MA (Mayfly Algorithm) population result.
[0199] According to one aspect of the present application, step S32 is further:
[0200] Step S32a, taking the outflow of the reservoir as the initial population, introducing a sliding operator to perform sliding average processing on the initial population to ensure the smoothness of the outflow flow process;
[0201] Step S32b, calculate the system objective function, evaluate the fitness functions of all populations, and introduce external archives to screen non-inferior solution sets;
[0202] Step S32c, determining the iteration condition, if not satisfied, updating the speed and position of the female and male mayflies;
[0203] Step S32d, evaluate the solution, if it is a non-dominated solution, update the archive, otherwise there is a 50% probability of updating the archive;
[0204] Step S32e, introducing the Debye Waller factor to perform stable sorting on the mayflies, mating and evaluating the offspring, and determining whether the offspring is a non-dominated solution of its parent of the same sex, and if so, storing the non-dominated solution in an external archive;
[0205] Step S32f, satisfying the iteration condition, retaining the final non-dominated solution set as the real-time flood control scheduling solution set.
[0206] In this embodiment, due to the staged and intermittent flood discharge operation of reservoir scheduling, the downstream discharge flow of the reservoir during the reservoir scheduling process may be jagged, affecting the downstream ecological security, the mismatch between water supply and demand, and the flood risk management, which brings serious challenges. This application introduces a "sliding operator" to perform sliding average processing on the initial population, reduce the fluctuation of the outflow flow, ensure the smoothness of the outflow flow process, make the outflow flow more stable, thereby reducing the impact on the downstream ecology and water conservancy facilities, and optimizing water resources management and scheduling.
[0207] Intelligent optimization algorithms have good global search capabilities, can perform global searches in a wide range of spaces, avoid falling into local optimal solutions, can effectively solve nonlinear, high-dimensional and complexly constrained optimization problems, are highly flexible, and are therefore suitable for different fields, but the fast convergence speed of the algorithm often leads to insufficient robustness of the algorithm. How to optimize this problem is a problem that most existing intelligent optimization algorithms should solve. The present invention introduces the Debye Waller factor to perform robustness optimization on the sorting and mating of the multi-objective mayfly optimization algorithm, solves the Huaihe River Basin flood control risk scheduling model coupled with a simplified forecast scenario set of 380×1000 iterations×200 populations, and obtains 76 million scheduling scheme sets, which increases the stability of the multi-objective mayfly optimization algorithm in solving the flood control scheduling model, and facilitates the decision-making of subsequent flood control scheduling schemes.
[0208] In summary, the "sliding operator" is introduced to smooth the outflow of reservoirs to ensure the stability of the outflow flow; the multi-objective mayfly optimization algorithm is used to solve the risk dispatch model of the complex flood control system in the Huaihe River Basin, and the optimized solution set is obtained; in order to solve the instability of the intelligent algorithm in the process of solving the dispatch model, the Debye Waller factor is introduced to improve the multi-objective mayfly intelligent optimization algorithm, enhance the robustness of the algorithm in the process of flood control risk dispatch in the Huaihe River Basin, and enhance the stability of the model solution. The problem of unstable convergence of most existing intelligent optimization algorithms is solved, ensuring the stability and safety of the flood control dispatch process in the Huaihe River Basin.
[0209] According to one aspect of the present application, the clustering method based on the coupling of system target decision-physical attribute-spatial geometric feature in step S4 extracts the typical scheduling scheme corresponding to the reduced error scenario set, further comprising:
[0210] Step S41: based on the flood control objectives of the Z subsystems of large-scale complex flood control in the basin, a set of scheduling schemes is preferentially selected to obtain a set of decision solutions for each subsystem;
[0211] Step S42, according to the physical properties of the solution set, the intra-cluster spacing of the spatial point set distribution is as small as possible, and the inter-cluster spacing is as large as possible, and the decision solution set of each subsystem is divided into M categories using Fuzzy C-Means Clustering, where M is a positive integer;
[0212] Step S43, using an adaptive curvature entropy point cloud simplification algorithm to calculate the curvature entropy in the M-type solution set, removing redundant points, keeping the spatial characteristics of the solution set unchanged, and extracting 0.1% of the representative point set from the M-type solution set;
[0213] Step S44, the point set finally extracted is a typical scheduling solution based on clustering decision-making coupled with system goal decision-physical attribute-spatial geometric feature.
