CN-121581569-B - Reservoir group scheduling rule extraction method based on runoff space-time two-dimensional random simulation

CN121581569BCN 121581569 BCN121581569 BCN 121581569BCN-121581569-B

Abstract

The invention discloses a reservoir group dispatching rule extraction method based on runoff space-time two-dimensional random simulation, which comprises the steps of optimizing runoff edge distribution functions of reservoirs one by one and time intervals, optimizing runoff Copula joint distribution functions of the same reservoir, different time intervals and the same time intervals and different reservoirs, utilizing the optimized runoff Copula joint distribution functions to combine inverse transformation sampling of a Monte Carlo method, generating a plurality of groups of conventional scene and extreme scene runoff sequence data through random simulation, establishing a reservoir group optimizing dispatching model, taking the randomly simulated plurality of groups of runoff sequence data as model input, obtaining a corresponding plurality of groups of reservoir group optimizing dispatching results through model optimizing solving, utilizing a data mining method to determine key decision factors, utilizing a machine learning method to establish a reservoir group dispatching rule model, utilizing the determined key decision factors to input a plurality of time interval state quantities at present and before and after as reservoir group dispatching rule models, predicting to obtain reservoir group dispatching operation results, and supporting decision practice.

Inventors

LI XIANG
Mi Zijia
YIN DONGQIN
Xi Xinyue
HUANG HAIBING
SHEN YANQING
SAN MEIYING
ZHANG YI
JIN SHAOBO

Assignees

中国水利水电科学研究院
青海黄河上游水电开发有限责任公司

Dates

Publication Date: 20260512
Application Date: 20251202

Claims (6)

