Reservoir dispatching rule intelligent extraction method

Abstract

The invention belongs to the field of flood and drought disaster prevention and water resource management and allocation, and relates to a reservoir scheduling rule intelligent extraction method, which comprises the following steps of: 1) performing scheduling mode classification on an optimal water storage and drainage strategy obtained by deterministic optimization scheduling by adopting a double-layer clustering method; 2) extracting corresponding scheduling rules of different scheduling modes; 3) adopting a two-step method to judge a scheduling mode which is most suitable for being adopted in the face of a time period; and 4) according to the scheduling mode which is determined in the step 3) and is most suitable for being adopted, making a decision on the discharged flow of the confronting time interval. The invention provides the intelligent extraction method for the reservoir scheduling rule, which improves the determination accuracy and can provide technical support for the development of the comprehensive benefits of the reservoir.

Description

Technical Field

[0001] The invention belongs to the field of flood and drought prevention and water resources management and allocation, and relates to an intelligent extraction method for reservoir dispatching rules, and in particular to an intelligent extraction method for reservoir dispatching rules based on double-layer clustering and two-step discrimination. Background Art

[0002] Reservoir operation rules are an indispensable technical guide for reservoir operation. Usually, operation rules are extracted based on experience or statistical analysis of the results of reservoir deterministic optimization operation (commonly known as implicit random optimization method). Deterministic optimization operation is to calculate the optimal storage and release strategy of the reservoir based on the historical inflow runoff sequence using deterministic optimization methods such as dynamic programming. Due to the randomness and non-repetitiveness of the future inflow runoff series of the reservoir, the results of deterministic optimization operation are often difficult to directly guide the actual operation of the reservoir. Therefore, it is necessary to analyze the relationship and characteristics between the historical optimal storage and release strategy of the reservoir and the historical water inflow and water demand of the reservoir, and extract a set of operation rules suitable for guiding the future operation of the reservoir. By mining the results of reservoir deterministic optimization operation, reservoir operation regulations with practical guiding significance can be extracted, which can effectively enhance the practicality of theoretical optimization operation results, provide more accurate decision support for the actual operation of the reservoir, and maximize the comprehensive benefits of multifunctional reservoirs.

[0003] However, it is difficult for traditional implicit stochastic optimization methods to accurately express the nonlinear relationship between decision variables such as reservoir discharge flow, reservoir water level and inflow when extracting reservoir scheduling rules. Artificial intelligence models such as artificial neural networks, random forests and deep learning are significantly better than traditional implicit stochastic optimization methods in describing this nonlinear relationship, and have achieved remarkable results in the practice of extracting reservoir scheduling rules. However, there are two problems with these artificial intelligence models, so that their accuracy and practicality need to be improved: 1) In the extraction of scheduling rules, it is very easy to generate new problems that violate scheduling common sense, such as not satisfying the upper and lower boundary constraints of water level and not satisfying the upper and lower limit constraints of discharge flow; 2) There is a problem that the same artificial intelligence model is used for scheduling rules in different scheduling periods such as the pre-flood drawdown period and the post-flood water storage period, resulting in the need to improve the generalization ability. Summary of the invention

[0004] In order to solve the above two technical problems existing in the background technology, the present invention provides a new dispatching rule extraction method, which not only improves the simulation accuracy of the dispatching rules but also enhances the rationality of the dispatching rules, reduces the probability of violating the reservoir dispatching constraints, and can provide technical support for the comprehensive benefits of the reservoir.

[0005] In order to achieve the above object, the present invention adopts the following technical solution:

[0006] A method for intelligently extracting reservoir dispatching rules, characterized in that the method comprises the following steps:

[0007] 1) The double-layer clustering method is used to classify the dispatching modes of the optimal water storage and release strategy obtained by deterministic optimal dispatching;

[0008] 2) Extract corresponding scheduling rules for different scheduling modes;

[0009] 3) Use a two-step method to determine the most suitable scheduling mode for the time period;

[0010] 4) Make decisions on the discharge flow during the period.

