A method for back-calculating operating parameters and historical sequence data of adjacent hydropower stations based on particle swarm optimization
The particle swarm algorithm with a restart function and inverse optimization model effectively infers upstream hydropower station data, addressing prediction challenges and enhancing downstream power generation planning accuracy and efficiency.
Patent Information
- Application Number
- CN202411226697.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-03
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-09-03
AI Technical Summary
Downstream hydropower stations face challenges in predicting the power generation plans of upstream stations due to information opacity, leading to increased engineering risks and power generation uncertainties, which existing methods fail to address effectively.
A method using a particle swarm algorithm with a restart function to infer operating parameters and historical sequence data of upstream hydropower stations, incorporating an inverse optimization double-layer model and engineering experience to reconstruct the model, reducing resource consumption and avoiding local optima.
The method provides accurate and timely estimation of upstream hydropower station data, supporting downstream stations in formulating effective power generation plans and reducing water abandonment risks.
Smart Images

Figure CN118763743B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of hydropower station dispatching and operation, and in particular relates to a method for inferring operating parameters and historical sequence data of adjacent hydropower stations based on a particle swarm algorithm. Background Art
[0002] Hydropower is the largest clean energy in the world and in my country. With the growth of global energy demand and the need to reduce carbon emissions, countries around the world will use hydropower as one of their energy development strategies. Hydropower is often built in a cascade form. The hydropower and electricity between upstream and downstream hydropower stations are closely coupled. Cascade hydropower requires upstream and downstream coordinated power generation to achieve the highest efficiency. However, a river basin cascade is usually developed by multiple property units, and upstream and downstream power stations often belong to different property owners. The power generation scheduling methods between different property owners vary greatly, making it extremely difficult to closely cooperate between cascade power stations. Especially after the reform of the power system, the power market model encourages competition between power stations, making the upstream and downstream mismatch phenomenon increasingly serious, which is mainly manifested in: the upstream power station arranges power generation based on its own characteristics, power generation targets and runoff forecasts, and lacks consideration of the power generation plan of the downstream power station; while the downstream power station is highly dependent on the outflow flow of the upstream power station, so the power generation process is greatly restricted by the upstream; the opacity of the power generation plan of the upstream power station to the downstream power station increases the probability of extreme situations (severe lack of power generation capacity or large amounts of water abandonment) of the downstream power station, bringing serious engineering risks. How to help downstream power stations eliminate or alleviate the information uncertainty of the power generation plans of upstream power stations through effective means and improve the feasibility of power generation arrangements of downstream power stations is a practical engineering problem that power generation companies urgently need to solve in the current electricity market reform, and it is also the core problem that the present invention aims to solve.
[0003] In the existing literature and technical data, the methods to solve the mismatch problem of cascade power stations are mainly focused on the design of power market mechanisms and the response of power generation enterprises themselves. In terms of mechanism design, Brazil introduced the "power redistribution" and "guaranteed capacity" mechanisms, and the National Power Dispatching Center uniformly carried out real-time economic dispatching, weakening the hydropower connection between upstream and downstream; Northern Europe managed the risks brought by the mismatch problem of upstream and downstream by introducing a multi-time scale trading market. However, the above countries have different ideas for the development of power markets from my country, and their experience is difficult to apply directly. In terms of the response of power generation enterprises themselves, when optimizing the dispatch of power generation systems, the most common idea is to take the behavior of upstream power stations into account and use probability or fuzzy mathematics methods to approximate the simulation of upstream adjacent power stations. However, this method relies heavily on data fitting from a statistical perspective, ignores the hydropower coupling relationship between cascade power stations, and does not fully explore the deep reasons behind the model. Therefore, downstream power stations urgently need a practical method based on physical laws to reverse the information of upstream adjacent hydropower stations, so as to better predict the power generation plan of upstream power stations and reduce their own power generation uncertainty.