[0214] In this embodiment, step S4 is specifically as follows:
[0215] The large-scale and complex flood control system in the Huaihe River Basin is decomposed into three coordinated subsystems A, B, and C. The three subsystems A, B, and C are mainly based on the flood control dispatch safety of the overall system, but each subsystem has the goal of coordinated dispatch with the overall system. Therefore, when making decisions on the plan, preference selection is first made based on the subsystem goals, and the preference solution set for system A is 24.25 million, the preference solution set for system B is 22.5 million, and the preference solution set for system C is 25.25 million.
[0216] Since the number of solution sets of each subsystem obtained according to the subsystem target decision is large, it still cannot meet the flood control scheduling decision requirements. Therefore, this application divides the physical attributes of the subsystem point set based on the idea of "the spacing within the cluster is as small as possible, and the spacing between clusters is as large as possible". The decision solution set of subsystem A is divided into 2 categories, the decision solution set of subsystem B is divided into 2 categories, and the decision solution set of subsystem C is divided into 3 categories through the Fuzzy C-Means Clustering method.
[0217] In order to meet the real-time requirements of the dispatching decision-making in the Huaihe River Basin, the solution set in the subsystem-cluster N (N is the number of clusters of the subsystem by the Fuzzy C-MeansClustering method) is extracted. In order to ensure the spatial characteristics of the extracted point set, the adaptive curvature entropy point cloud simplification algorithm is used to calculate the curvature entropy in each subsystem solution set, remove redundant points, keep the spatial characteristics of the solution set within the subsystem unchanged, and extract 0.1% of the representative point set from the cluster solution set: the representative point set of subsystem A-cluster 1 is 12,250; the representative point set of subsystem A-cluster 2 is 12,000; the representative point set of subsystem B-cluster 1 is 10,450; the representative point set of subsystem B-cluster 2 is 12,050; the representative point set of subsystem C-cluster 1 is 8,750; the representative point set of subsystem C-cluster 2 is 8,100; the representative point set of subsystem C-cluster 3 is 8,400. In this way, the clustering method based on the coupling of target decision-making-physical attributes-spatial geometric characteristics is used to peel off many scheduling solution sets layer by layer, and the extracted point set has both high spatial coverage and practical significance, providing strong technical support for flood control scheduling decision-making.
[0218] In short, the preference selection of the scheduling scheme is made according to the target decision of the subsystem, so that the preference solution set has more practical characteristics; the Fuzzy C-Means Clustering method is used to classify the preference solution set, and the adaptive and efficient optimization clustering is used to improve the scheduling decision time, which is convenient for the extraction of subsequent scheduling schemes; the adaptive curvature entropy point cloud simplification algorithm is used to extract the point set, which preserves the spatial form of the point set and ensures the extraction accuracy to obtain a representative solution set. This embodiment solves the existing technical problems: how to ensure the actual characteristics of the extracted point set; how to ensure high coverage and high representativeness of the decision-making scheme. Compared with the traditional clustering method, the point set extracted by the clustering method based on the system attributes is closer to the flood control scheduling requirements and avoids the situation of mechanically extracting physical points. Maintaining the spatial characteristics of the solution set and ensuring the uniformity of the extracted point set facilitates the accurate decision-making of the scheduling plan for the complex flood control system in the Huaihe River Basin, and provides a practical and feasible method for accurate decision-making in real-time flood control.
[0219] According to another aspect of the present application, a real-time risk dispatching system for a large-scale complex flood control system in a river basin includes:
[0220] at least one processor;
[0221] and a memory communicatively coupled to at least one of the processors;
[0222] Among them, the memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the real-time risk scheduling method for a large-scale complex flood control system in a river basin as described in any of the above technical solutions.
