A method, system and electronic equipment for optimal scheduling of reservoir groups

Abstract

This invention discloses a method, system, and electronic equipment for optimal scheduling of a group of reservoirs. First, it obtains different combinations of initial and final water levels, inflow rates, and reservoir output under different combinations. Then, it uses the LSTM algorithm to simulate the mapping between the initial and final water levels, inflow rates, and output, constructing a response surface model. Finally, it discretizes the water levels of each reservoir during the scheduling period, calculates the reservoir output under different discrete state combinations at each stage using the response surface model, and calls a dynamic programming algorithm to calculate the optimal scheduling scheme. This invention uses the LSTM method to construct the response surface, reducing redundant calculations in the original DP recursive equations, thereby significantly improving the computational efficiency of reservoir scheduling while ensuring the quality of the scheduling scheme.

Description

Technical Field

[0001] This invention belongs to the field of efficient water resource utilization and optimized scheduling of hydropower systems, specifically relating to a method, system and electronic equipment for optimized scheduling of reservoir groups. Background Technology

[0002] Due to rapid economic growth, China's demand for electricity has increased significantly in recent years, leading to a surge in the development of hydropower resources. By the end of 2022, the total installed capacity of hydropower exceeded 413.5 GW, accounting for approximately 16.1% of the total installed capacity of all energy sources and approximately 35.3% of the installed capacity of clean energy. The total power generation from hydropower exceeded 1352.2 TWh, providing approximately 53.2% of clean energy, and hydropower has advantages in relative reliability and stability compared to other energy sources. Furthermore, in the era of carbon peaking and carbon neutrality, hydropower systems will maintain rapid growth to meet the increasing demand for clean and sustainable energy. However, the rapid expansion and capacity increase of the power system have made the computational modeling of reservoir scheduling and operation extremely difficult, posing a significant challenge to power system operation. Therefore, reservoir scheduling has become one of the most critical and demanding engineering tasks in China's power system.

[0003] Dynamic programming (DP) is a well-known optimization algorithm that performs well in solving multi-stage decision optimization problems with staged and nonlinear characteristics. Since the benefits or costs of each stage of the problem are separable and monotonic, the original multi-stage problem can be decomposed into several relatively independent and simple subproblems. These subproblems can then be solved recursively using DP to obtain the global optimum in the discrete space. The reservoir scheduling problem conforms to a standard multi-stage decision process, and DP is widely used for optimal reservoir scheduling decisions. In the DP method, the state of the reservoir (e.g., water level) is first discretized into many disjoint states, and then potential combinations of discrete states are traversed to determine the global optimum. However, as the number of discrete states and reservoirs increases, the computational complexity of the DP method in multi-dimensional decision problems grows exponentially, leading to the well-known dimensionality problem. In other words, the execution time of the DP method in practical applications is very long, and the memory usage required for computation may exceed the maximum processing power of a typical computer. To alleviate the dimensionality problem, many dynamic programming (DP) variants have been successfully developed in recent years, such as POA (Progressive Optimization Algorithm), DDDP (Discrete Differential Dynamic Programming), and DPSA (Dynamic Programming with Successive Approximation). While these DP variants can improve computational efficiency to varying degrees, the search process remains sensitive to the initial operating trajectory of each reservoir, making it challenging to ensure a globally optimal solution. Therefore, developing improved strategies that balance the computational efficiency and optimality of DP methods for reservoir operation is crucial. Summary of the Invention

[0004] The technical problem to be solved by this invention is: to address the problem of excessive computational complexity in existing DP algorithms, a reservoir group optimization scheduling method coupled with LSTM and dynamic programming is provided, which better reduces redundant calculations in the original DP recursive equations, thereby significantly improving the execution efficiency of reservoir operation.

[0005] To solve the above technical problems, the present invention provides the following technical solution: a method for optimal scheduling of a reservoir group, comprising the following steps:

[0006] S1, according to the reservoir The upper and lower limits of water level and the maximum and minimum values ​​of historical inflow are used to discretize the state of water level and flow, obtaining a dataset containing different combinations of initial water level, final water level, and inflow. Then, using the dataset, the water balance equation and reservoir characteristic curves are used to calculate the reservoir under different combinations. Exertion, of which, ; This refers to the number of cascade reservoir groups.