[0011] Preferably, the specific implementation method of step 1) adopted in the present invention is:

[0012] 1.1) Obtain the optimal water storage and release strategy obtained by deterministic optimization scheduling;

[0013] 1.2) Adopting the aggregate hierarchical clustering method, considering the similarity of reservoir inflow and outflow, the dispatching mode of the entire optimal storage and release strategy is preliminarily clustered to determine the major types of dispatching modes;

[0014] 1.3) Using the subtractive clustering method, the large categories of scheduling patterns determined in step 1.2) are divided into smaller subcategories, each of which represents a scheduling pattern, taking into account the subtle differences in scheduling period, inflow and outflow.

[0015] Preferably, the specific implementation method of step 1.1) adopted in the present invention is to optimize the time-based runoff sequence using mathematical programming models such as linear programming, dynamic programming, and two-level programming. If a two-level programming model is used, the upper layer can aim at minimizing the flood control reservoir capacity occupancy rate, and the lower layer can aim at maximizing the annual power generation. The particle swarm algorithm can be used to solve the two-level programming model to obtain the optimal water storage and release strategy.

[0016] Preferably, the expression of the major category of the scheduling mode in step 1.2) adopted by the present invention is:

[0017]

[0018] in:

[0019] Q in (t) is the reservoir inflow during period t;

[0020] Q out (t) is the outflow from the reservoir during period t in the optimal storage and release strategy;

[0021] AHC(k) is the aggregate hierarchical clustering method to obtain the k-th scheduling pattern category;

[0022] AHC is agglomerative hierarchical clustering;

[0023] T is the total number of time periods in the historical runoff series.

[0024] Preferably, the expression of the subclass in step 1.3) adopted by the present invention is:

[0025]

[0026] in:

[0027] Q in (t) is the reservoir inflow during period t;

[0028] Q out (t) is the outflow from the reservoir during period t in the optimal storage and release strategy;

[0029] Z t is the reservoir water level in period t in the optimal storage and release strategy;

[0030] D t is the specific date of period t;

[0031] C(i, m) is the mth type of scheduling mode after subdividing AHC(i) by the subtractive clustering method;

[0032] SCM is a subtractive clustering method.

[0033] Preferably, the dispatching modes in step 1) adopted in the present invention include AHC (1) in which the outflow is almost equal to the inflow, AHC (2) in which the downstream flow is fixed, and AHC (3) in which the outflow is a nonlinear function of the inflow.

[0034] Preferably, when the main category of the scheduling mode adopted in the present invention is AHC(1) in which the outflow and the inflow are almost equal, the subcategory corresponding to AHC(1) in which the outflow and the inflow are almost equal is C(1,1), and the step 2) uses the following expression to extract the corresponding scheduling rules of different scheduling modes:

[0035] Q d (t) = Q in (t)

[0036] in:

[0037] Q d (t) is the simulated outbound flow in period t under the extraction scheduling mode;

[0038] Q in (t) is the reservoir inflow during period t.

[0039] Preferably, when the main category of the dispatching mode adopted in the present invention is AHC (2) with a fixed downstream flow, the AHC (2) with a fixed downstream flow adopts different fixed downstream flows according to the sub-category C (2, m), and the step 2) adopts the following expression to extract the corresponding dispatching rules of different dispatching modes:

[0040] Q d (t)∈{Q min , Q max , Q3, …, Q n}

[0041] in:

[0042] Q d (t) is the simulated outbound flow in period t under the extraction scheduling mode;

[0043] Q min is the minimum outflow of the reservoir;

[0044] Q max is the maximum outflow of the reservoir;

[0045] Q n It is the fixed discharge flow of the reservoir to meet other functional needs.