1. The reservoir group scheduling rule extraction method based on runoff space-time two-dimensional random simulation is characterized by comprising the following steps: S1, taking a reservoir group with hydraulic and electric power connection as a research object, screening runoff edge distribution functions of reservoirs one by one and time intervals, and screening runoff Copula joint distribution functions of the same reservoir, different time intervals and the same time interval and different reservoirs based on the runoff edge distribution functions; S2, considering space-time two-dimensional characteristics of runoffs, utilizing a screened runoff Copula joint distribution function, combining inverse transformation sampling of a Monte Carlo method, and generating a plurality of groups of runoff sequence data containing all reservoirs and all time periods through multiple random simulation; s3, determining an objective function and constraint conditions of reservoir group optimization, establishing a reservoir group optimization scheduling model, taking a plurality of groups of runoff sequence data which are randomly simulated for a plurality of times as model input, and carrying out model solving by using an optimization algorithm to obtain a plurality of groups of corresponding reservoir group optimization scheduling results, wherein the corresponding reservoir group optimization scheduling results comprise all reservoirs, all time period end reservoir capacity, time period average discharging flow, time period average power generation flow, time period average water discharge flow and time period average output; s4, determining key decision factors by adopting a data mining method according to multiple groups of runoff sequence data randomly simulated for multiple times and multiple groups of reservoir group optimization scheduling calculation results corresponding to the runoff sequence data, and establishing a reservoir group scheduling rule model by utilizing a machine learning method; S5, taking state quantities of a plurality of time periods before and after the determined key decision factors as input of a reservoir group dispatching rule model, and predicting to obtain a reservoir group dispatching operation result, wherein the state quantities comprise positions of the time periods in one year, actual measurement and forecast of the storage flow of all reservoirs in the time periods before and after, actual measurement of the initial storage capacity in the current time periods before and after, and actual measurement of average discharge flow in the time periods before; Step S1 further comprises: s11, selecting various probability distribution functions, at least comprising a Normal function, a Gamma function, a Log-Normal function, a GEV function and a Weibull function, and respectively calculating runoff edge distribution functions of reservoirs and time periods; S12, calculating parameters of the runoff edge distribution functions respectively by adopting a maximum likelihood method, and performing result inspection on various distribution functions by using root mean square error as an evaluation index to screen out the most suitable runoff edge distribution functions; S13, selecting a plurality of runoff Copula joint distribution functions based on the runoff edge distribution function, wherein the runoff Copula joint distribution functions at least comprise Clayton Copula functions, frank Copula functions and Gumbel Copula functions, and the runoff Copula joint distribution functions of the same reservoir, different time periods, the same time period and different reservoirs are calculated respectively; S14, calculating parameters of the runoff Copula combined distribution functions by adopting a maximum likelihood method, and performing result inspection on the runoff Copula combined distribution functions by using root mean square error as an evaluation index to screen out the most suitable runoff Copula combined distribution functions of the same reservoir, different time periods and the same time period and different reservoirs; Solving the expression of the runoff amount of the reservoir i period t+1 when the runoff amount of the reservoir i period t is known The method comprises the following steps: Wherein, the A runoff Copula joint distribution function of any reservoir i in a period t and a period t+1; And The runoff of any reservoir i in the period t and the period t+1 are respectively; When the runoff of the time period tbank i is known, solving the expression of the runoff of the time period tbank i+1 The method comprises the following steps: Wherein, the The runoff Copula combined distribution function is used for reservoirs i and i+1 in any period of time; And The runoff of the reservoir i and the reservoir i+1 in any period t respectively; the objective function of reservoir group optimization is the maximum power generation amount, and the expression is: ; ; ; ; ; ; Wherein, the The method is characterized in that the method is a maximum target of the generated energy of a reservoir group with hydraulic power and electric power connection, i is a reservoir index, I is the total reservoir number in the reservoir group, t is the period index, T is the total time period number; in period of time for power station of reservoir i Is a power generation system; The time period is long; Is the density of water; gravitational acceleration; Generating efficiency for a power station of the reservoir i; 、 And The discharging flow of the reservoir i in the period t, the power generation flow of the power station and the water discarding flow are respectively; 、、 And The average water head, the average water level before the dam, the average water level after the dam and the average water head loss of a power station of the reservoir i in a period t are respectively; the average storage capacity of the reservoir i in the period t; And The storage capacity of the reservoir i at the beginning and the end of the period t is respectively; 、、、 And And 、、、 And Coefficients that are fitted using polynomials; the constraint conditions comprise a water quantity balance equation, an initial/termination reservoir capacity constraint, a lower drainage flow constraint, a power generation flow constraint and a hydropower station output constraint; The expression of the water balance equation is as follows: ; the initial/terminated library capacity constraint is expressed as: ; ; the expression of the reservoir capacity constraint is as follows: ; the expression of the lower leakage flow constraint is: ; The expression of the power generation flow constraint is as follows: ; the expression of the hydropower station output constraint is as follows: ; Wherein, the Is a reservoir In the time period Is a warehouse-in flow rate; Is a reservoir In the time period Is a vapor amount of (a); Is a reservoir Stock capacity at the beginning of the dispatch period; Is a reservoir Target stock capacity at the end of the dispatch period; And Respectively reservoirs In the time period Minimum and maximum storage capacities of (2); And Respectively reservoirs In the time period Minimum and maximum bleed down flows of (2); And Respectively reservoirs During the period of time Minimum and maximum power generation flow rates of (2); And Respectively reservoirs During the period of time Minimum and maximum generated power of (a).
2. The method for extracting the reservoir group scheduling rule based on the runoff space-time two-dimensional random simulation according to claim 1, wherein the step S2 is a detailed implementation method for generating a plurality of groups of runoff sequence data containing all reservoirs and all time periods, comprising the following steps: S21, generating random numbers a 1 E (0, 1) for radial flow of adjacent periods of the random simulation reservoir 1 by using a linear congruence random number generator, mapping the random numbers into random variables of target distribution by using an inverse function of a cumulative distribution function in edge distribution, and solving The runoff x 1,1 of the reservoir 1 period 1 can be obtained; Generating a random number a 2 , a 3 , ..., a T epsilon (0, 1) can obtain the following equation set: solving the equation set to obtain the runoff x 1,2 , x 1,3 , ..., x 1,T-1 , x 1,T of the reservoir 1 period 2~T; S22, generating a random number b 1 E (0, 1) by using a linear congruent random number generator for randomly simulating runoffs of adjacent periods of the reservoir 2, and enabling Knowing the runoff x 1,1 of the reservoir 1 period 1, solving the equation to obtain the runoff x 2,1 of the reservoir 2 period 1; generating a random number b 2 , b 3 , ..., b T e (0, 1) can result in the following system of equations: Solving the equation set to obtain the runoff x 2,2 , x 2,3 , ..., x 2,T-1 , x 2,T of the reservoir 2 in the period 2~T; s23, sequentially performing runoff simulation of the reservoir 3~I according to S22.