[0004] In view of this, the present invention discloses a practical method for inferring the operating parameters and historical sequence data of adjacent upstream hydropower stations, and uses three adjacent hydropower stations in a hydropower base in my country but belonging to two property owners as the engineering background for application calculation. The results show that the present invention can efficiently and accurately infer the operating parameters and historical sequence data of upstream power stations based on historical observation data, public data and disclosed operating steps, meeting the accuracy and practicality requirements of power station dispatch optimization. The present invention can provide technical reference for my country's power market reform and provide reference for the technical solution of my country's "dual carbon" goal. Summary of the invention
[0005] The technical problem to be solved by the present invention is that the downstream hydropower station uses its own historical observation data and public data to reversely estimate the operating parameters and historical sequence data of the upstream adjacent hydropower station. From the perspective of the downstream power station, the present invention first establishes an inverse optimization double-layer model for the upstream adjacent hydropower station taking into account the influence of natural runoff, and integrates parameter analysis based on engineering experience to reconstruct the model. Finally, a particle swarm algorithm with a restart function is used to achieve rapid estimation of the historical sequence data and operating parameters of the upstream adjacent hydropower station, providing technical support for the hydropower station to formulate an effective power generation production plan.
[0006] Technical solution of the present invention:
[0007] The present invention discloses a method for inferring the operating parameters and historical sequence data of adjacent hydropower stations based on a particle swarm algorithm, which mainly includes three main steps: establishing an inference model based on its own observation data and public data, determining parameters based on engineering experience and reconstructing the inference model, and solving the reconstructed inference model using a particle swarm algorithm with a restart function. Finally, the rapid estimation of the operating parameters and historical sequence data of the upstream adjacent hydropower station can be achieved, providing technical support for the hydropower station to formulate an effective power generation production plan. The application scope of the present invention is a cascade hydropower generation system composed of three hydropower stations, which are respectively called the first power station, the middle power station, and the last power station from upstream to downstream according to their spatial distribution. The cascade hydropower generation system can be a local component of a larger-scale cascade hydropower generation system. In the cascade hydropower generation system, the first power station and the last power station belong to the same property right subject, and the middle power station belongs to another property right subject. In the present invention, the middle power station is the adjacent upstream power station of the last power station, so it is also called the target power station to be inferred by the last power station, referred to as the "target power station". The specific steps of the present invention are as follows:
[0008] Step 1: Establish an inference model based on own observation data and public data;
[0009] The inference model refers to an inverse optimization double-layer model for inferring the operating parameters of the upstream adjacent hydropower station and their historical sequence data.
[0010] In the inverse optimization two-layer model, the upper model takes the minimum deviation of the power generation sequence as the objective function, and is responsible for optimizing and iterating the operating parameters and historical sequence data of the target power station; the lower model takes the actual scheduling process of the target power station as the background, takes the maximum power generation as the objective function, simulates the physical power generation behavior of the target power station based on the parameters passed in by the upper model, and returns the scheduling results to the upper model as the basis for the next step of optimization and iteration of the upper model.
[0011] As shown in formula (1) to formula (14), formula (1) to formula (14) are a whole and together constitute the inference model. Among them, the objective function of the upper model is formula (1), and the constraint is formula (2) and the entire lower model; the objective function of the lower model is formula (3), and the constraint condition is formula (4) to formula (14).
[0012] (1)
[0013] st (2)
[0014] (3)
[0015] st (4)
[0016] (5)
[0017] (6)
[0018] (7)
[0019] (8)
[0020] (9)
[0021] (10)
[0022] (11)
[0023] (12)
[0024] (13)
[0025] (14)
[0026] The subscripts "up" and "dn" of the parameters refer to the first power station and the last power station respectively; if there is no subscript "up" or "dn", the parameters are the relevant parameters of the target power station; is the set of all time periods, for a certain period of time; is the set of real numbers; Target power station The actual power generation during the period; Target power station Simulated power generation for the time period; Target power station The power generation flow during the period; is the inverse of the water consumption rate; The duration corresponding to a time period; Target power station Storage capacity during the time period; and The target power station is Inflow and outflow during the period; Target power station The amount of water abandoned during the period; is the water level-storage capacity relationship curve of the target power station, is the tailwater level-discharge relationship curve of the target power station, It is the water consumption rate inverse-water head relationship curve of the target power station. , , All three curves are fitted with a linear function. are the slope and intercept corresponding to each curve respectively; Target power station Water level above the dam, tailwater level, and net head during the period; is the head loss of the target power station; are the upper and lower limits of the target power station storage capacity respectively; The installed capacity of the target power station; is the discharge flow of the first power station; for The natural runoff between the first power station and the target power station in the time period; for The inflow of the power station at the end of the time period; for The natural runoff between the target power station and the last power station in the time period; for The sum of natural runoff between the first power station and the last power station in a time period.
