Optimized regulation and control method based on operation scene model considering uncertain source load

Abstract

The invention relates to an optimized regulation and control method based on a model considering an uncertain source load operation scene, and the method comprises the steps: solving a predicted value of conventional load power through a load power prediction method, such as a neural network method, and solving predicted values of wind power output and photovoltaic output through a distributed power output prediction method, such as a data statistics prediction method; a prediction error distribution function of wind power output and load power is combined, a source load operation scene generation model considering prediction errors is established, a large number of original source load operation scenes are generated, an improved hierarchical K-means clustering algorithm is used for extracting the original source load operation scenes, and a source load uncertainty model based on the multi-scene technology is established; compared with the prior art, the method has the advantages that the calculation complexity is simplified while the uncertainty influence caused by prediction errors is fully considered, the dependence of a traditional clustering algorithm on the number of target scenes is avoided, and the strong uncertainty of the source side and the load side of the power distribution network is accurately described.

Description

Technical Field

[0001] The present invention belongs to the technical field of power system dispatching automation, and specifically relates to an optimization control method based on a source-load operation scenario model taking uncertainty into consideration. Background Art

[0002] With the construction of new power systems, the penetration rate of distributed generation (DG), flexible loads, energy storage and other resources represented by wind power and photovoltaics will continue to increase, resulting in the randomness, diversity and strong uncertainty of the source-load operation of the power system. At the same time, due to the significant uncertainty factors brought by distributed generation and flexible loads represented by wind power and photovoltaics, the real-time balance of power in the distribution network is seriously affected. When studying the power balance method, it is of great significance to accurately characterize the uncertainty of source and load. Therefore, in the process of building new power systems in the future, it is necessary to accurately model and predict the influencing factors such as flexible loads and distributed new energy. The source-load operation scenario generation and extraction method considering uncertainty models and analyzes the uncertainty of source-load prediction, and describes the random process of source-load operation through multi-scenario technology, in which the source side is the uncertainty of wind and solar output, and the load side is the uncertainty of load demand. Uncertainty modeling consists of two parts: scenario generation and reduction. It is an effective way to solve the problem of optimal planning and operation of power systems containing renewable energy. Summary of the invention

[0003] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide an optimization control method based on a source-load operation scenario model taking into account uncertainty.

[0004] The purpose of the present invention is to overcome the existing problems in the current construction process of new power systems, such as the significant uncertainties brought about by the flexible load of the distribution network and the large-scale use of DG, which makes it difficult to characterize the uncertainty of the source and load of the distribution network. An optimization and control method based on a source-load operation scenario model considering uncertainty is provided. This method models and analyzes the uncertainty of source-load prediction, and describes the random process of source-load operation through multi-scenario technology, so as to accurately characterize the uncertainty of source and load in the distribution network.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] The present invention provides an optimization control method based on a source-load operation scenario model taking into account uncertainty, comprising the following steps:

[0007] Step S1: construct an initial sampling scenario set for wind and solar power output and load forecast;

[0008] Step S2: combining the forecast error distribution of wind and solar power output and the forecast error distribution of load, and calculating the corresponding errors of the initial sampling scenario set of wind and solar power output and load forecast by moment estimation method;

[0009] Step S3: According to the wind / solar output and load forecast initial sampling scene set in step 1 and the error value calculated in step 2, the actual initial sampling scene set of wind / solar output and load is calculated;

[0010] Step S4: using the Latin hypercube sampling method to sample the actual initial sampling scenario set of wind and solar power output and load, to obtain a wind power output sampling scenario set, a photovoltaic output sampling scenario set and a load sampling scenario set;

[0011] Step S5: Use an improved hierarchical K-means clustering algorithm to extract and reduce the wind power output sampling scenario set, the photovoltaic output sampling scenario set and the load sampling scenario set, respectively, and construct a source-load uncertainty model of multi-scenario technology based on the extracted and reduced scenario sets, and optimize and regulate the power grid system based on the constructed source-load uncertainty model.

[0012] Furthermore, the step S1 specifically includes: combining the historical data of distributed power sources and loads in the past, calculating the predicted value of conventional load power by a neural network method, calculating the predicted values ​​of wind power output and photovoltaic output by a data statistical prediction method, and constructing an initial sampling scenario set for wind and photovoltaic output and load prediction.

[0013] Furthermore, the prediction errors of wind and solar power output all obey the TLS distribution, and the distribution formula is:

[0014]

[0015] Among them, v t is the shape parameter of the wind-solar TLS distribution during period t, σ DG,t is the variance of the wind and solar TLS distribution during period t, μ t is the location parameter of the wind-solar TLS distribution during period t, and Γ is the gamma function.

