A rolling-corrected medium- and long-term implicit stochastic joint optimization scheduling method

Abstract

The present invention discloses a rolling-corrected medium- and long-term implicit random joint optimization scheduling method, comprising the following steps: establishing a deterministic medium- and long-term optimization scheduling model with the maximum system power generation as the objective function, solving the deterministic medium- and long-term optimization scheduling model using a step-by-step optimization algorithm (POA), optimizing parameters, establishing a medium- and long-term wind-solar-cascade hydropower implicit random joint scheduling function model, and using the SVM method for solving the problem, and a rolling simulation correction strategy. The present invention optimizes the scheduling operation of cascade hydropower stations that fully receive wind and solar power output at medium- and long-term scales by establishing a PSO-SVM medium- and long-term implicit random scheduling function model based on parameter optimization and combining it with a rolling simulation correction strategy. Through comparative analysis, the results show that the scheduling operation rules derived from the scheduling function model proposed in the present invention combined with the rolling simulation correction strategy are consistent with the current existing reservoir scheduling laws, and are an effective way to guide the actual operation of cascade hydropower under the "bundled transmission" mode of wind, solar, and water.

Description

Technical Field

[0001] The present invention relates to the technical field of joint dispatching of cascade reservoirs, and in particular to a rolling-corrected medium- and long-term implicit random joint optimization dispatching method. Background Art

[0002] As global climate change continues to intensify and extreme weather events become more frequent, accelerating the green transformation of my country's energy mix has become a key initiative. In recent years, the proportion of installed renewable energy capacity in China has continued to rise. By the end of 2021, China's installed renewable energy capacity reached 1.063 billion kilowatts, with both wind and photovoltaic power generation exceeding 300 billion kilowatts. The newly installed capacity of these two sources reached 101 million kilowatts, accounting for 27% and 31.1% of the country's new power generation capacity, respectively. With the gradual expansion of wind and photovoltaic power station development, new renewable energy generation, represented by these two sources, has been integrated into the grid on a large scale. Multi-energy complementary power generation systems, exemplified by solar-hydro / wind-hydro / wind-solar-hydro, have become a crucial component of modern power systems. Currently, research on the operation of integrated power generation systems such as "wind power-photovoltaic-pumped storage" is extensive and relatively mature. However, the direct grid connection of wind and solar power, coupled with regulation by the main grid, requires high levels of grid stability and dispatchability due to the randomness and uncertainty of wind and solar output, posing significant challenges to grid dispatch. Different from the main grid dispatching to supplement wind and solar power, using basin hydropower to provide auxiliary supplements for wind and photovoltaic power, and establishing an integrated power generation system of "wind, solar, hydropower and comprehensive complementarity followed by 'bundling and external transmission'" is an effective way to absorb wind power and photovoltaic power, and can obtain a smooth output curve, effectively avoiding the impact of wind and solar output fluctuations on the power grid.

[0003] The prior art discloses a multi-time-scale nested wind, solar, and hydropower complementary scheduling method. This method addresses the scheduling and operation of cascade power stations at three time scales: long-term, short-term, and real-time. Different scheduling models are established for different time scales. Finally, nested scheduling and step-by-step information feedback are used to achieve the complementary scheduling goal of coordinated power and electricity compensation, taking into account multiple uncertainties. This technology's long-term optimal scheduling uses a daily scheduling period and a year as the scheduling period. It provides water level constraints for short-term scheduling by changing the water level control method of the basin's reservoirs. Under the water level constraint, short-term optimal scheduling uses hydropower compensation to adjust the random, intermittent, and fluctuating wind and solar output within the day. Real-time scheduling tracks the day-ahead power generation plan and uses hydropower to compensate for the forecast deviation of wind and solar output to meet the daily power generation plan. Simultaneously, real-time scheduling provides compensation for wind and solar output to meet the daily power generation plan, obtaining a new operating water level and feeding it back to the short-term optimal scheduling model. The short-term optimal scheduling model then feeds the operating water level back to the long-term optimal scheduling model, achieving step-by-step nested scheduling and information feedback.

[0004] The prior art also discloses a method and system for calculating wind, solar, and hydropower generation plans. This system establishes a hierarchical structure of power stations and power sources within a river basin, and simultaneously establishes and solves a wind, solar, and hydropower complementary scheduling model to meet the grid-oriented operation and planning requirements for cascade hydropower stations and wind and photovoltaic power generation in the river basin. This technology establishes two hierarchical models, one targeting the maximum sum of all combined power sources and the other maximizing the power generation of each combined power source. Using acquired runoff data and predicted wind and solar output (daily, hourly, and 15-minute intervals), the system completes the wind, solar, and hydropower complementary power generation operation and generation plan through gradual optimization.

[0005] However, the existing wind-solar-hydro complementary scheduling and planning technologies are all aimed at the operation problems of cascade hydropower stations under complementary conditions. Whether it is long-term-short-term-real-time multi-time scale nested scheduling, or hierarchical and graded solution to achieve complementary power generation planning, its core is still the operation of power stations at the short-term scale, and does not involve the medium- and long-term joint optimization scheduling operation of cascade hydropower stations. It cannot solve the medium- and long-term joint scheduling problems of cascade hydropower stations under the wind-solar grid connection in large river basins, nor can it provide overall guidance and annual plan formulation basis for the operation of cascade reservoir hydropower stations. Summary of the Invention

[0006] In order to solve the problems existing in the prior art, the present invention provides a rolling-corrected medium- and long-term implicit random joint optimization scheduling method, which solves the medium- and long-term joint scheduling problem of cascade hydropower stations under the grid-connected wind and solar power in large river basins, and at the same time provides overall guidance and annual plan formulation basis for the operation of cascade reservoir hydropower stations, solving the problems mentioned in the above background technology.