[0027] For the lower model, These 7 parameters are assigned by the upper model.
[0028] Compared with the existing public literature, the inference model disclosed in the present invention takes into account the natural runoff factor, and the addition of this factor will significantly improve the applicability of the inference model. The relationship between natural runoff is given by equations (12) to (14), and indirectly affects the values of power generation-related physical quantities such as water level, reservoir capacity, power generation flow, and abandoned water flow.
[0029] Step 2: Parameter determination and inference model reconstruction based on engineering experience;
[0030] The inference model built in step 1 has only conceptual meaning due to the presence of a large number of functions and variables and cannot be directly applied. Therefore, step 2 will use engineering experience to determine the value of the parameters and then transform the inference model in step 1 into a solvable form, which includes the following three steps:
[0031] Step 2.1: Determine the distribution of natural runoff between upstream and downstream of the cascade;
[0032] Adjacent cascade hydropower stations are in the same hydrological environment. According to engineering experience, the distribution of natural runoff among the three hydropower stations is proportional to the basin area controlled by each station. Therefore, two equations are added to the inference model established in step 1, expressed as equations (15) and (16).
[0033] (15)
[0034] (16)
[0035] Where: They are the basin areas controlled by the first power station, the target power station and the last power station respectively.
[0036] Step 2.2: Determination of water discard parameters;
[0037] According to engineering experience, power plants will not actively abandon water when arranging power generation plans, and will only be forced to abandon water in extreme cases. Considering that the lower model simulates the active power generation behavior of the target power plant, the formula (6) .
[0038] Step 2.3: Reconstruct the inference model;
[0039] Combine equations (5) to (10), (12) to (16), and substitute them into equation (4) to obtain equation (17).
[0040] (17)
[0041] Further, let:
[0042] (18)
[0043] Then equation (17) is equivalently transformed into equation (19).
[0044] (19)
[0045] At this point, all parameters to be inferred are unified into formula (19). Further, the inference model established in step 1 is reconstructed into formula (20) to formula (25). Formula (20) to formula (25) are a whole, which together represent the reconstructed inference model. Formula (25) is the same as formula (19). The reconstructed inference model is still an inverse optimization two-layer model, in which the objective function of the upper model is formula (20), and the constraints are formula (21) and the entire lower model; the objective function of the lower model is formula (22), and the constraints are formula (23) to formula (25).
[0046] (20)
[0047] st (twenty one)
[0048] (twenty two)
[0049] st (twenty three)
[0050] (twenty four)
[0051] (25)
[0052] Step 3: Use the particle swarm optimization algorithm with restart function to solve the reconstructed inference model;
[0053] The particle swarm algorithm is an easy-to-implement and robust evolutionary computation method, which is widely used in solving non-convex optimization problems. However, if the standard particle swarm algorithm published in the literature is directly used to solve the reconstructed inference model in step 2, the following two problems will be encountered, thereby reducing the usability of the model:
[0054] Problem a: The standard particle swarm algorithm requires a large population, which takes a lot of time and resources to calculate.
