Cascade water, wind and light medium and long term robust scheduling method generated in combination with extreme scene

Abstract

The invention discloses a cascade water, wind and light medium and long term robust scheduling method generated in combination with an extreme scene, and belongs to the field of reservoir scheduling in hydroelectric power generation, and the method comprises the steps: S1, constructing a water, wind and photovoltaic output extreme scene set based on a Copula-GAN hybrid method, and obtaining a comprehensive scene set in combination with historical measured data; s2, taking maximized on-grid electric quantity, minimum maximum abandoned electric quantity, minimum maximum power shortage, minimum time period average power shortage and minimum time period average abandoned electric quantity as optimization targets, introducing a risk penalty term aiming at an extreme scene set while considering constraint conditions, and further constructing a multi-target robust optimization model considering extreme risk penalty, solving by adopting a cuckoo algorithm; s3, taking the comprehensive scene set as input, and obtaining an optimization result through a multi-target robust optimization model considering extreme risk penalty; and S4, on the basis of an optimization result, constructing a linear scheduling function as a robust scheduling rule by taking the available energy as an independent variable and the period discharge flow as a dependent variable.

Description

Technical Field

[0001] This invention belongs to the field of hydropower reservoir scheduling technology, specifically a robust long-term scheduling method for cascade hydropower, wind power, and solar power generated in combination with extreme scenarios. Background Technology

[0002] With the proposal of dual-carbon targets and the continuous advancement of new power system construction, the grid-connected scale of renewable energy, represented by wind power and photovoltaics, continues to expand. However, due to the influence of climate and geographical location, the output of wind and photovoltaic power exhibits randomness and volatility, and their large-scale integration poses a severe challenge to the safe and stable operation of the power grid. To address the uncertainty of renewable energy output, scenario analysis has been widely applied in power system planning and operation research. This method constructs a representative set of wind, solar, and hydropower output scenarios to simulate the system's operating state under different conditions, providing an important basis for formulating scheduling schemes for multi-energy complementary systems of hydropower, wind power, and solar power.

[0003] However, existing scene generation methods often only focus on scene generation for a single resource, or only generate typical scenes for composite wind and solar resources without fully considering extreme situations, and lack effective means for robust optimization and scheduling of water-wind-solar complementary systems under extreme climatic conditions. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing research, which focuses only on scene generation for single resources or only on typical scenes of composite wind and solar resources without fully considering extreme situations. Furthermore, it lacks effective means for robust optimization and scheduling of water-wind-solar complementary systems under extreme climatic conditions. This invention proposes a cascade water-wind-solar system long-term robust scheduling method that combines extreme scene generation. Coordinate the power generation plans of cascade hydropower with wind and solar energy under extreme wind and solar power output conditions to ensure that the power system can still achieve a reliable balance of power generation and output under extreme scenarios.

[0005] To address the aforementioned technical problems, this invention provides a robust long-term scheduling method for cascaded water, wind, and solar power systems generated in extreme scenarios, comprising: S1. Construct extreme scenario sets for hydropower, wind power, and photovoltaic power output based on the Copula-GAN hybrid method, and obtain a comprehensive scenario set by combining historical measured data; S2. The optimization objectives are to maximize the amount of electricity generated online, minimize the maximum amount of electricity wasted, minimize the maximum amount of electricity shortage, minimize the average amount of electricity shortage during a given period, and minimize the average amount of electricity wasted during a given period. While considering the constraints, a risk penalty term is introduced for the extreme scenario set. Then, a multi-objective robust optimization model considering extreme risk penalties is constructed and solved using the Cuckoo algorithm. S3. Using the comprehensive scenario set as input, the optimization results are obtained through a multi-objective robust optimization model that considers extreme risk penalties; S4. Based on the optimization result, a linear scheduling function is constructed as a robust scheduling rule, with available energy as the independent variable and period discharge as the dependent variable.

[0006] As preferred, the S1 specific steps are as follows: S11. The original runoff, wind speed and light intensity data are subjected to wavelet transform denoising, and the denoised data is subjected to standardization processing, and is converted into water power output, wind power output and photovoltaic output according to the corresponding conversion relationship; S12. The threshold of extreme event is defined based on the historical output quantile, the occurrence period of extreme event is marked, and the feature vector is constructed; S13. The marginal probability distribution of water power output, wind power output and photovoltaic output is fitted respectively, and the joint probability distribution is obtained by using vine-copula structure; the noise vector joint Copula sampling vector is taken as the generator input, and the basic scenario set is generated based on GAN; S14. The feature vector is taken as the conditional label, the noise vector is input into CGAN to generate the basic extreme scenario set; the Wasserstein distance is taken as the similarity measure, and the extreme scenario subset is extracted from the basic extreme scenario set by K-medoids clustering; S15. The historical output data, the basic scenario set and the extreme scenario subset are fused, and the comprehensive scenario set is constructed according to the seasonal characteristics.