[0007] To achieve the above-mentioned object, the present invention provides the following technical solution: a rolling-corrected medium- and long-term implicit random joint optimization scheduling method, comprising the following steps:

[0008] Step 1: Based on the situation of wind and photovoltaic grid connection, a deterministic medium- and long-term optimization scheduling model with the maximum system power generation as the objective function is established;

[0009] Step 2: Use the stepwise optimization algorithm (POA) to solve the deterministic medium- and long-term optimization scheduling model. Convert the multi-stage problem into multiple two-stage problems. Fix the variables of the current other stages, search and optimize the decision variables of the selected two stages, and after solving the current two-stage problem, enter the next two-stage calculation. Use the previous result as the initial condition for the next optimization. Repeat the optimization cycle until convergence, and obtain the optimal operation trajectory as the sample data set of the implicit random scheduling function.

[0010] Step 3: Parameter Optimization: Based on the Pattern Recognition and Regression Toolbox, the kernel function is introduced. The support vector machine (SVM) method is used to transform the nonlinear problem into a linear problem in a high-dimensional space and solve it. In the SVM model, the three main parameters that affect the regression prediction results are the penalty coefficient C, the kernel function parameter γ, and the insensitive loss coefficient p. The particle swarm optimization algorithm (PSO) based on biological phenomena is selected to optimize the main parameters of the SVM medium- and long-term implicit random scheduling function model. The combination of the main parameters is determined and the optimal parameter combination (C, γ, p) is optimized.

[0011] Step 4: Under the condition of full acceptance of wind and photovoltaic power, a medium- and long-term wind-solar-cascade hydropower implicit random joint scheduling function model is established, and the SVM method is used to solve it to obtain the water level regression prediction for each time period;

[0012] Step 5: Compared with the previous period, the water level values ​​obtained by regression prediction during the flood season suddenly increased too much. Therefore, the water level in each period was corrected through a rolling simulation correction strategy.

[0013] Preferably, the deterministic medium- and long-term optimization scheduling model is expressed as follows:

[0014]

[0015] Where, E is the total power generation of the complementary power generation system; is the total output of hydropower station i in time period t; is the total photovoltaic output during period t; is the total wind power output in period t; n is the total number of hydropower stations; T is the calculation period, which is one year; △t is the time interval between each period.

[0016] Preferably, in step 2, the specific calculation steps of the stepwise optimization algorithm POA are as follows:

[0017] (1) Determine the initial conditions and constraints;

[0018] (2) According to the order of power stations, optimize the water level Z at different times of each power station and fix the water level Z at time t and time t+2. i,t and Z i,t+2 No change, adjust the water level Z at time t+1 i,t+1 , so that the system power generation in the t and t+1 periods is maximized;

[0019] (3)Z i,t+1 ,Z i,t+1 ,…,Z i,T The calculation of is consistent with step (2). After traversing all reservoirs and time periods, a new trajectory is obtained. This new trajectory is used as the initial trajectory for a new round of iteration and steps (1) to (2) are repeated.

[0020] (4) Repeat step (3) until the end time, thereby obtaining the water level process of each reservoir in the cascade, the power generation flow process and the total power of the cascade under the initial conditions and constraint conditions;

[0021] (5) Compare the two rounds of trajectories. If the difference between the two does not meet the accuracy requirements, enter the next round of iteration; otherwise, the iteration is terminated and the optimization is completed.

[0022] Preferably, the constraints include water balance constraints, reservoir capacity constraints, discharge flow constraints, water level constraints, output constraints and initial and final water level constraints.

[0023] Preferably, the water balance constraint expression is:

[0024] V i,t+1 =V i,t +(q i,t -Q i,t -S i,t )×Δt

[0025]

[0026] Where V i,t+1 and V i,t are the initial and final storage capacities of reservoir i in period t; q i,t is the average inflow of reservoir i in period t; Q i,t is the power generation flow of power station i in period t; S i,t is the water discharge of reservoir i in period t; Q s i-1,t is the discharge of reservoir i-1; q' i-1,t is the interval flow between reservoir i and reservoir i-1;

[0027] The storage capacity constraint expression is:

[0028]

[0029] Where V i min and V i max are the upper and lower limits of the storage capacity of reservoir i in time period t respectively;

[0030] The downstream flow constraint expression is:

[0031] Q i min ≤Q i,t ≤Q i max

[0032] Where Q i min and Q imax are the upper and lower limits of the discharge flow of reservoir i in time period t;

[0033] The water level constraint expression is:

[0034] Z i min ≤Z i,t ≤Z i max

[0035] Where Z i min and Z i max are the upper and lower limits of the water level of reservoir i in time period t;

[0036] The output constraint expression is:

[0037] N i min ≤N i,t ≤N i max

[0038] Where N i min and N i max are the upper and lower limits of the output of power station i in time period t;

[0039] The initial and final water level constraint expressions are:

[0040]

[0041] Where Z i start and Z i end are the initial and final water levels of the flood season given to reservoir i, which are constants.

[0042] Preferably, in step three, the steps of optimizing the parameters of the particle swarm optimization algorithm PSO are as follows:

[0043] 1) Divide the sample set into training samples and prediction samples, and the training samples serve as the data samples for parameter optimization;

[0044] 2) Set the parameter value range, determine the fitness function of the optimization problem, and transform the problem of finding the optimal parameters into solving the maximum value of the fitness function;

[0045] 3) PSO optimization obtains the optimal parameter combination (C, γ, p).