[0055] Problem b: Solving the underlying model is a complex problem that is high-dimensional, non-convex, and nonlinear, and takes a long time to solve;
[0056] Another idea is to use the particle swarm algorithm that has been published in the literature. Compared with the standard particle swarm algorithm, the advantage of the particle swarm algorithm is that it only requires a small number of particles, which greatly reduces resource consumption, but the cost is that the diversity of the population is easily destroyed and the solution process is prone to "premature". According to practical experience, directly using the particle swarm algorithm to solve the reconstructed inference model in step 2, the search process will converge quickly, and the searchable range is small, that is, the "premature" phenomenon still exists. In order to solve the above problems, the present invention discloses a particle swarm algorithm with a restart function, which transforms the particle swarm algorithm to maintain the diversity of the population during the solution process and avoid the algorithm from falling into a local optimal solution. The specific solution steps are as follows:
[0057] Step 3.1: Initialize the population. Set the number of particles in the particle swarm algorithm to M and the maximum number of cycles to D. Generate M particles, each of which has two attributes: position and velocity. The initial position and velocity are randomly generated within the domain.
[0058] Step 3.2: Calculate the fitness of each particle. Calculate the fitness of each particle according to the objective function (20) of the upper model.
[0059] Step 3.3: Update the global optimum and individual optimum. Compare the fitness values of all particles with the current global optimum fitness value and the historical individual optimum fitness value, and update the global optimum and individual optimum values respectively.
[0060] Step 3.4: Update the speed and position of each particle. The speed and position of each particle are updated through the memory item, self-cognition item and group cognition item.
[0061] Step 3.5: Evaluate the degree of population convergence. After each iteration, the particle swarm algorithm with restart function evaluates the degree of population convergence. If the degree of convergence does not exceed the threshold, no operation is performed on the population; if the degree of convergence exceeds the threshold, it means that the algorithm has fallen into a local optimum. At this time, the population is regenerated. The newly generated particle population replaces the original particle population, but all the memories of the original particle population are retained.
[0062] Step 3.6: Repeat steps 3.2 to 3.5 until the number of iterations reaches D or the accuracy requirement is met.
[0063] Beneficial effects of the present invention: The present invention combines engineering practice and discloses a practical method for inferring the operating parameters and historical sequence data of upstream adjacent hydropower stations using its own observation data and public data. The method first establishes an inverse optimization two-layer model for the upstream adjacent hydropower station, and then reconstructs the model based on engineering experience, so that the model solution parameters are greatly reduced. During the solution process, a particle swarm algorithm with a restart function is adopted. On the one hand, a small number of particles are used to greatly reduce resource consumption. On the other hand, multiple restarts are performed to avoid the "premature" phenomenon to improve the solution efficiency and solution quality. The present invention can quickly provide reasonable estimation results of the operating parameters and historical sequence data of the upstream adjacent hydropower station through a small amount of observations and public historical data, which meets the timeliness and practicality requirements, and provides a new technical approach for hydropower stations to better predict future states, increase power generation, and reduce the risk of water abandonment. BRIEF DESCRIPTION OF THE DRAWINGS
[0064] Figure 1 It is a topological structure diagram of a hydroelectric power generation system to which the present invention is applicable;
[0065] Figure 2 It is a flow chart of particle swarm algorithm with restart function;
[0066] Figure 3 is the deviation correlation diagram between the objective function value of the inference model and the inference result of the operating parameters of the B hydropower station;
[0067] Figure 4 It is a graph of the parameter evolution of the particle swarm algorithm with restart function;
[0068] Figure 5 It is the overall flow chart of the present invention. DETAILED DESCRIPTION
[0069] The present invention is further described below in conjunction with the accompanying drawings and implementation examples.
[0070] The cascade reservoir composed of three adjacent power stations from two different ownership units in the middle and lower reaches of a hydropower base in my country was selected as the research object to verify the present invention. The three hydropower stations are A hydropower station, B hydropower station, and C hydropower station from upstream to downstream, which correspond to the first power station, middle power station, and last power station referred to in the present invention, respectively. The detailed topological structure is shown in Figure 1 In this cascade hydropower generation system, the first power station and the last power station belong to the property right subject A, and the middle power station belongs to the property right subject B. The middle power station is the adjacent upstream power station of the last power station, so it is also called the target power station to be inferred by the last power station, referred to as the "target power station".