[0007] S2. Using the dataset obtained in step S1, and the reservoirs under different combinations The power output is simulated using the LSTM algorithm to map the initial and final water levels, inflow rate, and power output of the reservoir, thus constructing a model of the reservoir. The response surface model; let Return to step S1 to construct the response surface model for the next reservoir, until... ;

[0008] S3, During the scheduling period The water levels of each reservoir are discretized, and the estimated water levels of the reservoirs under different combinations of discrete states at each stage are calculated using a response surface model. Output effort, and use dynamic programming algorithm to calculate the optimal efficiency for the stage.

[0009] Furthermore, the aforementioned step S1 includes the following sub-steps:

[0010] S1.1, During the scheduling and operation period, the reservoir... The initial and final water levels are discrete within their upper and lower limits. There are points, represented as follows: and ,in, and They represent the first The initial water level discrete point and the first One final water level discrete point;

[0011] S1.2. Based on historical inbound flow statistics, the inbound flow is discretized to the maximum flow rate. and minimum flow Between 1 point, represented as ,in For reservoir The Individual discrete points of inbound flow;

[0012] S1.3 Obtain the discrete point combination of water level and flow rate The reservoir flow rate is obtained by substituting it into the outflow balance equation in formula (1), and then the downstream water level and head are obtained by using the tailwater level-flow curve. Finally, the reservoir output value for each combination is calculated using the output formula in formula (2). ;

[0013] (1)

[0014] (2)

[0015] in the formula For outbound flow, Because of the long duration, This is the output coefficient. The difference in water head between the upstream and downstream of the reservoir.

[0016] Furthermore, the aforementioned step S2 includes the following sub-steps:

[0017] S2.1, Based on the combination of discrete points of water level and flow rate and the reservoir output value for each combination. Build dataset ;

[0018] S2.2. Using the initial water level, the final water level, and the inflow rate as inputs, and the reservoir output as output, a response surface model is constructed and trained using the LSTM algorithm. .

[0019] S2.3, Repeat steps S1 to S2 to build the next reservoir. The response surface model, until .

[0020] Furthermore, the aforementioned step S3 includes the following sub-steps:

[0021] S3.1 Initialization The number of discrete states is , For the number of stages, Based on inbound flow Initial water level , and the end water level , Perform state combination, where , Given the state vectors formed by the initial and final water levels of each reservoir, the entire set of states is calculated using a response surface model. Corresponding reservoir output ;

[0022] S3.2 Calculate based on the recursive equation of the response surface model Optimal benefit at each stage:

[0023] ,

[0024] In the formula, LSTM represents the response surface model. Let t be the initial state set of the reservoir group at stage t. For reservoir group Phase end state set, for The optimal benefit function for a given stage. for The optimal benefit function for a given stage;

[0025] S3.3、 Then determine whether t is equal to T. If t is equal to T, the optimal scheduling scheme for the reservoir group within the entire scheduling range is obtained. If t is not equal to T, return to step S3.1.

[0026] Another aspect of the present invention proposes a reservoir group optimization scheduling system, comprising: a response surface modeling data acquisition module, a response surface model construction module, and a response surface dynamic programming algorithm module;

[0027] The response surface modeling data acquisition module, based on the reservoir The upper and lower limits of water level and the maximum and minimum values ​​of historical inflow are used to discretize the state of water level and flow, obtaining a dataset containing different combinations of initial water level, final water level, and inflow. Then, using the dataset, the water balance equation and reservoir characteristic curves are used to calculate the reservoir under different combinations. Exertion, of which, ; This refers to the number of cascade reservoir groups.

[0028] The response surface model construction module is used to utilize the dataset obtained by the response surface modeling data acquisition module, as well as reservoirs under different combinations. The power output is simulated using the LSTM algorithm to map the initial and final water levels, inflow rate, and power output of the reservoir, thus constructing a model of the reservoir. The response surface model; let The response surface modeling data acquisition module is then returned to execute, and the response surface model is constructed for the next reservoir until... .

[0029] The response surface dynamic programming algorithm module, during the scheduling period The water levels of each reservoir are discretized, and the estimated water levels of the reservoirs under different combinations of discrete states at each stage are calculated using a response surface model. Output effort, and use dynamic programming algorithm to calculate the optimal efficiency for the stage.

[0030] The present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any of the methods of the present invention.

[0031] The present invention also proposes a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method described in the present invention.