[0046] Preferably, the fitness function is to minimize the root mean square error between the predicted value and the sample value, and the expression is:

[0047]

[0048] Where ld and lu are the upper and lower limits of each parameter.

[0049] Preferably, the medium- and long-term wind-solar-cascade hydropower implicit random joint dispatch function model is expressed as follows:

[0050]

[0051] Where Z i,t ,Z i,t+1 are the initial and final water levels of reservoir i in time period t; Z i+1,t is the initial water level of reservoir i+1 in time period t; Q i,t+1 is the average inflow to reservoir i during period t; Q q,t+1 is the inflow from reservoir i to reservoir i+1.

[0052] Preferably, in step 4, the steps of solving the problem using the SVM method are as follows:

[0053] ① Normalize the sample data to a range of [0,1], and the expression is:

[0054]

[0055] Where: x, y are sample data before and after normalization; x max ,x min ,y max ,y min are the maximum and minimum values ​​of the corresponding data;

[0056] ② Set the parameter values ​​of the PSO algorithm and the range of the optimized parameters, and use the Gaussian radial basis function to train the training samples 20 times to obtain 20 sets of optimized parameters;

[0057] ③ The parameter combination corresponding to the minimum fitness function value is considered optimal, and is input into the medium- and long-term wind-solar-cascade hydropower implicit random joint scheduling function model for regression prediction.

[0058] Preferably, in step 5, the rolling simulation correction strategy includes the following:

[0059] Strategy A: When the reservoir water level in a certain period rises too much compared to the previous period, resulting in the discharge flow not meeting the safe operation requirements of the power station, the water level is lowered to the guaranteed output of the hydropower station in the current period based on the water level drop at the end of the given period.

[0060] Strategy B: When the output of a certain period is less than the guaranteed output of the power station, the water level at the end of the period is lowered by a given water level drop until the output conditions are met.

[0061] The present invention has the following beneficial effects: By establishing a medium- and long-term implicit stochastic dispatch function model based on parameter optimization using a PSO-SVM, combined with a rolling simulation correction strategy, it optimizes the dispatch and operation of cascade hydropower stations that fully receive wind and solar power over the medium and long term. Comparative analysis shows that the dispatch and operation rules derived from the proposed dispatch function model combined with the rolling simulation correction strategy conform to existing reservoir dispatch rules and are an effective method for guiding the actual operation of cascade hydropower stations under the "bundled delivery" model of wind, solar, and hydropower. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 This is a schematic diagram of the implicit random optimization scheduling and rolling calculation process of the present invention;

[0063] Figure 2 Schematic diagram of the water level process comparison for verifying the POA and PSO-SVM models in the embodiment of the present invention, (a) Jinping I Reservoir; (b) Ertan Reservoir;

[0064] Figure 3 Schematic diagram of the comparison of water level processes of strategies A, B and the optimized scheduling model POA, (a) Jinping I Reservoir; (b) Ertan Reservoir;

[0065] Figure 4 Schematic diagram of the water level output process of Jinping I Reservoir, (a) Strategy A, (b) Strategy B;

[0066] Figure 5 Schematic diagram of the water level output process of Ertan Reservoir, (a) Strategy A, (b) Strategy B;

[0067] Figure 6 Schematic diagram of annual power generation comparison: (a) Jinping I Reservoir, (b) Ertan Reservoir, and (c) wind-solar-hydro multi-energy complementary system. DETAILED DESCRIPTION

[0068] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0069] Example 1

[0070] Against the backdrop of the continuous increase in the scale of wind and photovoltaic grid connection, the use of cascade hydropower to supplement the two and then "bundle and transmit" them to form a "wind, solar and water" multi-energy complementary energy system is an effective way to maximize the utilization of clean energy. The present invention innovatively establishes a medium- and long-term implicit random joint scheduling function model for cascade hydropower stations under wind and solar grid connection, optimizes the three main parameters of the model using the PSO algorithm, solves the model using the SVM method and combines the rolling simulation correction strategy to derive the operating rules of the cascade reservoirs, which are consistent with the reservoir scheduling law. Taking the cascade power stations in the lower reaches of the Yalong River as the research object, it shows that the implicit random scheduling function and its correction strategy proposed in the study are reasonable, providing theoretical and technical guidance for the actual operation of the reservoir.

[0071] The present invention provides a technical solution: a rolling-corrected medium- and long-term implicit random joint optimization scheduling method, comprising the following steps:

[0072] Step 1: Based on the situation of wind and photovoltaic grid connection, establish a deterministic medium- and long-term optimization scheduling model with the maximum system power generation as the objective function

[0073] Step 2: Use the stepwise optimization algorithm (POA) to solve the deterministic medium- and long-term optimization scheduling model. Convert the multi-stage problem into multiple two-stage problems. Fix the variables of the current other stages, search and optimize the decision variables of the selected two stages, and after solving the current two-stage problem, enter the next two-stage calculation. Use the previous result as the initial condition for the next optimization. Repeat the optimization cycle until convergence, and obtain the optimal operation trajectory as the sample data set of the implicit random scheduling function.

[0074] Step 3: Parameter Optimization: Based on the Pattern Recognition and Regression Toolbox, the kernel function is introduced. The support vector machine (SVM) method is used to transform the nonlinear problem into a linear problem in a high-dimensional space and solve it. In the SVM model, the three main parameters that affect the regression prediction results are the penalty coefficient C, the kernel function parameter γ, and the insensitive loss coefficient p. The particle swarm optimization algorithm (PSO) based on biological phenomena is selected to optimize the main parameters of the SVM medium- and long-term implicit random scheduling function model. The combination of the main parameters is determined and the optimal parameter combination (C, γ, p) is optimized.