[0071] The cascade of three power stations is a typical large-scale hydropower cascade with multiple ownership entities. Due to the close connection between electricity and water power among cascade hydropower, the power generation plan and reservoir discharge process of B hydropower station have a significant impact on the power generation process of C hydropower station, which has a fundamental impact on the feasibility of the power generation plan formulated by C hydropower station.
[0072] In the actual application of the inference model in the following text, the parameters of the first power station in the upstream position, Hydropower Station A, are represented by letters with the subscript "up", and the parameters of the last power station in the downstream position, Hydropower Station C, are represented by letters with the subscript "dn". Letters without the subscript "up" or "dn" represent the parameters of the target power station to be inferred, Hydropower Station B. Figure 5 As shown, the specific steps of applying the present invention are as follows:
[0073] Step 1: Establish an inference model based on own observation data and public data;
[0074] Step 1.1 build the upper model;
[0075] The upper model passes the unknown parameters to the lower model, and optimizes and iterates based on the deviation of the results obtained by the lower model. Therefore, the upper model aims to minimize the deviation between the simulated power generation sequence and the actual power generation sequence. According to the actual physical meaning, determine The objective function of the upper model is equation (26), and the constraints are equation (27) and the entire lower model mentioned in step 1.2. Equations (26) to (39) are a whole, which together constitute the inference model of the B hydropower station.
[0076] (26)
[0077] st (27)
[0078] It is the private information of Hydropower Station B, which is unknown information to the property right subject A. In the inference model, it is given by the upper model and passed to the lower layer.
[0079] Step 1.2 build the lower model;
[0080] The lower model simulates the power generation dispatching operation process of the target power station. Therefore, taking the maximum power generation of the target power station as the optimization criterion, the objective function of the lower model is formula (28).
[0081] (28)
[0082] There are 9 constraints as follows.
[0083] 1) Power generation function constraints:
[0084] (29)
[0085] 2) Water balance constraints:
[0086] (30)
[0087] 3) Downflow flow constraints:
[0088] (31)
[0089] 4) Water level-reservoir capacity curve constraints:
[0090] (32)
[0091] In the formula, The two parameters are private information of Hydropower Station B, and from the perspective of the property right subject A, they are unknown information.
[0092] 5) Tailwater level-discharge flow constraint:
[0093] (33)
[0094] In the formula, The two parameters are private information of Hydropower Station B, and from the perspective of the property right subject A, they are unknown information.
[0095] 6) Inverse of water consumption rate - head curve constraint:
[0096] (34)
[0097] In the formula, The two parameters are private information of Hydropower Station B, and from the perspective of the property right subject A, they are unknown information.
[0098] 7) Constraints on the head calculation formula:
[0099] (35)
[0100] In the formula, It is the private information of Hydropower Station B. From the perspective of the property right subject A, it is unknown information.
[0101] 8) Upper and lower limit constraints:
[0102] (36)
[0103] 9) Natural runoff constraints:
[0104] (37)
[0105] (38)
[0106] (39)
[0107] Step 2: Delimitation of inference model parameters and model reconstruction based on engineering experience;
[0108] Step 2.1: Determine the distribution of natural runoff between upstream and downstream of the cascade;
[0109] Adjacent cascade hydropower stations are in the same hydrological environment, so the distribution of natural runoff among the three hydropower stations is proportional to the basin area controlled by each station, which can be expressed as formula (40) and formula (41):
[0110] (40)
[0111] (41)
[0112] Step 2.2: Determination of water discard parameters;
[0113] According to engineering experience, power stations do not actively abandon water when arranging power generation plans, and are forced to abandon water only in extreme cases. Considering that the lower model simulates the active power generation behavior of hydropower station B, the formula (6) .
[0114] Step 2.3: Reconstruct the inference model;
[0115] By combining equations (30) to (35) and equations (37) to (41) and substituting them into equation (29), we obtain equation (42).
[0116] (42)
[0117] Further, let:
[0118] (43)
[0119] Formula (42) can be equivalently transformed into formula (44).