[0007] As preferred, the S13 is realized by a generative adversarial network (GAN) generation module: The composition of GAN includes a generator and a discriminator, and the discriminator is used to help the generator to better learn the distribution of the original data, the input is data, and the output is the probability of judging that the data is the original data.

[0008] As preferred, the S14 is realized by a conditional generative adversarial network (CGAN): the noise joint extreme event feature is taken as the input, the extreme event feature is taken as the conditional label to constrain the generation process, so that the conditional generator learns the extreme event distribution; the output is the extreme scenario set, and the discriminator distinguishes the real extreme scenario from the generated extreme scenario, and the generation quality is improved through adversarial training.

[0009] As preferred, in the S15: hydrological feature values are calculated for the historical data and the basic scenario set and the extreme scenario set, including monthly average flow, monthly average output fluctuation rate and output peak-valley difference; the scenarios with feature values within the preset flood period threshold range are classified into the flood period comprehensive scenario set, and the scenarios with feature values within the preset dry period threshold range are classified into the dry period comprehensive scenario set.

[0010] As preferred, the maximization of total online power is specifically represented as: ; ; The minimization of the maximum curtailment amount is specifically represented as: ; ; The minimization of the maximum shortage amount is specifically represented as: ; ; The minimization of the period average shortage amount is specifically represented as: ; The minimization of the period average curtailment amount is specifically represented as: ; In the formula: , , are the hydropower, wind power and photovoltaic output of the power plant in the t period, is the shortage amount of the scene in the t period.

[0011] As a preferred, the risk penalty term introduced for the extreme scenario set is specifically represented as: ; In the formula, R is the total penalty of the system extreme risk, and are the penalty coefficients of the shortage and curtailment, respectively.

[0012] As a preferred, the optimization result is the corresponding optimized reservoir water level process, water, wind and light output process, and the six variables are available water, available energy, reservoir water level, reservoir storage, discharge flow and system output, which are obtained by corresponding calculation.

[0013] As a preferred, in S4, the Pearson correlation coefficients of the six variables are calculated, and the independent variables highly correlated with the decision variables are screened according to a preset threshold. Because the available energy AE can best reflect the input conditions of the three resources of water, wind and light in the system, the system available energy is finally taken as the independent variable, and the period discharge flow is taken as the decision variable to construct a linear scheduling function.

[0014] As a preferred, the basic type of the linear scheduling function is as follows: ; In the formula, k and K are the number and total number of the scheduling function; and are the decision variables of the scheduling function, respectively; and are the parameters of the scheduling function.

[0015] The beneficial effects of the present application are: 1、 The present scheme uses scene generation technology to generate extreme scenes, proposes a cascade water and wind light long-term robust scheduling rule combined with extreme scene generation, derives a linear scheduling function of the water and wind light complementary system, and provides a new idea for improving the safety and reliability of large-scale wind and light grid connection.

[0016] 2、 The present scheme takes the maximization of total on-grid power of cascade hydropower stations under extreme climate conditions, minimization of maximum power abandonment, minimization of maximum power shortage, minimization of period average power shortage, and minimization of period average power abandonment as objective functions, constructs a robust optimization model, extracts a robust scheduling rule, and enhances the power supply reliability under extreme climate conditions. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 The flowchart of the present application; Figure 2 The cascade hydropower station scheduling rule of the lower reaches of the Jinsha River Figure One (Taking Wudongde as an example); Figure 3 The cascade hydropower station scheduling rule of the lower reaches of the Jinsha River Figure Two ; Figure 4 The cascade hydropower station scheduling rule of the lower reaches of the Jinsha River Figure Three ; Figure 5 The cascade hydropower station scheduling rule of the lower reaches of the Jinsha River Figure Four ; Figure 6 The cascade hydropower station scheduling rule of the lower reaches of the Jinsha River Figure Five ; Figure 7 The cascade hydropower station scheduling rule of the lower reaches of the Jinsha River Figure Six . DETAILED DESCRIPTION

[0018] In order to make the purpose, technical scheme and advantages of the present application clearer and more apparent, the present application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific embodiments described herein are only one of the best embodiments of the present application, which are used to explain the present application and do not limit the protection scope of the present application. All other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0019] Example 1: as shown in Figure 1 -7, a cascade water and wind light long-term robust scheduling method combined with extreme scene generation, comprising: S1, based on the Copula-GAN hybrid method, an extreme scenario set of hydroelectric, wind power and photovoltaic output is constructed, and a comprehensive scenario set is obtained combined with historical measured data.