[0223] The preferred embodiments of the present invention are described in detail above; however, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all belong to the protection scope of the present invention.
Claims
1. A real-time risk dispatching method for a large-scale complex flood control system in a river basin, characterized by: The following steps are involved: Step S1, collecting flood data and engineering characteristic data of various flood control projects in the complex flood control system, determining that flood forecast error is a risk factor in flood control scheduling, using the MA-parzen window method to estimate the probability density function of the flood forecast error, obtaining distribution parameters, generating an initial error scenario set based on Monte Carlo random simulation of flood forecast error and inputting it into a pre-built scenario network reduction model, using the gradient method to drive the model, and obtaining a reduced error scenario set; Step S2, constructing a Martini3-Zhukov's WBS two-layer model and performing system decomposition and dimensionality reduction on the complex flood control system to obtain Z subsystems, constructing an objective function library and a constraint condition library based on the subsystems, using a weight adjustment method with dynamic criterion elimination to determine the weight of each objective of the subsystem, and converting the linear weighting into an overall objective function for evaluating the safety of the subsystem, and constructing a real-time multi-objective risk scheduling model for a large-scale complex flood control system in a river basin; Step S3, optimizing the real-time multi-objective risk scheduling model of a large-scale complex flood control system in a watershed based on cluster parallel technology of multi-task perfect nested loops, superimposing the simplified error scenario set on the forecast inflow of each flood control project and inputting the optimized real-time multi-objective risk scheduling model of a large-scale complex flood control system in a watershed, solving it by using an improved multi-objective mayfly algorithm based on Debye Waller factor optimization, and obtaining a real-time flood control scheduling solution set corresponding to the simplified error scenario set by the iteration number population; Step S4, extracting typical dispatching schemes corresponding to the reduced error scenario set based on the clustering method of system target decision-physical attribute-spatial geometric feature coupling, making decisions on the real-time flood control dispatching scheme set based on the typical dispatching schemes, and selecting the optimal scheme as the real-time risk dispatching scheme for the large-scale complex flood control system in the basin; The step S2 is further as follows: Step S21, constructing a Martini3-Zhukov's WBS two-layer model and decomposing the complex flood control system by dimensionality reduction based on the Martini3-Zhukov's WBS two-layer model to obtain Z collaborative subsystems, where Z is a positive integer greater than 2; Step S22: construct the objective function library of each subsystem respectively, use the weight adjustment method of dynamic criterion elimination to evaluate the weight of each objective, and convert the linear weighted method into the total objective function for evaluating the safety of the subsystem; Step S23, respectively constructing a constraint library of the real-time flood control risk dispatch model of each subsystem, including: water balance constraint, reservoir discharge constraint, reservoir characteristic water level constraint, initial / final water level control constraint of the dispatch period, reservoir discharge amplitude constraint, flood storage and detention area water balance constraint, flood storage and detention area capacity constraint and flood control point safety constraint; Step S24: construct a real-time multi-objective risk dispatch model for a large-scale complex flood control system in a river basin based on the overall objective function and the constraint condition library; The step S21 is further as follows: Step S21a, generalizing the watershed and distinguishing the geographical space of the complex flood control system, and determining the number of subsystem targets to be decomposed; Step S21b, partition the complex flood control basin system according to the spatial geographical location. Due to the irregularity of the basin boundary, the smaller the partition, the better, but it should not be less than the number of flood control projects. Zhukov's method is used to gradually cluster adjacent partitions. According to the principle of minimizing the area difference between subsystems, the large-scale complex flood control system in the basin is decomposed into Z subsystems, where Z is a positive integer greater than 2; Step S21c, continue to decompose the subsystem according to the attributes of the flood control project, and encode the flood control project to establish an index system; Step S21d, using the Martini3 model to analyze the dependencies between flood control projects in the subsystem based on geographical location, project attributes, and target effectiveness. The closer the dependencies, the greater the probability that they are a subsystem. Step S21e, check the rationality of the decomposition of the subsystem in the above step S21b, determine the members of each subsystem in the complex flood control basin, if the subsystem decomposition is wrong, in order to coordinate and balance the flood control tasks and reasonably allocate flood control projects, use the shortest Euclidean distance method to re-subordinate the flood control projects to the subsystem, determine the final Z subsystem internal structures, and count the number of reservoirs, flood control points and flood storage areas in the subsystem; The step S22 is further as follows: Step S22a, constructing the target function library of each subsystem of large-scale complex flood control in the basin, including the flood control safety of the reservoir group within the subsystem, the safety of the downstream flood protection area and the safety of the flood storage area; Step S22b: Determine the weight of each target using the weight adjustment method of dynamic criterion elimination, and construct the total objective function of the subsystem by linear weighting.