[0046] Preferably, when the scheduling mode used in the present invention is AHC (3) in which the outflow is a function of the inflow, the AHC (3) in which the outflow is a function of the inflow is simulated using an extreme learning machine for each subclass C (3, m), and the step 2) uses the following expression to extract the corresponding scheduling rules of different scheduling modes:

[0047] Q d (t) = ELM[Q in (t), D(t)]

[0048] in:

[0049] It's Q d (t) is the simulated outbound flow in period t under the extraction scheduling mode;

[0050] Q in (t) is the reservoir inflow during period t;

[0051] It is the extraction scheduling mode D t is the specific date of period t;

[0052] ELM is extreme learning machine.

[0053] Preferably, the specific implementation method of step 3) adopted by the present invention is:

[0054] 3.1) Use decision tree and hidden Markov model to predict the scheduling mode to be adopted in the facing period;

[0055] 3.2) Adopt the adaptive enhancement algorithm to make a final judgment on the prediction result obtained in step 3.1) to obtain the most suitable scheduling mode for the time period.

[0056] Preferably, the specific implementation method of step 4) adopted in the present invention is to calculate the downstream flow based on the information such as reservoir inflow, reservoir water level and scheduling period in the facing period and the scheduling rules corresponding to the most suitable scheduling mode determined in step 3).

[0057] The advantages of the present invention are:

[0058] The present invention provides a method for intelligently extracting reservoir dispatching rules, and in particular, relates to a method for intelligently extracting reservoir dispatching rules based on double-layer clustering and two-step discrimination. The double-layer clustering successively adopts the aggregation hierarchical clustering method and the subtraction clustering method to realize the fine classification of the dispatching mode contained in the deterministic optimization dispatching results, and the two-step discrimination method successively adopts the decision tree-hidden Markov model and the adaptive enhancement algorithm to realize the accurate discrimination of the dispatching mode in the facing period. The present invention fully considers the characteristics that the reservoir outflow-inflow relationship and the dispatching mode are quite different due to the different dispatching targets in different dispatching periods such as the pre-flood drawdown period, the flood season, the post-flood water storage period, and the dry season. The double-layer clustering method is used to finely classify the dispatching modes and the reservoir outflow-inflow relationship in different dispatching periods of the deterministic optimization dispatching results, and different models are used to extract the dispatching rules under different dispatching modes respectively. The two-step discrimination method is used to determine the dispatching mode and its corresponding dispatching rules that the reservoir should adopt in the facing period in the actual dispatching process, so as to finally determine the decision on the discharge flow of the reservoir in the facing period. Compared with single-layer clustering methods such as k-means and fuzzy clustering, the double-layer clustering method of the present invention can perform more detailed classification of the dispatching mode of different dispatching periods and the relationship between the outflow and inflow of the reservoir, and use different models to extract the dispatching rules under different dispatching modes respectively, which lays a good model structure foundation for effectively improving the simulation accuracy of dispatching rules, guarantees the scientificity and rationality of pattern classification to a greater extent, fully explores the characteristics and regularity of clustering data, can extract dispatching rules that are more in line with reality, and provide technical support for the comprehensive benefits of reservoirs. The two-step discrimination method for the dispatching mode in the face of the time period provided by the present invention greatly reduces the error rate of pattern discrimination. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a flow chart of the reservoir dispatching rule extraction method provided by the present invention;

[0060] Figure 2 It is a schematic diagram of the process of classifying the dispatching modes of the optimal water storage and release strategy using the double-layer clustering method adopted by the present invention;

[0061] Figure 3 It is a schematic diagram of the annual distribution of the dispatching mode categories extracted by the present invention and the optimal-simulated discharge process; DETAILED DESCRIPTION

[0062] The specific implementation of the method of the present invention is described in detail below through examples and in conjunction with the accompanying drawings.