[0075] Step 4: Under the condition of full acceptance of wind and photovoltaic power, a medium- and long-term wind-solar-cascade hydropower implicit random joint scheduling function model is established, and the SVM method is used to solve it to obtain the water level regression prediction for each time period;

[0076] Step 5: Compared with the previous period, the water level values ​​obtained by regression prediction during the flood season suddenly increased too much. Therefore, the water level in each period was corrected through a rolling simulation correction strategy.

[0077] Furthermore, in step 2, the specific calculation steps of the stepwise optimization algorithm POA are as follows:

[0078] (1) Determine the initial conditions and constraints;

[0079] (2) According to the order of power stations, optimize the water level Z at different times of each power station and fix the water level Z at time t and time t+2. i,t and Z i,t+2 No change, adjust the water level Z at time t+1 i,t+1 , so that the system power generation in the t and t+1 periods is maximized;

[0080] (3)Z i,t+1 ,Z i,t+1 ,…,Z i,T The calculation of is consistent with step (2). After traversing all reservoirs and time periods, a new trajectory is obtained. This new trajectory is used as the initial trajectory for a new round of iteration and steps (1) to (2) are repeated.

[0081] (4) Repeat step (3) until the end time, thereby obtaining the water level process of each reservoir in the cascade, the power generation flow process and the total power of the cascade under the initial conditions and constraint conditions;

[0082] (5) Compare the two rounds of trajectories. If the difference between the two does not meet the accuracy requirements, enter the next round of iteration; otherwise, the iteration is terminated and the optimization is completed.

[0083] Furthermore, the constraints specifically include:

[0084] Water balance constraints:

[0085] V i,t+1 =V i,t +(q i,t -Q i,t -S i,t )×Δt

[0086]

[0087] Where V i,t+1 and V i,t are the initial and final storage capacities of reservoir i in period t; q i,t is the average inflow of reservoir i in period t; Q i,t is the power generation flow of power station i in period t; S i,t is the water discharge of reservoir i in period t; Q s i-1,t is the discharge of reservoir i-1; q' i-1,t is the interval flow between reservoir i and reservoir i-1;

[0088] Storage capacity constraints:

[0089] V i min ≤V i,t ≤V i max

[0090] Where V i min and V i max are the upper and lower limits of the storage capacity of reservoir i in time period t respectively;

[0091] Downflow flow restriction:

[0092] Q i min ≤Q i,t ≤Q i max

[0093] Where Q i min and Q i max are the upper and lower limits of the discharge flow of reservoir i in time period t;

[0094] Water level constraint:

[0095] Z i min ≤Z i,t ≤Z i max

[0096] Where Z i min and Z i max are the upper and lower limits of the water level of reservoir i in time period t;

[0097] Output constraints:

[0098] N i min ≤N i,t ≤N i max

[0099] Where N i min and N i max are the upper and lower limits of the output of power station i in time period t;

[0100] Initial and final water level constraints:

[0101]

[0102] Where Z i start and Zi end are the initial and final water levels of the flood season given to reservoir i, which are constants.

[0103] Furthermore, in step three, the steps for optimizing the parameters of the particle swarm optimization algorithm PSO are as follows:

[0104] 1) Divide the sample set into training samples and prediction samples, and the training samples serve as the data samples for parameter optimization;

[0105] 2) Set the parameter value range, determine the fitness function of the optimization problem, and transform the problem of finding the optimal parameter into solving the maximum value of the fitness function. The fitness function is to minimize the root mean square error between the predicted value and the sample value, and the expression is:

[0106]

[0107] Where ld and lu are the upper and lower limits of each parameter value;

[0108] 3) PSO optimization obtains the optimal parameter combination (C, γ, p).

[0109] Furthermore, the medium- and long-term wind-solar-cascade hydropower implicit random joint dispatch function model is expressed as follows:

[0110]

[0111] Where Z i,t ,Z i,t+1 are the initial and final water levels of reservoir i in time period t; Z i+1,t is the initial water level of reservoir i+1 in time period t; Q i,t+1 is the average inflow to reservoir i during period t; Q q,t+1 is the inflow from reservoir i to reservoir i+1.

[0112] Furthermore, in step 4, the steps of solving the problem by the SVM method are as follows:

[0113] ① Normalize the sample data to a range of [0,1], and the expression is:

[0114]

[0115] Where: x, y are sample data before and after normalization; x max ,x min ,y max ,y min are the maximum and minimum values ​​of the corresponding data;

[0116] ② Set the parameter values ​​of the PSO algorithm and the range of the optimized parameters, and use the Gaussian radial basis function to train the training samples 20 times to obtain 20 sets of optimized parameters;

[0117] ③ The parameter combination corresponding to the minimum fitness function value is considered optimal, and is input into the medium- and long-term wind-solar-cascade hydropower implicit random joint scheduling function model for regression prediction.

[0118] Furthermore, in step 5, the rolling simulation correction strategy includes the following:

[0119] Strategy A: When the reservoir water level in a certain period rises too much compared to the previous period, resulting in the discharge flow not meeting the safe operation requirements of the power station, the water level is lowered to the guaranteed output of the hydropower station in the current period based on the water level drop at the end of the given period.

[0120] Strategy B: When the output of a certain period is less than the guaranteed output of the power station, the water level at the end of the period is lowered by a given water level drop until the output conditions are met.