[0016] Furthermore, the load distribution probability density function is:

[0017]

[0018] in, is the predicted value of load at time t, is the mean of all load forecast values ​​at time t, σ L,t is the variance of the conventional load during period t.

[0019] Furthermore, the calculation formula of step S3 is:

[0020]

[0021] in, are the predicted actual values ​​of wind power output, photovoltaic output and conventional load, are the predicted values ​​of wind power output, photovoltaic output and conventional load, Δξ w,t , Δξ pv,t , Δξ L,t They are the prediction errors of wind power output, photovoltaic output and conventional load respectively.

[0022] Furthermore, the actual initial sampling scenario set of wind and solar power output and load includes predicted actual values ​​of multiple daily wind power outputs, daily photovoltaic outputs and daily conventional loads.

[0023] Furthermore, in step S4, the Latin hypercube sampling method is used to sample the actual initial sampling scene set of wind and solar power output and load, which specifically includes the following steps:

[0024] Step S41: Integrate the load distribution probability density function to obtain the cumulative distribution function Y k =F m (P Q ), the value range of the distribution function [0,1] is divided into M sub-intervals, each of which has the same probability and the same length, where the daily load forecast actual value scenario set P = [P1, P2, ..., P Q ];

[0025] Step S42: Draw a random value Y in each sub-interval according to uniform distribution k , a total of M times, and then the inverse function F of CDF -1 m (Y k ) to obtain the actual value of the corresponding daily load forecast P Q ;

[0026] Step S43: Randomly extract the actual values ​​of the daily load forecast selected in step S42 from each interval and form a load sampling scenario set P = [P1, P2, ..., P M ].

[0027] Furthermore, the construction method of the wind power output sampling scenario set and the photovoltaic output sampling scenario set is the same as that of the load sampling scenario set.

[0028] Furthermore, in step S5, an improved hierarchical K-means clustering algorithm is used to extract and reduce the wind power output sampling scene set, the photovoltaic output sampling scene set and the load sampling scene set, respectively, which specifically includes the following steps:

[0029] Step S51: The two scenes with the largest distance in the original scene set are regarded as the initial cluster centers. The distance calculation formula is:

[0030]

[0031] Among them, P s is the initial cluster center of the sth class, P j is the initial cluster center of the jth class, p s,t 、p j,t is the load value at each moment of these two scenarios;

[0032] Step S52: In the remaining scene set, the scene with the largest product of distances to the two initial cluster centers is selected as the third cluster center, and so on until K cluster centers are selected, where K is the initial number of clusters;

[0033] Step S53: assign the remaining scenes to the nearest cluster center, set the number of iterations λ=1, and obtain the λth clustering measure function value J(λ), the calculation formula is:

[0034]

[0035] Among them, M s is the number of scenes in the sth category, is the jth data vector in the sth category; C s is the cluster center of the sth class;

[0036] Step S54: Perform the next level of clustering: select the class with the largest radius among all clusters. The class radius r is calculated as:

[0037] r s =max||P j -C s ||,j=1,2,...,M s

[0038] Among them, r s is the radius of the cluster center of the sth cluster, P j is the cluster center of the jth class;

[0039] Secondly, the two scenes with the largest distance in the cluster are selected as new cluster centers. Finally, clustering is performed again using the cluster centers, and λ=λ+1 is set to obtain the λ+1th J(λ+1) value.

[0040] Step S55: define δ=(J (λ) -J (λ+1) ) / J (λ), determine whether δ is greater than δ0. If so, execute step S54. If less than or equal to, output the occurrence probability and cluster center of each typical scene, and use the cluster center as the typical day scene set.

[0041] Furthermore, the δ0 is a preset threshold value, which is set according to the clustering measure function value change curve.

[0042] Compared with the prior art, the present invention has the following advantages:

[0043] (1) Compared with Monte Carlo simulation and other methods, the Latin Hypercube Sampling (LHS) technique used in the present invention has a small number of samples and avoids duplication, and can still reflect the theoretical distribution of random variables, ensuring that all sampling areas are evenly covered by sampling points. The results will be more representative.

[0044] (2) The improved hierarchical K-means clustering algorithm proposed in the present invention first uses the maximum distance method to obtain the initial cluster value, and then obtains the optimal cluster center based on the hierarchical K-means to avoid the dependence of the traditional clustering algorithm on the number of target scenes.

[0045] (3) The source-load operation scenario generation model considering prediction errors proposed in this invention can effectively reduce the uncertainty on both sides of the source and load of the distribution network, and realize the accurate extraction of the source-load day-ahead operation scenario. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a flow chart of a method for generating and extracting source-load operation scenarios taking uncertainty into consideration in an embodiment;

[0047] Figure 2 It is a source load prediction curve diagram of the embodiment in which the day-ahead prediction is performed by point prediction at intervals of 1 hour;

[0048] Figure 3 A set of source-load prediction operation scenarios of the traditional K-means clustering algorithm in the embodiment;

[0049] Figure 4 It is a source-load prediction operation scenario set of the improved hierarchical K-means clustering algorithm in the embodiment; DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are 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 creative work should fall within the scope of protection of the present invention.