[0063] See also Figure 1 The present invention proposes an intelligent extraction method for reservoir dispatching rules, and the specific steps are as follows:

[0064] Step 1: Classification of dispatching modes. That is, the optimal water storage and release strategy obtained by deterministic optimization dispatching is classified by using a double-layer clustering method, taking into account the relationship between the optimal water storage and release strategy and influencing factors such as the inflow rate. Figure 1 As shown, the specific solution steps included in this step are as follows:

[0065] Step 1.1, establish a deterministic optimization model to solve the optimal water storage and release strategy. In this embodiment, the historical inflow runoff sequence is optimized using a two-level programming (BLP) model. The upper layer aims to minimize the flood control reservoir capacity occupancy rate, and the lower layer aims to maximize the annual power generation. The particle swarm algorithm (PSO) is used to solve the two-level programming model.

[0066] Step 1.2: Considering the similarity of reservoir inflow and outflow, the Agglomerative Hierarchical Clustering method (AHC) is used to preliminarily classify the dispatching modes into k dispatching mode categories AHC(k):

[0067]

[0068] Among them, Q in (t) is the reservoir inflow during period t, Q out (t) is the outflow of the reservoir in period t in the optimal water storage and release strategy, and AHC(k) is the kth type of scheduling mode obtained by the aggregated hierarchical clustering method.

[0069] Usually, a reservoir has at least three major types of operation modes, including AHC (1) in which the outflow and inflow are almost equal, AHC (2) in which the outflow is a function of the inflow, and AHC (3) in which the outflow is a function of the inflow.

[0070] Step 1.3, considering the subtle differences within the scheduling mode category AHC(k), the subtractive clustering method (SCM) is used to finely classify the scheduling modes, so that each category is aggregated into 1 to m scheduling modes:

[0071]

[0072] Among them, Z t is the reservoir water level in period t of the optimal storage and release strategy, D t is the specific date of period t, and C(i, m) is the mth type of scheduling mode after AHC(i) is subdivided by the subtractive clustering method.

[0073] Step 2: Extract dispatch rules. That is, select appropriate model algorithms based on the characteristics of each dispatch mode to extract the corresponding dispatch rules of different dispatch modes. For the three major types of dispatch modes, the appropriate algorithms for reservoirs are as follows:

[0074] For the dispatch mode category AHC(1) where the outflow and inflow are almost equal, the corresponding subcategory C(1,1) uses the identity function to determine the outflow flow, which is specifically expressed as:

[0075] Q d (t) = Q in (t)

[0076] Where: Q d (t) is the simulated outbound flow in period t under the extraction scheduling mode;

[0077] For the dispatch mode category AHC(2) with a fixed downstream flow rate in a specific dispatch period, the outflow flow rate adopts different fixed downstream flows according to the sub-category C(2, m). The specific expression is:

[0078] Q d (t)∈{Q min , Q max , Q3, …, Q m}

[0079] Where: Q min is the minimum discharge flow of the reservoir, Q max is the maximum discharge flow of the reservoir, Q m Fixed discharge flow to meet other functional requirements of the reservoir;

[0080] For the scheduling mode category AHC(3) where the outflow is a function of the inflow, each subcategory C(3, m) is simulated using an extreme learning machine (ELM). The specific expression is:

[0081] Q d (t) = ELM[Q in (t), D(t)]

[0082] Step 3: Dispatch mode determination. That is, a two-step method is used to determine the most suitable dispatch mode for the period based on the inflow, water level and dispatch period information of the period. Figure 1 As shown, the specific solution steps included in this step are as follows:

[0083] Step 3.1, input the information of inflow, water level and dispatching period and the classification results of dispatching mode into the decision tree (DT) model to make a preliminary judgment on the dispatching mode;

[0084] Step 3.2, input the information of inflow, water level and dispatching period and the classification results of dispatching mode into the hidden Markov model (HMM) to make a preliminary judgment on the dispatching mode;

[0085] Step 3.3, input the preliminary judgment results of step 3.1 and step 3.2 into the adaptive boost algorithm (AdaBoost) to make a final judgment on the scheduling mode to be adopted in the time period, and obtain the most suitable scheduling mode for the time period.