[0121] Example 2

[0122] Step 1: Establish a deterministic optimization scheduling model

[0123] (1) Objective function

[0124] The medium- and long-term optimal scheduling uses optimization theories and methods to seek the optimal operation scheduling mode, optimal strategy and optimal decision for the system, hydropower station and its reservoir. The results can be used as the overall guidance for the operation of the reservoir and hydropower station and the basis for formulating annual plans. It can also provide the load distribution of the power station for short-term scheduling calculations and provide a basis for water consumption estimation. The traditional optimization scheduling method does not take into account wind and photovoltaic inputs and is not suitable for clean energy complementary power generation systems such as wind-water / photovoltaic-water / wind-photovoltaic-water. Therefore, the present invention takes into account the situation of wind and photovoltaic grid connection, and establishes a deterministic medium- and long-term optimal scheduling model (Formula 1-1) with the maximum system power generation as the objective function. The optimization scheduling results obtained by solving the model are used as sample data.

[0125]

[0126] Where, E is the total power generation of the complementary power generation system; is the total output of hydropower station i in period t; O t P is the total photovoltaic output during period t; O t W is the total wind power output in period t; n is the total number of hydropower stations; T is the calculation period, which is one year; △t is the time interval between each period.

[0127] (2) Constraints

[0128] ① Water balance constraints:

[0129] V i,t+1 =V i,t+(q i,t -Q i,t -S i,t )×Δt (1-2)

[0130]

[0131] Where V i,t+1 and V i,t are the initial and final storage capacities of reservoir i in period t; q i,t is the average inflow of reservoir i in period t; Q i,t is the power generation flow of power station i in period t; S i,t is the water discharge of reservoir i in period t; Q s i-1,t is the discharge of reservoir i-1; q' i-1,t is the flow rate between reservoir i and reservoir i-1.

[0132] ② Storage capacity constraints:

[0133] V i min ≤V i,t ≤V i max (1-4)

[0134] Where V i min and V i max are the upper and lower limits of the storage capacity of reservoir i in time period t.

[0135] ③ Downflow flow restriction

[0136] Q i min ≤Q i,t ≤Q i max (1-5)

[0137] Where Q i min and Q i max are the upper and lower limits of the discharge flow of reservoir i in time period t.

[0138] ④ Water level constraint

[0139] Z i min ≤Z i,t ≤Z i max (1-6)

[0140] Where Z i min and Z imax are the upper and lower limits of the water level of reservoir i in time period t.

[0141] ⑤ Output constraint

[0142] N i min ≤N i,t ≤N i max (1-7)

[0143] Where N i min and N i max are the upper and lower limits of the output of power station i in time period t.

[0144] ⑥ Initial and final water level constraints

[0145]

[0146] Where Z i start and Z i end are the initial and final water levels of the flood season given to reservoir i, which are constants.

[0147] Step 2: Deterministic optimization scheduling model solution method

[0148] As an improved algorithm for dynamic programming, the Stepwise Optimization Algorithm (POA) can transform solving a multi-stage problem into solving multiple two-stage problems. The main idea of ​​this method is to fix the variables of the current other stages, search and optimize the decision variables of the selected two stages, and after solving the current two-stage problem, enter the calculation of the next two-stage. The results of the previous optimization are used as the initial conditions for the next optimization, and the optimization is repeated until convergence. The specific calculation steps are as follows:

[0149] (1) Determine the initial conditions and constraints.

[0150] (2) According to the order of power stations, optimize the water level Z at different times of each power station. i,t and Z i,t+2 No change, adjust the water level Z at time t+1 i,t+1 , so that the system power generation in the t and t+1 periods is maximized.

[0151] (3)Z i,t+1 ,Z i,t+1 ,…,Z i,T The calculation of is consistent with step (2). After traversing all reservoirs and time periods, a new trajectory is obtained. This new trajectory is used as the initial trajectory for a new round of iteration and steps (1) to (2) are repeated.

[0152] (4) Repeat step (3) until the end time. Thus, the water level process of each reservoir in the cascade, the power generation flow process and the total power of the cascade under the initial conditions and constraint conditions are obtained.

[0153] (5) Compare the two rounds of trajectories. If the difference between the two does not meet the accuracy requirements, enter the next round of iteration; otherwise, the iteration is terminated and the optimization is completed.

[0154] The present invention divides a year into 36 calculation periods, fixes the water level values ​​at the initial and final moments, and uses the POA algorithm to repeatedly optimize until convergence to obtain the optimal operation trajectory as the sample data set of the implicit random scheduling function.

[0155] Step 3: Parameter optimization

[0156] (1) Support Vector Machine

[0157] Support vector machines (SVMs) are a machine learning method based on statistical theory and are widely used in pattern recognition and regression problems. Based on the Pattern Recognition and Regression Toolbox, kernel functions are introduced and the SVM method is used to transform nonlinear problems into linear problems in a high-dimensional space and solve them. The steps are as follows:

[0158] ①Introduce the Gaussian radial basis kernel function to map the sample points to the high-dimensional space, and its expression is:

[0159] K(χ i ,χ j )=exp(-γ||χ i -χ j || 2 ),γ>0 (3-1)

[0160] Where: K(χ i ,χ j ) is the kernel function; γ is the kernel function parameter; χ i , χ j are the independent variables and dependent variables, and i=j is the dimension.

[0161] ②After kernel function conversion, the decision function becomes:

[0162]

[0163] Where: w is the weight coefficient and b is the bias phase.

[0164] ③Use Lagrange multiplier method to solve convex quadratic programming problems

[0165] ④ Obtain the regression estimation function through learning, and use the regression estimation function to perform nonlinear classification or regression prediction.

[0166] (2) Particle Swarm Optimization

[0167] Particle Swarm Optimization (PSO) is a widely used swarm intelligence algorithm. Its basic principle is to seek the optimal solution to the optimization problem through information transmission and sharing among individuals in the population. It has the characteristics of few parameters, strong versatility, simple operation, and easy implementation. This paper selects the PSO algorithm based on biological phenomena to optimize the main parameters of the long-term implicit random scheduling function model of the SVM. The specific steps for solving the optimization problem are as follows:

[0168] ① Initialize the particle swarm, including the particle position x i , speed v i And the population size M, calculate the particle fitness value F it (i).