[0120] (44)
[0121] At this time, all parameters to be inferred are unified into formula (44). Further, the inference model established in step 1 is reconstructed into formula (45) to formula (50). Formula (45) to formula (50) are a whole and together constitute the reconstructed inference model. Formula (49) is the same as formula (44), where the objective function of the upper model is formula (45), and the constraints are formula (46) and the lower model; the objective function of the lower model is formula (47), and the constraints are formula (48) to formula (50).
[0122] (45)
[0123] st (46)
[0124] (47)
[0125] st (48)
[0126] (49)
[0127] (50)
[0128] Step 3: Use the particle swarm algorithm with restart function to solve the reconstructed inference model, such as Figure 2 As shown;
[0129] Step 3.1: Initialize the population. Set the number of particles in the particle swarm algorithm to M and the maximum number of iterations to D. Generate M particles, each of which has two attributes: position and velocity. The initial position and velocity are randomly generated within the definition domain.
[0130] Step 3.2: Calculate the fitness of each particle. Calculate the fitness of each particle according to the objective function of the upper model.
[0131] Step 3.3: Update the global optimum and individual optimum. Compare the fitness values of all particles with the current global optimum fitness value and the historical individual optimum fitness value, and update the global optimum and individual optimum values respectively.
[0132] Step 3.4: Update the speed and position of each particle. The speed and position of each particle are updated through the memory item, self-cognition item and group cognition item.
[0133] Step 3.5: Evaluate the degree of population convergence. After each iteration, the particle swarm algorithm with restart function evaluates the degree of population convergence. If the degree of convergence does not exceed the threshold, no operation is performed on the population. If the degree of convergence exceeds the threshold, it means that the algorithm has fallen into a local optimum. At this time, the population is regenerated. The newly generated particle population is used to replace the original particle population, but all the memories of the original particle population are retained.
[0134] Step 3.6: Repeat steps 3.2 to 3.5 until the number of iterations reaches D or the accuracy is met.
[0135] The installed capacities of the three power stations are shown in Table 1.
[0136] Table 1 Installed capacity of the hydropower station used in the example
[0137]
[0138] The real data of Hydropower Station A and Hydropower Station C from January 2011 to December 2017 for a total of 84 months are used, one of which is one month. The length of is 84. Set the number of particles M of the particle swarm algorithm with restart function to 8 and the maximum number of iterations D to 1000.
[0139] Table 2 shows the errors of the operating parameters and historical sequence data of Hydropower Station B, and their values meet the engineering requirements. Tables 3 and 4 respectively give the five months with the smallest and largest deviations in the power generation inference. The five months with smaller inference deviations are all concentrated in the flood season. In fact, when the water inflow is relatively abundant during the flood season, in order to meet the requirements of flood control, the power station generally controls the water level below the flood limit water level. In order to maximize the value of water resources and reduce water abandonment, the power station generally generates electricity at or close to the installed capacity. At this time, the power generation is relatively fixed, and the inference model has a better inference effect. The five months with larger inference deviations are all concentrated in the dry season. The dispatching and operation of the power station during the dry season should consider many factors. On the one hand, it is necessary to consider participating in the current electricity market to obtain benefits. On the other hand, it is also necessary to reasonably reserve the reservoir capacity based on the forecast of water inflow to meet future water use and power generation requirements. Since the inference model proposed in this paper is an ex post model, all water inflow processes are considered to be known and the power generation process of the target power station is controlled according to the optimal dispatching method. However, in the actual dispatching process, the uncertainty of water inflow will make the power generation plan of the power station more conservative. It is necessary to find the dispatching plan with the largest mathematical expectation value of power generation in the process of water inflow in multiple scenarios. Therefore, it is almost impossible to carry out the actual plan according to the optimal solution of actual water inflow, which also explains the reason for the large deviation of power inference during the dry season.