[0086] Step 4: Decision on the discharge flow of the reservoir. That is, the discharge flow is calculated based on the information of reservoir inflow, reservoir water level and dispatching period in the facing period and the dispatching rules corresponding to the most suitable dispatching mode determined in step 3.3).

[0087] Taking the extraction of the dispatching rules of the Three Gorges Reservoir as an example, the effectiveness and rationality of the invented method are illustrated. The Three Gorges Reservoir has the functions of flood control, power generation, shipping, water resource utilization and ecological environment protection. In the actual dispatching operation, in order to comprehensively consider the above functions, four relatively obvious dispatching periods are formed, namely the pre-flood drawdown period, the flood season, the post-flood water storage period and the dry season. Each dispatching period has a relatively fixed dispatching mode according to the upstream and downstream water inflow and downstream water demand. The embodiment adopts a total of 30 years (including 20 years of training and 10 years of verification) of daily inflow runoff sequence to classify the dispatching mode of the Three Gorges Reservoir, and extracts the dispatching rules under each type of dispatching mode, and calculates the corresponding downstream flow. Figure 2 The clustering process of the two-layer clustering method is shown. First, the aggregative hierarchical clustering method AHC is used to divide the strategy set data into three categories. On this basis, the subtractive clustering method SCM is used to further subdivide the strategy set data into seven types of scheduling modes. Figure 3 The temporal distribution of the seven types of dispatching modes in 2000 and the optimal storage and release process under the dispatching mode are intuitively displayed. In the figure, the C(2,1) mode is in the dry season, and the reservoir maintains the ecological base flow and discharges; the C(1,1) mode and the C(1,2) mode reservoirs discharge according to the incoming flow, so that the reservoir water level maintains the normal storage level or the flood limit water level; the C(2,2) mode is in the flood season, and the reservoir discharges according to the downstream safety discharge volume; the C(3,1) mode belongs to the pre-flood drawdown period, the reservoir discharge flow exceeds the reservoir inflow flow, and the reservoir water level gradually decreases; the C(3,2) mode and the C(3,3) mode belong to the post-flood water storage period, the reservoir discharge flow is less than the reservoir inflow flow, and the reservoir water level gradually rises. Figure 3The optimal discharge flow calculated by the deterministic optimization model in 2000 and the simulated discharge flow under the extracted dispatching rules are also shown. It can be seen from the figure that after the dispatching mode is subdivided, the dispatching rules of some dispatching modes can be extracted more accurately, and the data regularity in the deterministic optimization dispatching results is fully mined, so that the generalization ability of the model is enhanced, and the extracted dispatching rules are more in line with reality. The accuracy of the two-step method mode discrimination during the verification period reached 93.2%, the discharge flow determinism coefficient reached 0.97, and the discharge flow qualification rate (absolute error less than 10%) was 97.4%.

Claims

1. A method for intelligently extracting reservoir dispatching rules, characterized in that: The reservoir dispatching rule intelligent extraction method comprises the following steps: 1) The double-layer clustering method is used to classify the dispatching modes of the optimal water storage and release strategy obtained by deterministic optimal dispatching; 2) Extract corresponding scheduling rules for different scheduling modes; 3) Use a two-step method to determine the most suitable scheduling mode for the time period; 4) Make decisions on the discharge flow during the period.

2. The intelligent extraction method of reservoir dispatching rules according to claim 1 is characterized in that: The specific implementation method of step 1) is: 1.1) Obtain the optimal water storage and release strategy obtained by deterministic optimization scheduling; 1.2) Adopting the aggregate hierarchical clustering method, considering the similarity of reservoir inflow and outflow, the dispatching mode of the entire optimal storage and release strategy is preliminarily clustered to determine the major types of dispatching modes; 1.3) Using the subtractive clustering method, the scheduling mode categories determined in step 1.2) are divided into smaller subcategories, each of which represents a scheduling mode.