[0020] S1 specific steps are as follows: S11, wavelet transform is performed on the original runoff, wind speed and solar intensity data to remove measurement errors and random fluctuations. The denoised data is standardized to unify the dimension. And according to the corresponding conversion relationship, it is converted into hydroelectric output Phydro, wind power output Pwind and photovoltaic output Ppv respectively; S12, based on the historical output quantile, the threshold of extreme event is defined, the occurrence period tE of extreme event is marked, and the feature vector E=[Phydro,tE,Pwind,tE,Ppv,tE]; S13, the marginal probability distribution of Phydro, Pwind and Ppv is fitted respectively, the joint probability distribution C(uH, uW, uP) is obtained by using vine-copula structure; the noise vector z~N(0, 1) is combined with the Copula sampling vector [uH, uW, uP] as the generator input, and the basic scenario set is generated based on GAN; S14, taking the feature vector E as the conditional label, combining the noise vector input CGAN generates the basic extreme scenario set Ωraw; the Wasserstein distance is used as the similarity measure, and the extreme scenario subset ΩE is extracted from Ωraw by K-medoids clustering; S15, the historical output data, the basic scenario set Ω and the extreme scenario subset ΩE are fused, and the comprehensive scenario set is constructed according to the seasonal characteristics.

[0021] In S11, the corresponding runoff→hydroelectric output, wind speed→wind power output, and solar intensity→photovoltaic output conversion relationship is as follows: Runoff→hydroelectric output ; In the formula: K is the comprehensive output coefficient (usually 8.5), Q is the power flow (m3 / s), and H is the net water head of upstream and downstream (m).

[0022] Wind speed→wind power output ; ; Where: vhub is the hub height wind speed (unit: m / s), which is converted from the weather station wind speed, vin is the cut-in wind speed (usually 3 m / s), vout is the cut-out wind speed (usually 25 m / s), vr is the rated wind speed (usually 11 m / s), Pmax is the installed capacity of wind power (unit: MW), Zhub is the height of the wind turbine hub (unit: m), Z is the height of the weather station anemometer (unit: m), and m is the stability coefficient (taking the empirical value 0.18).

[0023] Irradiance→Photovoltaic output Where: Pmax is the installed capacity of photovoltaic power station (unit: MW), Rrad is the solar radiation intensity (unit: W / m 2 ), Rstc is the radiation intensity under standard test conditions, Tair is the air temperature at the weather station (unit: ℃), Tstc is the air temperature under standard test conditions, Tnom is the normal operating temperature of the panel, and ap is the power temperature coefficient, i.e. the power change rate per ℃ change.

[0024] S13 is implemented by a generative adversarial network (GAN) generation module: The composition of GAN includes a generator G (gennerator) and a discriminator D (disciminator). D is to help G better learn the distribution of the original data. The input is data, and the output is the probability of judging that the data is original data. The working principle of GAN is: the generator G tries to fit the true data distribution as much as possible to deceive the discriminator D; the discriminator D inputs data as real samples and generator generated samples, and tries to distinguish whether it is a real sample or a generated sample. The two update and iterate each other to reach Nash equilibrium.

[0025] Take the noise vector z ~ N(0, 1) + Copula sampling vector =[uH,uW,uI] as the input of the generator G, and take the measured data and generated data as the input of the discriminator D; The GAN adversarial process can be represented by the following formula: Where: x is a sample of the measured historical data distribution pdata; and z is a noise sample from a random noise distribution pz.

[0026] S14 is implemented by a conditional generative adversarial network (CGAN): Take the noise joint extreme event features as input, and take the extreme event features As a conditional label constraint generation process, the conditional generator GC learns the extreme event distribution; the output is an extreme scenario set Ωraw, the discriminator D distinguishes between real extreme scenarios and generated extreme scenarios, and the generation quality is improved through adversarial training; The Wasserstein distance is used as the similarity measure, and the Ωraw is compressed into an extreme scenario subset ΩE through K-medoids clustering.

[0027] In S15, hydrological feature value calculation is performed on the historical data and the basic scenario set and the extreme scenario set, including monthly average flow, monthly average output fluctuation rate, and output peak-valley difference; scenarios with feature values within a preset flood season threshold range are classified into a flood season comprehensive scenario set, and scenarios with feature values within a preset dry season threshold range are classified into a dry season comprehensive scenario set.