[0032] Compared with the prior art, the beneficial technical effects of the present invention using the above technical solution are as follows:

[0033] 1. Compared with DP using traditional computation methods, the RSDP algorithm uses the response surface constructed by LSTM to reduce redundant calculations in the original DP recursive equation, thereby significantly improving the execution efficiency of reservoir operation.

[0034] 2. Mathematically, the established response surface model is a high-dimensional nonlinear mapping function that reflects the complex relationship between reservoir output and influencing factors (initial water level, final water level, and inflow).

[0035] 3. Since the initial water level, final water level, and inflow are known, other relevant information about reservoir operation and scheduling, such as reservoir output and outflow, can be determined based on the response surface model. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the response surface model construction.

[0037] Figure 2 This is a schematic diagram of the RSDP method. Detailed Implementation

[0038] To better understand the technical content of the present invention, specific embodiments are described below in conjunction with the accompanying drawings.

[0039] In this invention, various aspects of the invention are described with reference to the accompanying drawings, in which numerous illustrative embodiments are shown. Embodiments of the invention are not limited to those depicted in the drawings. It should be understood that the invention is implemented through any of the various concepts and embodiments described above, as well as the concepts and embodiments described in detail below, because the concepts and embodiments disclosed herein are not limited to any particular implementation. Furthermore, some aspects of the invention disclosed may be used alone or in any suitable combination with other aspects of the invention disclosed.

[0040] This invention provides a method for optimal scheduling of a group of reservoirs, comprising the following steps:

[0041] S1, according to the reservoir The upper and lower limits of water level and the maximum and minimum values ​​of historical inflow are used to discretize the state of water level and flow, obtaining a dataset containing different combinations of initial water level, final water level, and inflow. Then, using the dataset, the water balance equation and reservoir characteristic curves are used to calculate the reservoir under different combinations. Exertion; among them, ; This refers to the number of cascade reservoir groups.

[0042] S2. Using the dataset obtained in step S1, and the reservoirs under different combinations The power output is simulated using the LSTM algorithm to map the initial and final water levels, inflow rate, and power output of the reservoir, thus constructing a model of the reservoir. The response surface model; let Return to step S1 to construct the response surface model for the next reservoir, until... .

[0043] S3, During the scheduling period The water levels of each reservoir are discretized, and the estimated water levels of the reservoirs under different combinations of discrete states at each stage are calculated using a response surface model. Output effort, and use dynamic programming algorithm to calculate the optimal efficiency for the stage.

[0044] Furthermore, the aforementioned step S1 includes the following sub-steps:

[0045] S1.1, During the scheduling and operation period, the reservoir... The initial and final water levels are discrete within their upper and lower limits. There are points, represented as follows: and ,in, and They represent the first The initial water level discrete point and the first One final water level discrete point;

[0046] S1.2. Based on historical inbound flow statistics, the inbound flow is discretized to the maximum flow rate. and minimum flow Between 1 point, represented as ,in For reservoir The Individual discrete points of inbound flow;

[0047] S1.3 Obtain the discrete point combination of water level and flow rate The reservoir flow rate is obtained by substituting it into the water balance equation in formula (1), and then the downstream water level and head are obtained by using the tailwater level-flow curve. Finally, the reservoir output value for each combination is calculated using the output formula in formula (2). ;

[0048] (1)

[0049] (2)

[0050] in the formula For outbound flow, Because of the long duration, This is the output coefficient. The difference in water head between the upstream and downstream of the reservoir.

[0051] Furthermore, the aforementioned step S2 includes the following sub-steps:

[0052] S2.1, Based on the combination of discrete points of water level and flow rate and the reservoir output value for each combination. Build dataset ;

[0053] S2.2. Using the initial water level, the final water level, and the inflow rate as inputs, and the reservoir output as output, a response surface model is constructed and trained using the LSTM algorithm. .

[0054] S2.3, Repeat steps S1 and S2 to build the next reservoir. The response surface model, until .

[0055] refer to Figure 2 Furthermore, the aforementioned step S3 includes the following sub-steps:

[0056] S3.1 Initialization The number of discrete states is , For the number of stages, Based on inbound flow Initial water level , and the end water level , Perform state combination, where , Given the state vectors formed by the initial and final water levels of each reservoir, the entire set of states is calculated using a response surface model. Corresponding reservoir output ;

[0057] S3.2 Calculate based on the recursive equation of the response surface model Optimal benefit at each stage:

[0058] ,

[0059] In the formula, LSTM represents the response surface model. Let t be the initial state set for stage t. for Phase end state set, for The optimal benefit function for a given stage. for The optimal benefit function for a given stage.