3. The intelligent extraction method of reservoir dispatching rules according to claim 2 is characterized in that: The specific implementation method of step 1.1) is to optimize the two-level programming model based on the time-based runoff sequence. The upper level aims to minimize the flood control reservoir capacity occupancy rate, and the lower level aims to maximize the annual power generation. The particle swarm algorithm is used to solve the two-level programming model to obtain the optimal water storage and release strategy.

4. The intelligent extraction method of reservoir dispatching rules according to claim 3 is characterized in that: The expression of the major category of scheduling mode in step 1.2) is: in: Q in (t) is the reservoir inflow during period t; Q out (t) is the outflow from the reservoir during period t in the optimal storage and release strategy; AHC(k) is the k-th scheduling pattern category obtained by the aggregative hierarchical clustering method; AHC is agglomerative hierarchical clustering; T is the total number of time periods in the historical runoff series.

5. The intelligent extraction method of reservoir dispatching rules according to claim 4 is characterized in that: The expression of the subclass in step 1.3) is: in: Q in (t) is the reservoir inflow during period t; Q out (t) is the reservoir outflow at time t in the optimal storage and release strategy; Z t is the reservoir water level in period t in the optimal storage and release strategy; D t is the specific date of period t; C(i, m) is the mth type of scheduling mode after subdividing AHC(i) by the subtractive clustering method; SCM is a subtractive clustering method.

6. The intelligent extraction method of reservoir dispatching rules according to claim 5 is characterized in that: The dispatching modes in step 1) include AHC (1) in which the outflow is almost equal to the inflow, AHC (2) in which the downstream flow is fixed, and AHC (3) in which the outflow is a nonlinear function of the inflow.

7. The intelligent extraction method of reservoir dispatching rules according to claim 6 is characterized in that: When the major category of the scheduling mode is AHC(1) where the outflow and the inflow are almost equal, the subcategory corresponding to AHC(1) where the outflow and the inflow are almost equal is C(1,1), and the step 2) uses the following expression to extract the corresponding scheduling rules of different scheduling modes: Q d (t)=Q in (t) in: Q d (t) is the simulated outbound flow in period t under the extraction scheduling mode; Q in (t) is the reservoir inflow during period t.

8. The intelligent extraction method of reservoir dispatching rules according to claim 6 is characterized by: When the dispatching mode category is AHC(2) with a fixed downstream flow, the AHC(2) with a fixed downstream flow adopts different fixed downstream flows according to the sub-category C(2, m), and the step 2) uses the following expression to extract the corresponding dispatching rules of different dispatching modes: Q d (t)∈{Q min ,Q max ,Q3,…,Q n } in: Q d (t) is the simulated outbound flow in period t under the extraction scheduling mode; Q min is the minimum outflow of the reservoir; Q max is the maximum outflow of the reservoir; Q n It is the fixed discharge flow of the reservoir to meet other functional needs.

9. The intelligent extraction method of reservoir dispatching rules according to claim 6 is characterized by: When the scheduling mode category is AHC (3) in which the outflow is a function of the inflow, the AHC (3) in which the outflow is a function of the inflow is simulated using an extreme learning machine for each subcategory C (3, m), and the step 2) uses the following expression to extract the corresponding scheduling rules of different scheduling modes: Q d (t)=ELM[Q in (t),D(t)] in: Q d (t) is the simulated outbound flow in period t under the extraction scheduling mode; Q in (t) is the reservoir inflow during period t; D t is the specific date of period t; The time period is the extraction scheduling mode, and ELM is the extreme learning machine.

10. The intelligent extraction method of reservoir dispatching rules according to claim 7, 8 or 9, characterized in that: The specific implementation method of step 3) is: 3.1) Use decision tree and hidden Markov model to predict the scheduling mode to be adopted in the facing period; 3.2) Adopt the adaptive enhancement algorithm to make a final judgment on the prediction result obtained in step 3.1) to obtain the most suitable scheduling mode for the time period.

Images