[0028] In S2, a multi-objective robust optimization model considering extreme risk penalty is constructed by introducing a risk penalty term for the extreme scenario set while maximizing online power generation, minimizing maximum curtailment, minimizing maximum power shortage, minimizing period average power shortage, and minimizing period average curtailment, and a cuckoo algorithm is used for solution.

[0029] Maximizing total online power generation is specifically represented as: ; ; Minimizing maximum curtailment is specifically represented as: ; ; Minimizing maximum power shortage is specifically represented as: ; ; Minimizing period average power shortage is specifically represented as: ; Minimizing period average curtailment is specifically represented as: ; Ω is a comprehensive scenario set, , are the power shortage and curtailment of period t in scenario w, respectively, is the online power generation, are the hydropower, wind power, and photovoltaic output of the power station in scenario w at period t, respectively, is the maximum load channel capacity.

[0030] S3, taking the integrated scenario set as input, obtaining the optimization result through a multi-objective robust optimization model considering extreme risk penalty.

[0031] The risk penalty term introduced for the extreme scenario set is specifically represented as: ; In the formula, R is the total penalty of the system extreme risk, and are the penalty coefficients of power shortage and power curtailment, respectively.

[0032] S4, based on the optimization result, a linear scheduling function is constructed as a robust scheduling rule with available energy as the independent variable and period discharge as the dependent variable. The optimization result is the corresponding optimized reservoir water level process, water, wind and light output process, and the six variables obtained through corresponding calculation are available water, available energy, reservoir water level, reservoir storage, discharge and system output.

[0033] In S4, the Pearson correlation coefficients of the six variables are calculated, and the independent variables highly correlated with the decision variables are selected according to the preset threshold. Since the available energy AE can best reflect the input conditions of the three resources of water, wind and light in the system, the linear scheduling function is finally constructed with the system available energy as the independent variable and the period discharge as the decision variable.

[0034] The basic form of the linear scheduling function is as follows: ; In the formula, k and K are the number and total number of the scheduling function; and are the decision variables of the scheduling function; and are the parameters of the scheduling function.

[0035] The scheme uses scenario generation technology to generate extreme scenarios, proposes a cascade water-wind-light medium and long term robust scheduling rule combined with extreme scenario generation, and derives the linear scheduling function of the water-wind-light complementary system, providing a new idea for improving the safety and reliability of large-scale wind and light grid connection.

[0036] Example 2. Appendix Figures 2-7 is the scheduling rule result of the water-wind-light complementary system in the lower reaches of the Jinsha River obtained by the scheduling rule extraction method proposed by the application. The results show that: ① the fitting results of February and March are acceptable, R 2 ≈0.6; ② the fitting results of the remaining months are good, R 2 all>0.85. The linear scheduling function can be constructed with the period available energy AE as the independent variable and the period discharge WR as the decision variable, that is, .

[0037] The research method proposes a robust long-term scheduling rule for cascade hydropower, wind power, and solar power generated under extreme scenarios. The objective function is to maximize the total on-grid power generation, minimize the maximum curtailment, minimize the maximum power shortage, minimize the average power shortage during the time period, and minimize the average curtailment during the time period of cascade hydropower stations under extreme weather conditions. A robust optimization model is constructed, and robust scheduling rules are extracted to enhance the reliability of power supply under extreme weather conditions.

[0038] The above-described specific embodiments are preferred embodiments of the present invention and are not intended to limit the specific scope of the present invention. The scope of the present invention includes, but is not limited to, these specific embodiments. All equivalent changes made in accordance with the shape, structure, and method of the present invention are within the protection scope of the present invention.

Claims

1. A robust long-term scheduling method for cascaded water, wind, and solar power systems generated in extreme scenarios, characterized in that... Includes the following steps: S1. Construct extreme scenario sets for hydropower, wind power, and photovoltaic power output based on the Copula-GAN hybrid method, and obtain a comprehensive scenario set by combining historical measured data; S2. The optimization objectives are to maximize the amount of electricity generated online, minimize the maximum amount of electricity wasted, minimize the maximum amount of electricity shortage, minimize the average amount of electricity shortage during a given period, and minimize the average amount of electricity wasted during a given period. While considering the constraints, a risk penalty term is introduced for the extreme scenario set. Then, a multi-objective robust optimization model considering extreme risk penalties is constructed and solved using the Cuckoo algorithm. S3. Using the comprehensive scenario set as input, the optimization results are obtained through a multi-objective robust optimization model that considers extreme risk penalties; S4. Based on the optimization results, a linear scheduling function is constructed as a robust scheduling rule, with available energy as the independent variable and the outflow during each time period as the dependent variable.