[0051] Embodiment 1:

[0052] This embodiment provides an optimization control method based on a source-load operation scenario model taking into account uncertainty, including the following steps:

[0053] Step S1: construct an initial sampling scenario set for wind and solar power output and load forecast;

[0054] Step S2: combining the forecast error distribution of wind and solar power output and the forecast error distribution of load, and calculating the corresponding errors of the initial sampling scenario set of wind and solar power output and load forecast by moment estimation method;

[0055] Step S3: According to the wind / solar output and load forecast initial sampling scene set in step 1 and the error value calculated in step 2, the actual initial sampling scene set of wind / solar output and load is calculated;

[0056] Step S4: using the Latin hypercube sampling method to sample the actual initial sampling scenario set of wind and solar power output and load, to obtain a wind power output sampling scenario set, a photovoltaic output sampling scenario set and a load sampling scenario set;

[0057] Step S5: Use an improved hierarchical K-means clustering algorithm to extract and reduce the wind power output sampling scenario set, the photovoltaic output sampling scenario set and the load sampling scenario set, respectively, and construct a source-load uncertainty model of multi-scenario technology based on the extracted and reduced scenario sets, and optimize and regulate the power grid system based on the constructed source-load uncertainty model.

[0058] Furthermore, the step S1 specifically includes: combining the historical data of distributed power sources and loads in the past, calculating the predicted value of conventional load power by a neural network method, calculating the predicted values ​​of wind power output and photovoltaic output by a data statistical prediction method, and constructing an initial sampling scenario set for wind and photovoltaic output and load prediction.

[0059] Furthermore, the prediction errors of wind and solar power output all obey the TLS distribution, and the distribution formula is:

[0060]

[0061] Among them, v t is the shape parameter of the wind-solar TLS distribution during period t, σ DG,t is the variance of the wind and solar TLS distribution during period t, μ t is the location parameter of the wind-solar TLS distribution during period t, and Γ is the gamma function.

[0062] Furthermore, the load distribution probability density function is:

[0063]

[0064] in, is the predicted value of load at time t, is the mean of all load forecast values ​​at time t, σ L,t is the variance of the conventional load during period t.

[0065] Furthermore, the calculation formula of step S3 is:

[0066]

[0067] in, are the predicted actual values ​​of wind power output, photovoltaic output and conventional load, are the predicted values ​​of wind power output, photovoltaic output and conventional load, Δξ w,t , Δξ pv,t , Δξ L,t They are the prediction errors of wind power output, photovoltaic output and conventional load respectively.

[0068] Furthermore, the actual initial sampling scenario set of wind and solar power output and load includes predicted actual values ​​of multiple daily wind power outputs, daily photovoltaic outputs and daily conventional loads.

[0069] Furthermore, in step S4, the Latin hypercube sampling method is used to sample the actual initial sampling scene set of wind and solar power output and load, which specifically includes the following steps:

[0070] Step S41: Integrate the load distribution probability density function to obtain the cumulative distribution function Y k =F m (P Q ), the value range of the distribution function [0,1] is divided into M sub-intervals, each of which has the same probability and the same length, where the daily load forecast actual value scenario set P = [P1, P2, ..., P Q ];

[0071] Step S42: Draw a random value Y in each sub-interval according to uniform distribution k , a total of M times, and then the inverse function F of CDF -1 m (Y k ) to obtain the actual value of the corresponding daily load forecast P Q ;

[0072] Step S43: Randomly extract the actual values ​​of the daily load forecast selected in step S42 from each interval and form a load sampling scenario set P = [P1, P2, ..., P M ].

[0073] Furthermore, the construction method of the wind power output sampling scenario set and the photovoltaic output sampling scenario set is the same as that of the load sampling scenario set.