[0169] ② Compare the F of the particles it (i) and individual extreme value P best (i), when the former is less than the latter, update the current individual extreme value, that is, P best (i) = F it (i). Where P best (i) is the fitness value of the i-th particle when it searches for the optimal position.

[0170] ③ Compare the F of the particles it (i) and the global extreme value g best (i), when the former is less than the latter, update the current global extreme value, i.e. g best (i) = P best (i). Where g best (i) is the fitness value of the entire particle swarm when it searches for the optimal position.

[0171] ④ When the individual optimum and the global optimum are found, update the particle position x i and speed v i , the expression is as follows:

[0172]

[0173] Where ω is the inertia weight; c1 and c2 are learning factors; r1 and r2 are random numbers in the range of [0, 1].

[0174] ⑤ Iterate the calculation until the termination condition is met and obtain the global optimal value as the fitness value.

[0175] (3) Parameter optimization process

[0176] When using the SVM method for classification calculations, there may be cases where samples remain linearly inseparable after high-dimensional space mapping, so it is necessary to properly adjust and set the SVM parameters. In the SVM model, the three main parameters that affect the regression prediction results are the penalty coefficient C, the kernel function parameter γ, and the insensitive loss coefficient p. To better determine the combination of these main parameters, PSO is combined with the SVM learning model and used to optimize the parameters. The main steps are as follows:

[0177] ① Divide the sample set into training samples and prediction samples, and the training samples serve as data samples for parameter optimization.

[0178] ② Set the parameter value range, determine the fitness function of the optimization problem, and transform the problem of finding the optimal parameter into solving the maximum value of the fitness function. The fitness function of the present invention is to minimize the root mean square error between the predicted value and the sample value, and the expression is:

[0179]

[0180] Where ld and lu are the upper and lower limits of each parameter.

[0181] ③PSO optimization obtains the optimal parameter combination (C, γ, p).

[0182] Step 4: Establish a medium- and long-term wind-solar-cascade hydropower implicit stochastic joint dispatch function

[0183] Cascade hydropower stations are closely connected in terms of hydraulics, and the mutual influence between their operating modes during joint operation cannot be ignored. Therefore, under the condition of full reception of wind and photovoltaic power, the following implicit stochastic optimization scheduling function model is established:

[0184]

[0185] Where: Z i,t ,Z i,t+1 are the initial and final water levels of reservoir i in time period t; Z i+1,t is the initial water level of reservoir i+1 in time period t; Q i,t+1 is the average inflow to reservoir i during period t; Q q,t+1 is the inflow from reservoir i to reservoir i+1.

[0186] This paper takes the cascade hydropower stations in the lower reaches of the Yalong River as the research object and explores the application of the model in Jinping I (JY) and Ertan (ET) reservoirs with seasonal regulation capabilities and above. The specific expression is:

[0187]

[0188] Where: Z JY,t ,Z JY,t+1 are the initial and final water levels of Jinping Level 1 in period t; QJY,t+1 is the average inflow flow of Jinping Level 1 during period t; Z ET,t ,Z ET,t+1 are the initial and final water levels of Ertan during period t; Q q,t+1 is the flow rate between Jinping Level 1 and Level 2 during time period t.

[0189] Step 5: SVM solution

[0190] Based on the Pattern Recognition and Regression Toolbox, the SVM method of Matlab2018a platform is used to solve the established scheduling function model. The steps are as follows:

[0191] (1) Normalize the sample data to a range of [0,1], and the expression is:

[0192]

[0193] Where: x, y are sample data before and after normalization; x max ,x min ,y max ,y min are the maximum and minimum values ​​of the corresponding data.

[0194] (2) Set the parameter values ​​of the PSO algorithm and the range of the optimized parameters, and use the Gaussian radial basis function to train the training samples 20 times to obtain 20 sets of optimized parameters.

[0195] (3) The parameter combination corresponding to the minimum fitness function value in step (2) is considered optimal and is input into the medium- and long-term implicit random scheduling function model for regression prediction.

[0196] Step 6: Rolling simulation correction

[0197] The water level values ​​obtained by SVM regression prediction during the flood season increased too much compared to the previous period, which is not conducive to the safe and stable operation of the hydropower station. To avoid this situation, the following two strategies are used to correct the water level in each period:

[0198] Strategy A: When the reservoir water level in a certain period rises too much compared to the previous period, resulting in the discharge flow not meeting the safe operation requirements of the power station, the water level is lowered to the guaranteed output of the hydropower station in the current period based on the water level drop at the end of the given period.

[0199] Strategy B: When the output of a certain period is less than the guaranteed output of the power station, the water level at the end of the period is lowered by a given water level drop until the output conditions are met.

[0200] The entire parameter optimization, model solution calculation and rolling simulation correction process is as follows Figure 1 shown.

[0201] Example 3

[0202] The present invention selects the cascade hydropower stations in the lower reaches of the Yalong River as the research object, takes the end-of-period water levels of Jinping I and Ertan as the decision variables, and the initial water level and inflow of Jinping I, the initial water level of Ertan, the flow between Jinping I and Ertan, and the wind and solar output within the period as independent variables. The calculation period is from November to October of each year, the calculation cycle is ten days, and 8500MW is the minimum output limit of the system. The POA algorithm is used to solve the deterministic optimization model to obtain a long series of optimization scheduling results and use them as data samples of the implicit random scheduling function. The data samples are divided into training samples (November 1959 to October 2002) and test samples (November 2002 to October 2012). The training samples are used to calibrate the optimal parameter combination of the model, and the test samples are used to test the effect of the implicit random scheduling model based on parameter optimization.