2. The real-time risk dispatching method for a large-scale complex flood control system in a river basin according to claim 1 is characterized in that: The step S1 is further as follows: Step S11, determine the study area, collect geographic spatial data and water system data of the study area, characteristic data of various flood control projects, hydrological data of flood control points and flood storage areas, and organize the historical flood process and forecast flood process of various flood control projects; Step S12, determining flood forecast error as a risk factor for flood control scheduling, estimating the probability density function of flood forecast error of each flood control project according to the MA-parzen window method, determining the distribution characteristics of flood forecast error, and obtaining forecast error distribution parameters; Step S13, based on the flood forecast error distribution characteristics and parameters of the flood control project, the Monte Carlo method is used to randomly simulate the forecast error to generate an initial forecast error scenario set; Step S14: construct a scenario network reduction model driven by the gradient method, input the initial forecast error scenario set into the scenario network reduction model, and generate a reduced error scenario set.
3. The real-time risk dispatching method for a large-scale complex flood control system in a river basin according to claim 2 is characterized in that: The step S14 is further as follows: Step S14a, using the forecast error of the Monte Carlo random simulation in step S13 as the initial node of the reduced scenario network; Step S14b, evaluate the prior probability of the node according to the MA-parzen window method, and determine the position of the network node layer by layer, where 1-0.8 is a first-level node, 0.8-0.6 is a second-level node, ..., 0.2-0 is a fifth-level node; calculate the conditional probability between adjacent level nodes, select the node with a large conditional probability value as the node branch, gradually establish the network branch between nodes, and construct an optimized network structure; Step S14c, using the gradient descent method to calculate the position of the current research node under a certain branch of the network, the congestion degree between the coupling node prior probability and the error, and the product of the three constitutes the evaluation parameter; Step S14d, based on the Bayesian principle, the posterior scenario probability is calculated, the position of the current research node under a certain branch of the network is calculated using the gradient ascent method, and the diagnostic parameters between each node are determined by multiplying the three together in combination with the congestion degree; Step S14e, respectively arranging the evaluation and diagnosis parameters in descending order, and eliminating the bottom a% of the rankings of the two, evaluating the representativeness and coverage of the retained nodes; if the evaluation passes, retain these nodes and their network branches, otherwise discard the nodes, determine the nodes that meet the conditions to build a simplified scenario network, and use MA-parzen to fit the scenario distribution to obtain the distribution parameters; a is the set value; Step S14f, repeating steps S14b-S14e, so that the reduced scenario network continues to evolve until the absolute error between the evolved reduced scenario and the initial scenario distribution parameter is greater than a preset value, then stopping the evolution and obtaining a reduced error scenario set.
4. The real-time risk dispatching method for a large-scale complex flood control system in a river basin according to claim 1 is characterized in that: The step S3 is further as follows: Step S31, using a perfect nested loop framework to build a multi-task parallel module of forecast error scenario sets, iteration times and populations, decomposing and reducing the flood control scheduling process from multiple angles, and optimizing the real-time multi-objective risk scheduling model of a large-scale complex flood control system in a basin by combining MATLAB distributed clusters and CPU parallel technology; Step S32, extract the simplified error scenario set and superimpose it on the forecast inflow of each flood control project, and input the obtained data into the optimized real-time multi-objective risk scheduling model of the large-scale complex flood control system in the basin, and use the improved multi-objective mayfly algorithm based on Debye Waller factor optimization to solve it, so as to obtain the real-time flood control scheduling plan set corresponding to the simplified error scenario set.