2. The long-term robust scheduling method for cascade water, wind, and solar power generation combined with extreme scenario generation as described in claim 1, characterized in that, The specific steps of S1 are as follows: S11. Perform wavelet transform to denoise the original runoff, wind speed, and light intensity data, and standardize the denoised data to convert them into hydropower output, wind power output, and photovoltaic output according to the corresponding conversion relationships. S12. Define the threshold for extreme events based on historical output quantiles, mark the time periods of extreme events, and construct feature vectors; S13. Fit the marginal probability distributions of hydropower output, wind power output, and photovoltaic output respectively, and obtain the joint probability distribution using a vine-copula structure; use the noise vector and Copula sampling vector as the generator input to generate a basic scene set based on GAN. S14. Using feature vectors as conditional labels, and combining them with noise vectors, input CGAN to generate a basic extreme scene set; using Wasserstein distance as a similarity metric, extract a subset of extreme scenes from the basic extreme scene set through K-medoids clustering; S15. Integrate historical output data, basic scenario sets, and extreme scenario subsets to construct a comprehensive scenario set based on seasonal characteristics.

3. The long-term robust scheduling method for cascade water, wind, and solar power generation combined with extreme scenario generation as described in claim 2, characterized in that, S13 is implemented through a Generative Adversarial Network (GAN) generation module: GANs consist of a generator and a discriminator. The discriminator helps the generator learn the distribution of the original data better. The input is the data, and the output is the probability that the data is the original data.

4. The long-term robust scheduling method for cascade water, wind, and solar power generation combined with extreme scenario generation as described in claim 2, characterized in that, S14 is implemented through a conditional generative adversarial network (CGAN): taking noise and extreme event features as input, and using the extreme event features as conditional labels to constrain the generation process, the conditional generator learns the distribution of extreme events. The output is a set of extreme scenarios. The discriminator distinguishes between real extreme scenarios and generated extreme scenarios, and the generation quality is improved through adversarial training.

5. A robust long-term scheduling method for cascaded water, wind, and solar power generation combined with extreme scenario generation, as described in claim 2, is characterized in that... In S15: hydrological characteristic values ​​are calculated for historical data and the basic scenario set and extreme scenario set, including monthly average flow, monthly average power output fluctuation rate, and power output peak-valley difference; Scenarios with feature values ​​within the preset flood season threshold range are classified into the flood season comprehensive scenario set, and scenarios with feature values ​​within the preset dry season threshold range are classified into the dry season comprehensive scenario set.

6. The long-term robust scheduling method for cascade water, wind, and solar power generation combined with extreme scenario generation as described in claim 1, characterized in that, The maximum total power consumption for internet access is specifically expressed as follows: ； ； The minimum maximum amount of power wasted is specifically expressed as: ； ； Minimizing the maximum power shortage is specifically expressed as: ； ； The minimum average power shortage over a given period is specifically expressed as follows: ； The minimum average amount of power wasted during the period is specifically expressed as follows: ； In the formula: , , These represent the hydropower, wind power, and photovoltaic power outputs of the power station during time period t. For the scene The power shortage during the next t period.

7. The long-term robust scheduling method for cascade water, wind, and solar power generation combined with extreme scenario generation as described in claim 1, characterized in that, The risk penalty term introduced for the extreme scenario set is specifically expressed as follows: ； In the formula, R represents the total penalty for extreme risks of the system. and These are the penalty coefficients for power shortages and power abandonment, respectively.

8. A robust long-term scheduling method for cascade water, wind, and solar power generation combined with extreme scenario generation, as described in claim 1, is characterized in that... The optimization results are the optimized reservoir water level process, water, wind and solar power output process, and six variables are obtained through corresponding calculations: available water, available energy, reservoir water level, reservoir storage, outflow and system output.

9. A robust long-term scheduling method for cascade water, wind, and solar power generation combined with extreme scenario generation, as described in claim 1, is characterized in that... In step S4, the Pearson correlation coefficients of the six variables are calculated, and independent variables that are highly correlated with the decision variables are selected based on preset thresholds.

10. A robust long-term scheduling method for cascade water, wind, and solar power generation combined with extreme scenario generation, as described in claim 9, is characterized in that... The basic form of the linear scheduling function is as follows: ； In the formula, k and K are the number and total number of scheduling functions, respectively; and These are the decision variables of the scheduling function; and These are the parameters of the scheduling function.

Images