[0203] Parameter settings

[0204] The main parameters of the particle swarm algorithm are: learning factors c1, c2, generally ranging from [0, 4]; population size; maximum number of iterations; inertia weight. According to the empirical method, the particle swarm parameters are set to: learning factor c1 = c2 = 2; speed inertia weight ω v =1.2, position inertia weight ω p =0.8; population size M = 10; maximum number of iterations A = 100; upper bound of penalty coefficient Kernel function parameters Insensitive loss coefficient p∈(0.01,0.05).

[0205] To reduce the impact of randomness on the optimization results, PSO was used to fit the training samples 20 times, and the parameter combination with the minimum root mean square error was selected. The optimal parameter combination for each decade of the implicit random scheduling function model was obtained, as shown in Table 1.

[0206] Table 1 Optimal parameter combinations for each decade

[0207]

[0208]

[0209] In the training sample data, the water levels at Jinping I during period #36 and Ertan during periods #11-12, #14-16, and #35-36 remained unchanged, at 1880 meters and 1200 meters, respectively. The root mean square error (RMS) for the 20 runs was zero, indicating no optimal parameter combination, indicating that both reservoirs can be operated directly at their corresponding water levels.

[0210] Result Analysis

[0211] The optimal parameters in Table 1 are input into the established SVM implicit random joint scheduling function model to obtain the medium- and long-term operation process of Jinping I and Ertan reservoirs. The results of the test period (2002-2011) are compared with the water level process of the POA algorithm optimized scheduling under the same time span, as shown in Figure 2. Figure 2 As shown in Figure 2, the water level variation trend obtained by the PSO-SVM model is roughly the same as the optimized scheduling result. However, some water levels during the flood season are too high, which is not conducive to the safe and stable operation of the hydropower station. Therefore, the two proposed strategies are used to correct some water levels during the test period.

[0212] After the rolling simulation correction of strategies A and B, combined with the existing optimization operation results, the water level processes of the two reservoirs under three conditions are obtained as follows: Figure 3 As shown in the figure, the water level processes of the two reservoirs are roughly the same as those under the corresponding scenarios, but the water level change trend under Strategy A is more consistent with the water level change trend under optimized scheduling. Furthermore, under both schemes, the water level of Jinping I Reservoir decreases evenly from the initial stage to the dead water level, while the water level of Ertan Reservoir maintains a higher level in the early stage and decreases to the dead water level in the later stage.

[0213] Figure 4-Figure 5 The water level output curves of Jinping I and Ertan reservoirs under two scenarios are presented. It can be seen that, compared to Strategy A, Strategy B shows a greater and earlier water level drawdown at Jinping I reservoir, as it achieves guaranteed output throughout the calculation period. Although guaranteed output was not achieved from late May to early June in some years due to upstream water inflow and normal reservoir operation, guaranteed output was generally achieved during the dry season. Under Strategy B, Ertan reservoir initially maintained a high water level, similar to Strategy A. Subsequently, water level fluctuations occurred from early January to early April before returning to normal storage levels. This was because upstream water inflows during this period could not meet the power station's guaranteed output, so water was diverted from the reservoir to achieve the output target. Finally, the water level drawdown concentratedly reached the dead level before the flood season. The output of the cascade power stations under the two scenarios differed significantly, but no single period exhibited excessively low output. Furthermore, under Strategy B, the output of the two power stations during the dry season was more stable, significantly improving the annual dry season output.

[0214] Figure 6 The annual power generation of Jinping I and II Tan reservoirs, as well as the wind-solar-hydro hybrid system, during the test period is shown. Clearly, the total annual power generation under Strategies A and B is essentially the same, slightly lower than the optimal scheduling model. This is because the POA algorithm must meet the system's minimum output requirements when optimizing scheduling. Overall, the water level and output curves derived from the PSO-SVM implicit stochastic scheduling function model combined with the rolling simulation correction strategy are consistent with the existing scheduling rules, and the simulation results are considered reasonable.

[0215] This paper establishes a medium- and long-term implicit stochastic dispatch function model based on parameter optimization using a PSO-SVM, combined with a rolling simulation correction strategy, to study the dispatch and operation rules for cascade hydropower stations that fully receive wind and solar power output over the medium and long term. Simultaneously, combined with the POA optimization dispatch results, a comparative analysis of water level processes, output processes, and annual power generation during the test period was conducted. The results show that the dispatch and operation rules derived from the proposed dispatch function model combined with the rolling simulation correction strategy are consistent with existing reservoir dispatch patterns and are an effective method for guiding the actual operation of cascade hydropower stations under the "bundled transmission" model of wind, solar, and water.