[0140] Table 2 Error between solution results and actual results
[0141]
[0142] Table 3 The five months with the smallest power generation estimation error
[0143]
[0144] Table 4 The five months with the largest power generation estimation errors
[0145]
[0146] Figure 3 This is the deviation correlation diagram between the objective function value (i.e., the power generation series deviation) and the parameter inference result. It can be seen from the figure that the two have a certain positive correlation, especially the parameter The correlation is the strongest. The correlation results show that when the objective function value is smaller, the inference of the parameters is more accurate, which shows that it is correct to set the objective function as the power generation series deviation in this method. This result also reveals that the example with the smallest objective function value can be used to obtain a more accurate inference of the parameters.
[0147] Figure 4 This is the evolution process of the parameters of the particle swarm algorithm with a restart function. In the early stage, the inference of the parameters often has a large improvement. In the middle and late stages, the parameters have fallen near the approximate optimal solution and began to improve slightly. Each improvement of the objective function is accompanied by the improvement of at least one parameter, and may be accompanied by the increase of one or more parameter deviations. The above phenomenon shows that the particle swarm algorithm with a restart function disclosed in the present invention is effective in anchoring parameters, but because the problems faced are non-convex and nonlinear, there are often a large number of local optimal solutions near the true value, so the algorithm still needs certain iterative steps to further improve the parameters.
[0148] On Intel Core TM The calculation is performed on an i9-12900H CPU, 16GB memory, and Windows 11 operating system. When M is 8, the calculation time for each generation can be controlled within 6 seconds. Since the parameters of the hydropower station are fixed, they only need to be calculated once and stored, and can be used in the future without repeated calculations. Therefore, this running time can meet the timeliness requirements of the property rights subject. This method is scalable and can gradually optimize the inference model according to the passage of time and the increasing number of observation data to obtain more accurate inference results.
[0149] The final results and solution process show that the inference model disclosed in the present invention can accurately model the target power plant, and the solution method provided is effective in finding and approaching the optimal solution, avoiding infeasible regions, and accelerating the solution speed. The present invention successfully estimates the operating parameters and historical sequence data of the target power plant.
Claims
1. A method for inferring the operating parameters and historical sequence data of adjacent hydropower stations based on particle swarm optimization, characterized in that: The application scope of the method for inferring the operating parameters and historical sequence data of adjacent hydropower stations based on the particle swarm algorithm is a cascade hydropower generation system consisting of three hydropower stations. According to their spatial distribution, they are respectively called the first power station, the middle power station, and the last power station from upstream to downstream, wherein the first power station and the last power station belong to the same property subject, and the middle power station belongs to another property subject; the middle power station is the adjacent upstream power station of the last power station, so it is also called the target power station to be inferred by the last power station, referred to as the "target power station"; the specific steps of the method for inferring the operating parameters and historical sequence data of adjacent hydropower stations based on the particle swarm algorithm are as follows: Step 1: Establish an inference model based on own observation data and public data; The inference model refers to an inverse optimization double-layer model for inferring the operating parameters of the upstream adjacent hydropower station and its historical sequence data; In the inverse optimization two-layer model, the upper model takes the minimum deviation of the power generation sequence as the objective function, and is responsible for optimizing and iterating the operating parameters and historical sequence data of the target power station; the lower model takes the actual scheduling process of the target power station as the background, takes the maximum power generation as the objective function, simulates the physical power generation behavior of the target power station based on the parameters passed in by the upper model, and returns the scheduling results to the upper model as the basis for the next step of optimization and iteration of the upper model; As shown in formula (1) to formula (14), formula (1) to formula (14) are a whole, and together constitute the inference model; wherein, the objective function of the upper model is formula (1), and the constraint condition is formula (2) and the entire lower model; the objective function of the lower model is formula (3), and the constraint condition is formula (4) to formula (14); (1) s.t. (2) (3) s.t. (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) The subscripts "up" and "dn" of the parameters refer to the first power station and the last power station respectively; if there is no subscript "up" or "dn", the parameters are the relevant parameters of the target power station; is the set of