3. The method for extracting reservoir group scheduling rules based on runoff space-time two-dimensional random simulation according to claim 2, wherein the step S2 further comprises generating a plurality of sets of extreme scenario runoff sequence data containing all reservoirs and all time periods, and the implementation method further comprises: Carrying out frequency statistical analysis on historical runoff data of all reservoirs and all time periods, and defining an extreme runoff judgment threshold value of any reservoir and any time period; Identifying and marking a plurality of groups of generated conventional scene multi-reservoir long-sequence simulated runoffs by taking an extreme runoff judgment threshold value of any reservoir and any period of time as a judgment basis, and screening to obtain extreme scene multi-reservoir runoff sequence data considering space-time two-dimensional correlation; According to the extreme scene multi-reservoir runoff sequence data, obtaining a single-period extreme scene and a multi-period extreme scene by adopting time-level cluster analysis, and obtaining the single-reservoir extreme scene and the multi-reservoir extreme scene by adopting space-level cluster analysis; The single-period extreme scene is defined as single-period extreme enlargement or single-period extreme withering as long as runoff of any period reaches an extreme threshold value for a group of runoff sequences comprising all reservoirs and all periods; The multi-period extreme scene is defined as multi-period extreme enlargement, multi-period extreme withering or multi-time Duan Feng withering transfer as long as runoff of more than one period reaches an extreme threshold value for a group of runoff sequences comprising all reservoirs and all periods; The extreme situation of the single reservoir is that for a group of runoff sequences comprising all reservoirs and all time periods, as long as the runoff of any reservoir reaches an extreme threshold value, the runoff sequence is defined as extreme full of the single reservoir, extreme withered of the single reservoir or full-withered transfer of the single reservoir; The multi-reservoir extreme scene is defined as multi-reservoir extreme abundant, multi-reservoir extreme withered or multi-reservoir withered transfer as long as runoff of more than one reservoir reaches an extreme threshold value for a group of runoff sequences comprising all reservoirs and all time periods.
4. The reservoir group scheduling rule extraction method based on runoff space-time two-dimensional random simulation according to claim 1, wherein the optimization algorithm is a genetic algorithm, and the detailed implementation method comprises the following steps: A1, initializing parameters of a genetic algorithm, and respectively taking a plurality of groups of runoff sequence data which comprise all reservoirs and all time periods and are randomly simulated for a plurality of times in the step S2 as a model to input; A2, taking all reservoirs, all period average discharging flow or period end reservoir capacity as decision variables, and randomly generating N different decision variables as initial populations ; A3, calculating the fitness of each individual in the population according to an objective function of the reservoir group optimal scheduling model; a4, selecting individuals from the population by adopting a tournament selection algorithm to generate a population with the scale of N/2 ; A5, selecting a population according to a preset crossover probability Selecting individuals to cross to generate offspring individuals, and adopting offspring individuals to replace the population Is a parent of the group; a6, the population is subjected to population selection according to the preset mutation probability The individuals in the population enter the next generation population by carrying out mutation operation and substituting the original individuals by the mutated individuals Chromosome without mutation directly enters next generation population ; A7, calculating population according to objective function of reservoir group optimal scheduling model The adaptation value of each individual in the database is added with 1 to the current evolution algebra Gen; a8, judging whether the evolution algebra Gen exceeds a set maximum evolution algebra or whether all adaptation values meet a set threshold, if so, entering a step A9, otherwise, returning to the step A3; a9, selecting individuals with optimal fitness in the population as optimal decision variables, and outputting all reservoirs corresponding to the optimal decision variables, the time period end reservoir capacity of all time periods, the time period average discharging flow, the time period average generating flow, the time period average water discarding flow and the time period average output as optimal scheduling results.
5. The method for extracting reservoir group scheduling rules based on radial flow space-time two-dimensional random simulation according to claim 1, wherein the step S4 further comprises: a1, constructing a multi-scheme sample set of data mining: constructing a multi-scheme sample set for data mining by adopting a plurality of groups of runoff series data randomly simulated for a plurality of times and a corresponding plurality of groups of reservoir group optimization scheduling calculation results, wherein each group of scheme samples comprises all reservoirs, all period initial reservoir capacities, period end reservoir capacities, period average warehousing flows, period average discharging flows, period average generating current scenes, period average water discarding flows and period average output; A2, determining key decision factors by adopting a data mining method: B1, calculating gray correlation degree of each decision factor Selecting the average drainage flow of all the time periods or the end storage capacity of the time periods of all the reservoirs as decision variables, selecting the positions of the time periods in one year, the storage flow of a plurality of time periods before and after the current time periods, the initial storage capacity of a plurality of time periods before and the average drainage flow of a plurality of time periods of all the reservoirs as candidate decision factors, and calculating gray association degrees between the candidate decision factors and the decision variables one by one, wherein the specific expression is as follows: Wherein, the The gray correlation degree of the candidate decision factor j is in a value range of 0, 1; the gray correlation coefficient of the candidate decision factor j when n is recorded; N is the total record number; The value at record n for the decision variable; the value of the candidate decision factor j when n is recorded; Is a resolution coefficient; B2, setting grey correlation threshold Select to satisfy Is taken as a key decision factor; a3, constructing a multi-scheme sample set of the machine learning model according to the decision variables and the key decision factors, dividing the multi-scheme sample set into a training set and a testing set according to a preset proportion, and training and verifying the machine learning model by adopting the training set and the testing set to obtain a reservoir group scheduling rule model; The verification method comprises the steps of adopting root mean square error, average absolute error and judgment coefficient as evaluation indexes to evaluate the effectiveness of the reservoir group scheduling rules of the machine learning method, and adopting reservoir capacity constraint out-of-limit rate and drainage flow out-of-limit rate as evaluation indexes to evaluate the feasibility of the reservoir group scheduling rules of the machine learning method.
6. The method for extracting reservoir group scheduling rules based on radial-flow space-time two-dimensional random simulation according to claim 5, wherein the machine learning model is LightGBM algorithm, and the optimal parameter combination of LightGBM algorithm is searched by adopting Bayesian optimization strategy.