[0216] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A rolling-corrected medium- and long-term implicit random joint optimization scheduling method, characterized in that: The steps include: Step 1: Based on the situation of wind and photovoltaic grid connection, a deterministic medium- and long-term optimization scheduling model with the maximum system power generation as the objective function is established; the deterministic medium- and long-term optimization scheduling model is expressed as follows: Where, E is the total power generation of the complementary power generation system; is the total output of hydropower station i in time period t; is the total photovoltaic output during period t; is the total wind power output in period t; n is the total number of hydropower stations; T is the calculation period, which is one year; △t is the time interval between each period; Step 2: Use the stepwise optimization algorithm (POA) to solve the deterministic medium- and long-term optimization scheduling model. Convert the multi-stage problem into multiple two-stage problems. Fix the variables of the current other stages, search and optimize the decision variables of the selected two stages, and after solving the current two-stage problem, enter the next two-stage calculation. Use the previous result as the initial condition for the next optimization. Repeat the optimization cycle until convergence, and obtain the optimal operation trajectory as the sample data set of the implicit random scheduling function. Step 3: Under the condition of full wind and photovoltaic power generation, a medium- and long-term wind-solar-cascade hydropower implicit random joint scheduling function model is established, and the SVM method is used to solve it to obtain the water level regression prediction for each time period; the medium- and long-term wind-solar-cascade hydropower implicit random joint scheduling function model is expressed as follows: Where Z i,t ,Z i,t+1 are the initial and final water levels of reservoir i in time period t; Z i+1,t is the initial water level of reservoir i+1 in time period t; Q i,t+1 is the average inflow to reservoir i during period t; Q q,t+1 is the interval inflow between reservoir i and reservoir i+1; The steps for solving the problem using the SVM method are as follows: ① Normalize the sample data to a range of [0,1], and the expression is: Where: x, y are sample data before and after normalization; x max ,x min ,y max ,y min are the maximum and minimum values ​​of the corresponding data; ② Set the parameter values ​​of the PSO algorithm and the range of the optimized parameters, and use the Gaussian radial basis function to train the training samples 20 times to obtain 20 sets of optimized parameters; ③ The parameter combination corresponding to the minimum fitness function value is considered optimal and input into the medium- and long-term wind-solar-cascade hydropower implicit random joint dispatch function model for regression prediction; The steps for optimizing the parameters of the particle swarm optimization algorithm PSO are as follows: 1) Divide the sample set into training samples and prediction samples, and the training samples serve as the data samples for parameter optimization; 2) Set the parameter value range, determine the fitness function of the optimization problem, and transform the problem of finding the optimal parameters into solving the maximum value of the fitness function; 3) PSO optimization obtains the optimal parameter combination (C, γ, p), where C represents the penalty coefficient, γ represents the kernel function parameter, and p represents the insensitive loss coefficient; Step 4: The water level values ​​of the flood season obtained by regression prediction are corrected for each period through a rolling simulation correction strategy.

2. The rolling-corrected medium- and long-term implicit-stochastic joint optimization scheduling method according to claim 1 is characterized by: In step 2, the specific calculation steps of the stepwise optimization algorithm POA are as follows: (1) Determine the initial conditions and constraints; (2) According to the order of power stations, optimize the water level Z at different times of each power station and fix the water level Z at time t and time t+2. i,t and Z i,t+2 No change, adjust the water level Z at time t+1 i,t+1 , so that the system power generation in the t and t+1 periods is maximized; (3)Z i,t+1 ,Z i,t+1 ,…,Z i,T The calculation of is consistent with step (2). After traversing all reservoirs and time periods, a new trajectory is obtained. This new trajectory is used as the initial trajectory for a new round of iteration and steps (1) to (2) are repeated. (4) Repeat step (3) until the end time, thereby obtaining the water level process of each reservoir in the cascade, the power generation flow process and the total power of the cascade under the initial conditions and constraint conditions; (5) Compare the two rounds of trajectories. If the difference between the two does not meet the accuracy requirements, enter the next round of iteration; otherwise, the iteration is terminated and the optimization is completed.

3. The rolling-corrected medium- and long-term implicit-stochastic joint optimization scheduling method according to claim 2 is characterized by: The constraints include water balance constraints, reservoir capacity constraints, discharge flow constraints, water level constraints, output constraints and initial and final water level constraints.

4. The rolling-corrected medium- and long-term implicit-stochastic joint optimization scheduling method according to claim 3 is characterized by: The water balance constraint expression is: V i,t+1 =V i,t +(q i,t -Q i,t -S i,t )×Δt Where V i,t+1 and V i,t are the initial and final storage capacities of reservoir i in period t; q i,t is the average inflow of reservoir i in period t; Q i,t is the power generation flow of power station i in period t; S i,t is the water discharge of reservoir i in period t; Q s i-1,t is the discharge of reservoir i-1; q' i-1,t is the interval flow between reservoir i and reservoir i-1; The storage capacity constraint expression is: In i min ≤V i,t ≤V i max Where V i min and V i max are the upper and lower limits of the storage capacity of reservoir i in time period t respectively; The downstream flow constraint expression is: Q i min ≤Q i,t ≤Q i max Where Q i min and Q i max are the upper and lower limits of the discharge flow of reservoir i in time period t; The water level constraint expression is: WITH i min ≤Z i,t ≤Z i max Where Z i min and Z i max are the upper and lower limits of the water level of reservoir i in time period t; The output constraint expression is: N i min ≤N i,t ≤N i max Where N i min and N i max are the upper and lower limits of the output of power station i in time period t; The initial and final water level constraint expressions are: Where Z i start and Z i end are the initial and final water levels of the flood season given to reservoir i, which are constants.

5. The rolling-corrected medium- and long-term implicit-stochastic joint optimization scheduling method according to claim 1 is characterized by: The fitness function is to minimize the root mean square error between the predicted value and the sample value, and the expression is: Where ld and lu are the upper and lower limits of each parameter.

6. The rolling-corrected medium- and long-term implicit-stochastic joint optimization scheduling method according to claim 1 is characterized by: In step 4, the rolling simulation correction strategy includes the following: Strategy A: When the reservoir water level in a certain period rises too much compared to the previous period, resulting in the discharge flow not meeting the safe operation requirements of the power station, the water level is lowered to the guaranteed output of the hydropower station in the current period based on the water level drop at the end of the given period. Strategy B: When the output of a certain period is less than the guaranteed output of the power station, the water level at the end of the period is lowered by a given water level drop until the output conditions are met.

Images