[0060] S3.3、 Then determine whether t is equal to T. If t is equal to T, the optimal scheduling scheme for the reservoir group within the entire scheduling range is obtained. If t is not equal to T, return to step S3.1.

[0061] This invention applies to the Longyangxia Hydropower Station, located on the main stream of the Yellow River in Qinghai Province, China. The Longyangxia Hydropower Station consists of 15 turbine-generator units, each with an installed capacity of 320MW, for a total installed capacity of 12.8 million kilowatts. The maximum allowable flow rate through each unit is approximately 1200 m³ / h. 3 Located in the Longyangxia Gorge downstream of the Sanmenxia Gorge on the Yellow River, the [unclear - possibly a specific location or unit] is the first cascade in the planned section of the upper Yellow River. Table 1 compares the computation time of the present invention and the DP method under different inflow conditions. The computation time comparison is shown in Table 1, where the rate ratio is the ratio of the DP computation time to the computation time of the present invention. As the number of discrete points increases, the computation time of both DP and the present invention gradually increases, but the DP algorithm's time increases more significantly. In contrast, the time increase of the present invention's method is slower, and the rate ratio also increases with the number of discrete points. Therefore, this demonstrates the high efficiency of the present invention's method in solving the reservoir optimization scheduling problem.

[0062] Table 1 Unit: s

[0063]

[0064] Secondly, taking the Longyangxia-Laxiwa cascade hydropower station as an example, the effectiveness of this method in the optimal scheduling of cascade reservoirs is verified. The Laxiwa Hydropower Station is located in Gonghe County, Qinghai Province, China, approximately 32.8 km from the Longyangxia Hydropower Station. It is the third large-scale hydropower station on the upper reaches of the Yellow River in Qinghai Province and is one of the key water conservancy projects of the People's Republic of China. The Laxiwa Hydropower Station has an installed capacity of 4,200,000 kW and an average annual power generation of 10.223 billion kWh, making it one of the largest hydropower stations in China. Table 2 compares the power generation results of this invention and the DP method in the application of cascade hydropower stations, and Table 3 compares the computation time of this invention and the DP method in the application of cascade hydropower stations. As shown in Table 3, the power generation of the DP and this invention algorithms in the optimal scheduling of cascade reservoirs is close; as shown in Table 3, when the number of discrete points is greater than 10, the computation time of this invention is less than that of the DP algorithm, and the rate ratio gradually increases with the increase of the number of discrete points. Therefore, this example proves the applicability of this invention in the optimal scheduling of cascade reservoirs.

[0065] Table 2 Unit: 100 million kWh

[0066]

[0067] Table 3 Unit: s

[0068]

[0069] Another aspect of this invention proposes a reservoir optimization scheduling system coupled with LSTM and dynamic programming, comprising: a response surface modeling data acquisition module, a response surface model construction module, and a response surface dynamic programming algorithm module;

[0070] The response surface modeling data acquisition module is used to obtain data based on the reservoir. The upper and lower limits of water level and the maximum and minimum values ​​of historical inflow are used to discretize the state of water level and flow, obtaining a dataset containing different combinations of initial water level, final water level, and inflow. Then, using the dataset, the water balance equation and reservoir characteristic curves are used to calculate the reservoir under different combinations. Exertion, of which, ; This refers to the number of cascade reservoir groups.

[0071] The response surface model construction module is used to construct a response surface model by using a dataset containing different combinations of initial water level, final water level, and inflow, as well as reservoir output under different combinations, and to simulate the mapping between initial water level, final water level, inflow and output using the LSTM algorithm.

[0072] The response surface dynamic programming algorithm module is used to discretize the reservoir water level during the scheduling period, calculate the estimated reservoir output under different discrete state combinations at each stage using the response surface model, and call the dynamic programming algorithm to calculate the optimal benefit for each stage.

[0073] The present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the steps of the method described in the present invention.

[0074] The present invention also proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the present invention.

[0075] While the present invention has been described above with reference to preferred embodiments, it is not intended to limit the invention. Those skilled in the art can make various modifications and refinements without departing from the spirit and scope of the invention. Therefore, the scope of protection of the present invention shall be determined by the claims.