Description

Reservoir group scheduling rule extraction method based on runoff space-time two-dimensional random simulation Technical Field The invention belongs to the field of reservoir group optimal scheduling, and particularly relates to a reservoir group scheduling rule extraction method based on runoff space-time two-dimensional random simulation. Background The reservoir engineering has comprehensive utilization benefits of flood control, water supply, power generation, ecology, shipping and the like. The reservoir group in the same system has the connection of water power, electric power, hydrology and the like, and the social, economic and ecological benefits can be maximized by developing the joint scheduling. In order to optimize the combined operation mode of the reservoir group, a large number of optimization solving methods are developed in the theoretical level in the past decades, including Linear Programming (LP), nonlinear programming (NLP), dynamic Programming (DP), intelligent optimization and the like. However, reservoir optimization scheduling theory results are difficult to directly apply in a practical level, and still depend on the experience of a scheduler to combine reservoir scheduling rules. Reservoir dispatching rules are important bases for guiding reservoir dispatching operation in practice. The traditional reservoir dispatching rule form mainly comprises a dispatching diagram and a dispatching function, and the method is relatively visual and easy to operate, so that the method is widely applied. The reservoir dispatching diagram generally takes time as a horizontal axis and water level as a vertical axis, the diagram is divided into different operation areas through key control lines, the areas correspond to different dispatching strategies, a manager is helped to make scientific decisions according to real-time water conditions, and maximization of comprehensive benefits of the reservoir and minimization of risks are achieved. The scheduling function describes the scheduling strategy in a preset display function form (such as multiple linear regression, etc.), and the mapping relationship between the reservoir scheduling decision variable (average discharging flow in the current period/end reservoir capacity in the current period) and the decision factor (the reservoir storage flow in the current period, the reservoir storage flow in the previous period, the reservoir storage capacity in the previous period, the discharging flow in the previous period, etc.) is established by comparing and selecting different function forms for multiple times, so as to guide the reservoir operation. In recent years, with the rapid development of artificial intelligence technology, reservoir dispatching rules are extracted based on data mining and machine learning (artificial neural network ANN, support vector machine SVM, random forest RF, long-short-term memory network LSTM and the like) to guide reservoir dispatching operation, and as the problems of randomness, nonlinearity, multiple variables and the like of reservoir dispatching decisions can be effectively solved, the data-driven research method starts to emerge continuously. The data mining technique can be used for analyzing and establishing the association relation between the decision variable and each decision factor, thereby laying a foundation for extracting the scheduling rule by a machine learning method, the method does not need to preset a fixed function form, but self-learns and identifies potential scheduling rules based on input data through strong mapping capability of the algorithm, and automatically generates an adaptive scheduling strategy by adjusting network structures and parameters of the model. Reservoir scheduling rule extraction based on data mining and machine learning can exhibit greater flexibility and adaptability than conventional scheduling methods. The limitation of the prior art is mainly that firstly, the traditional reservoir dispatching rules are generally established by utilizing historical design runoff data, and the runoff change caused by climate change and human activity influence, namely, the uncertainty and the randomness of the runoff are difficult to respond in time. For example, current Longsheep isthmus reservoir design schedules were compiled in 1998, and runoff sequence changes have disjointed the scheduling rules based on the hydrologic consistency assumption from the actual reservoir schedule. Secondly, the traditional reservoir dispatching rules are usually established in a multiple linear regression equal display function form, so that the characteristics of nonlinearity, multivariable and the like of reservoir dispatching decisions are difficult to effectively reflect. Thirdly, a reservoir dispatching rule is established by combining data mining with a machine learning method, and is generally dependent on a long-series hydrologic data. However, it is difficult to change runoff scenes in the s