all time periods, for a certain period of time; is the set of real numbers; Target power station The actual power generation during the period; Target power station Simulated power generation for the time period; Target power station The power generation flow during the period; is the inverse of the water consumption rate; The duration corresponding to a time period; Target power station Storage capacity during the time period; and The target power station is Inflow and outflow during the period; Target power station The amount of water abandoned during the period; is the water level-storage capacity relationship curve of the target power station, is the tailwater level-discharge relationship curve of the target power station, It is the water consumption rate inverse-water head relationship curve of the target power station. , , All three curves are fitted with a linear function. are the slope and intercept corresponding to each curve respectively; Target power station Water level above the dam, tailwater level, and net head during the period; is the head loss of the target power station; are the upper and lower limits of the target power station storage capacity respectively; The installed capacity of the target power station; is the discharge flow of the first power station; for The natural runoff between the first power station and the target power station in the time period; for The inflow of the power station at the end of the time period; for The natural runoff between the target power station and the last power station in the time period; for The sum of natural runoff between the first power station and the last power station in the time period; For the lower model, These 7 parameters are assigned by the upper model; Step 2: Parameter determination and inference model reconstruction based on engineering experience; The inference model built in step 1 has only conceptual meaning due to the presence of a large number of functions and variables and cannot be directly applied. Therefore, step 2 will use engineering experience to determine the value of the parameters and then transform the inference model in step 1 into a solvable form, which includes the following three steps: Step 2.1: Determine the distribution of natural runoff between upstream and downstream of the cascade; Adjacent cascade hydropower stations are in the same hydrological environment. According to engineering experience, the distribution of natural runoff among the three hydropower stations is proportional to the basin area controlled by each station. Therefore, two equations are added to the inference model established in step 1, expressed as equations (15) and (16); (15) (16) Where: The basin areas controlled by the first power station, the target power station and the last power station respectively; Step 2.2: Determination of water discard parameters; According to engineering experience, power plants will not actively abandon water when arranging power generation plans, and will only be forced to abandon water in extreme cases. Considering that the lower model simulates the active power generation behavior of the target power plant, the formula (6) ; Step 2.3: Reconstruct the inference model; Combine equations (5) to (10), (12) to (16), and substitute them into equation (4) to obtain equation (17); (17) Further, let: (18) Then transform equation (17) into equation (19); (19) At this time, all the parameters to be inferred are uniformly collected in formula (19); further, the inference model established in step 1 is reconstructed into formula (20) to formula (25), which are a whole and jointly represent the reconstructed inference model. Formula (25) is the same as formula (19); the reconstructed inference model is still an inverse optimization two-layer model, in which the objective function of the upper model is formula (20), and the constraints are formula (21) and the entire lower model; the objective function of the lower model is formula (22), and the constraints are formula (23) to formula (25); (20) s.t. (21) (22) s.t. (23) (24) (25) Step 3: Use the particle swarm optimization algorithm with restart function to solve the reconstructed inference model; Step 3.1: Initialize the population: Set the number of particles in the particle swarm algorithm to M and the maximum number of cycles to D; generate M particles, each of which contains two attributes: position and velocity. The initial position and velocity are randomly generated within the definition domain; Step 3.2: Calculate the fitness of each particle: Calculate the fitness of each particle according to the upper model objective function (20); Step 3.3: Update the global optimum and individual optimum: Compare the fitness values of all particles with the current global optimum fitness value and the historical individual optimum fitness value, and update the global optimum and individual optimum values respectively; Step 3.4: Update the speed and position of each particle: For each particle, the speed and position are updated through the memory item, the self-cognition item and the group cognition item; Step 3.5: Evaluate the degree of population convergence: After each iteration, the particle swarm algorithm with restart function evaluates the degree of population convergence. If the degree of convergence does not exceed the threshold, no operation is performed on the population; if the degree of convergence exceeds the threshold, it means that the algorithm has fallen into a local optimum. At this time, the population is regenerated; the newly generated particle population is used to replace the original particle population, but all the memories of the original particle population are retained; Step 3.6: Repeat steps 3.2 to 3.5 until the number of iterations reaches D or the accuracy requirement is met.