5. The real-time risk dispatching method for a large-scale complex flood control system in a river basin according to claim 4 is characterized in that: The step S31 is further as follows: Step S31a, building a MATLAB distributed cluster and determining the number of CPU cores of the nodes and cluster; Step S31b, compiling a multi-task parallel module consisting of a forecast error scenario set, iteration times and population nested loops; Step S31c: To ensure load balancing of scheduling parallelism, the forecast error scenario set of the outermost loop is allocated to the cluster nodes; Step S31d, based on the forecast error scenario set allocated in step S31c, allocating the number of MA iterations of the middle layer to different threads in each node; Step S31e: Using the threads allocated in step S31d, parallelly calculate the MA population within the range of MA iteration times corresponding to each thread.
6. The real-time risk dispatching method for a large-scale complex flood control system in a river basin according to claim 4 is characterized in that: The step S32 is further as follows: Step S32a, taking the outflow of the reservoir as the initial population, introducing a sliding operator to perform sliding average processing on the initial population to ensure the smoothness of the outflow flow process; Step S32b, calculate the system objective function, evaluate the fitness functions of all populations, and introduce external archives to screen non-inferior solution sets; Step S32c, determining the iteration condition, if not satisfied, updating the speed and position of the female and male mayflies; Step S32d, evaluate the solution, if it is a non-dominated solution, update the archive, otherwise there is a 50% probability of updating the archive; Step S32e, introducing the Debye Waller factor to perform stable sorting on the mayflies, mating and evaluating the offspring, and determining whether the offspring is a non-dominated solution of its parent of the same sex, and if so, storing the non-dominated solution in an external archive; Step S32f, satisfying the iteration condition, retaining the final non-dominated solution set as the real-time flood control scheduling solution set.
7. The real-time risk dispatching method for a large-scale complex flood control system in a river basin according to claim 1 is characterized in that: In step S4, the typical scheduling scheme corresponding to the reduced error scenario set is extracted based on the clustering method of system target decision-physical attribute-spatial geometric feature coupling, which is further: Step S41: based on the flood control objectives of the Z subsystems of large-scale complex flood control in the basin, a set of scheduling schemes is preferentially selected to obtain a set of decision solutions for each subsystem; Step S42, according to the physical properties of the solution set, the intra-cluster spacing of the spatial point set distribution is as small as possible, and the inter-cluster spacing is as large as possible, and the decision solution set of each subsystem is divided into M categories using Fuzzy C-Means Clustering; Step S43, using an adaptive curvature entropy point cloud simplification algorithm to calculate the curvature entropy in the M-type solution set, removing redundant points, keeping the spatial characteristics of the solution set unchanged, and extracting b% representative point sets from the M-type solution set; b is a preset value; Step S44, the point set finally extracted is a typical scheduling solution based on clustering decision-making coupled with system goal decision-physical attribute-spatial geometric feature.
8. Real-time risk dispatching system for large-scale complex flood control systems in river basins, characterized by: include: at least one processor; and a memory communicatively coupled to at least one of the processors; Wherein, the memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the real-time risk scheduling method for a large-scale complex flood control system in a river basin as described in any one of claims 1 to 7.
Citation Information
Patent Citations
Determination method for flood prevention forecast scheduling risk of reservoir
CN110991850A
Flood control system reservoir attribute real-time dynamic evaluation method based on rough set theory
CN111401630A
Practical method for real-time flood probability prediction
CN111898831A
Reservoir group flood control dispatching intelligent method and system
CN116882684A
Plain river network water engineering cluster multi-target scheduling rule making method and system
CN117035201A