Claims

1. A method for optimal scheduling of a group of reservoirs, characterized in that, Includes the following steps: S1, according to the reservoir The upper and lower limits of water level and the maximum and minimum values ​​of historical inflow are used to discretize the state of water level and flow, obtaining a dataset containing different combinations of initial water level, final water level, and inflow. Then, using the dataset, the water balance equation and reservoir characteristic curves are used to calculate the reservoir under different combinations. Exertion, of which, ; This refers to the number of cascade reservoir groups. S2. Using the dataset obtained in step S1, and the reservoirs under different combinations The power output is simulated using the LSTM algorithm to map the initial and final water levels, inflow rate, and power output of the reservoir, thus constructing a model of the reservoir. The response surface model; let Return to step S1 to construct the response surface model for the next reservoir, until... Step S2 includes the following sub-steps: S2.1, Based on the combination of discrete points of water level and flow rate and the reservoir output value for each combination. Build dataset ; For reservoir The Individual discrete points of inbound flow; S2.

2. Using the initial water level, the final water level, and the inflow rate as inputs, and the reservoir output as output, a response surface model is constructed and trained using the LSTM algorithm. ; S2.3, Repeat steps S1 to S2 to build the next reservoir. The response surface model, until ; S3, During the scheduling period The water levels of each reservoir are discretized, and the estimated water levels of the reservoirs under different combinations of discrete states at each stage are calculated using a response surface model. Output effort, call the dynamic programming algorithm to calculate the optimal benefit for this stage; step S3 includes the following sub-steps: S3.1 Initialization The number of discrete states is , For the number of stages, Based on inbound flow Initial water level , and the end water level , Perform state combination, where , Given the state vectors formed by the initial and final water levels of each reservoir, the entire set of states is calculated using a response surface model. Corresponding reservoir output ; S3.2 Calculate based on the recursive equation of the response surface model Optimal benefit at each stage: ， In the formula, LSTM represents the response surface model. Let t be the initial state set of the reservoir group at stage t. For the reservoir group Phase end state set, for The optimal benefit function for a given stage. for The optimal benefit function for a given stage; S3.3、 Then determine whether t is equal to T. If t is equal to T, the optimal scheduling scheme for the reservoir group within the entire scheduling range is obtained. If t is not equal to T, return to step S3.

1.

2. The reservoir group optimization scheduling method according to claim 1, characterized in that, Step S1 includes the following sub-steps: S1.1, During the scheduling and operation period, the reservoir... The initial and final water levels are discrete within their upper and lower limits. There are points, represented as follows: and ,in, and They represent the first The initial water level discrete point and the first One final water level discrete point; S1.

2. Based on historical inbound flow statistics, the inbound flow is discretized to the maximum flow rate. and minimum flow Between 1 point, represented as ; S1.3 Obtain the discrete point combination of water level and flow rate The reservoir flow rate is obtained by substituting it into the water balance equation in formula (1), and then the downstream water level and head are obtained by using the tailwater level-flow curve. Finally, the reservoir output value for each combination is calculated using the output formula in formula (2). ; ， ， In the formula, For outbound flow, Because of the long duration, This is the output coefficient. The difference in water head between the upstream and downstream of the reservoir.

3. A reservoir group optimization scheduling system applied to the reservoir group optimization scheduling method according to any one of claims 1-2, characterized in that, include: Response surface modeling data acquisition module, response surface model construction module, response surface dynamic programming algorithm module; The response surface modeling data acquisition module, based on the reservoir The upper and lower limits of water level and the maximum and minimum values ​​of historical inflow are used to discretize the state of water level and flow, obtaining a dataset containing different combinations of initial water level, final water level, and inflow. Then, using the dataset, the water balance equation and reservoir characteristic curves are used to calculate the reservoir under different combinations. Exertion; among them, ; This refers to the number of cascade reservoir groups. The response surface model construction module is used to utilize the dataset obtained by the response surface modeling data acquisition module, as well as reservoirs under different combinations. The power output is simulated using the LSTM algorithm to map the initial and final water levels, inflow rate, and power output of the reservoir, thus constructing a model of the reservoir. The response surface model; let The response surface modeling data acquisition module is then returned to execute, and the response surface model is constructed for the next reservoir until... ; The response surface dynamic programming algorithm module, during the scheduling period The water levels of each reservoir are discretized, and the estimated water levels of the reservoirs under different combinations of discrete states at each stage are calculated using a response surface model. Output effort, and use dynamic programming algorithm to calculate the optimal efficiency for the stage.

4. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 2.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 2.

Images