[0074] Furthermore, in step S5, an improved hierarchical K-means clustering algorithm is used to extract and reduce the wind power output sampling scene set, the photovoltaic output sampling scene set and the load sampling scene set, respectively, which specifically includes the following steps:

[0075] Step S51: The two scenes with the largest distance in the original scene set are regarded as the initial cluster centers. The distance calculation formula is:

[0076]

[0077] Among them, P s is the initial cluster center of the sth class, P j is the initial cluster center of the jth class, p S,t 、p j,t is the load value at each moment of these two scenarios;

[0078] Step S52: In the remaining scene set, the scene with the largest product of distances to the two initial cluster centers is selected as the third cluster center, and so on until K cluster centers are selected, where K is the initial number of clusters;

[0079] Step S53: assign the remaining scenes to the nearest cluster center, set the number of iterations λ = 1, and obtain the λth clustering measure function value J ( λ ) , the calculation formula is:

[0080]

[0081] Among them, M s is the number of scenes in the sth category, is the jth data vector in the sth category; C s is the cluster center of the sth class;

[0082] Step S54: Perform the next level of clustering: select the class with the largest radius among all clusters. The class radius r is calculated as:

[0083] r s =max||P j -C s ||,j=1,2,...,M s

[0084] Among them, r s is the radius of the cluster center of the sth cluster, P j is the cluster center of the jth class;

[0085] Secondly, the two scenes with the largest distance in the cluster are selected as new cluster centers. Finally, clustering is performed again using the cluster centers, and λ=λ+1 is set to obtain the λ+1th J(λ+1) value.

[0086] Step S55: define δ=(H (λ) -J (λ+1) ) / J (λ) , determine whether δ is greater than δ0. If so, execute step S54. If less than or equal to, output the occurrence probability and cluster center of each typical scene, and use the cluster center as the typical day scene set.

[0087] Furthermore, the δ0 is a preset threshold value, which is set according to the clustering measure function value change curve.

[0088] Embodiment 2:

[0089] The parts not mentioned in this embodiment are the same as those in Embodiment 1.

[0090] This embodiment relates to an optimization and control method based on a source-load operation scenario model that takes into account uncertainty. The method first obtains the predicted value of conventional load power through load power prediction methods such as neural network method, time series method, regression analysis method, etc., and then obtains the predicted value of wind power output and photovoltaic output through distributed power output prediction methods such as data statistical prediction method and artificial intelligence prediction method. On this basis, combined with the prediction error distribution function of wind power output and load power, a source-load operation scenario generation model that takes into account prediction errors is established and a large number of original source-load operation scenarios are generated. Finally, an improved hierarchical K-means clustering algorithm is used to extract the original source-load operation scenario, so that the uncertainty characteristics are effectively described using a few typical scenarios. A source-load uncertainty model based on multi-scenario technology is established, and the power grid system is optimized and controlled according to the source-load uncertainty model.

[0091] In order to solve the shortcomings of simple random sampling, which is inefficient and has poor accuracy, Latin Hypercube Sampling (LHS) is widely used. Compared with Monte Carlo simulation and other methods, LHS has a small number of samples and avoids duplication, and can still reflect the theoretical distribution of random variables, ensuring that all sampling areas are evenly covered by sampling points. The results will be more representative. In addition, in order to reduce the huge number of original scenes generated by LHS and the complicated calculations, K-means clustering algorithm technology is widely used, which can group data points into several clusters, so that the data points have high similarity within the same cluster and low similarity between different clusters, thereby achieving the purpose of dividing the data set.

[0092] like Figure 1 As shown, the method for constructing a source-load operation scenario model considering uncertainty in this embodiment specifically includes the following steps:

[0093] Step 1: Obtain the predicted value of conventional load power through load power prediction methods such as neural network method, time series method, regression analysis method, etc., and then obtain the predicted value of wind power output and photovoltaic output through distributed power output prediction methods such as data statistical prediction method, artificial intelligence prediction method, etc.

[0094] Step 2: Consider the prediction errors of wind power output, photovoltaic output and load energy consumption. According to current research results, the prediction errors of wind and photovoltaic output all obey the TLS (t location-scale) distribution, while the prediction error of load obeys a normal distribution with a mean of 0. The mean, variance and median are obtained by the moment estimation method, and then the prediction error is obtained by using the existing sampling method.

[0095] The moment estimation method takes the wind and solar output scenario as an example, assuming that the wind and solar output prediction value scenario set is Q = [Q1, Q2, …, Q n ], the mean μ, variance d and median k of the prediction error can be expressed as:

[0096]

[0097] The TLS distribution of the wind and solar prediction error Δξ can be expressed as:

[0098]

[0099] Among them, v t is the shape parameter of the wind-solar TLS distribution during period t, σ DG,t is the variance of the wind and solar TLS distribution during period t, μ t is the location parameter of the wind-solar TLS distribution during period t, and Γ is the gamma function.

[0100] The load distribution probability density function considering the prediction deviation is:

[0101]

[0102] in, is the predicted value of load at time t, is the mean of all load forecast values ​​at time t, σ L,t is the variance of the conventional load during period t.

[0103] Step 3: Based on the predicted value obtained in step 1 and the prediction error considered in step 2, the source-load operation scenario generation model considering the prediction error is used to transform the source-load prediction results with uncertainty errors into several deterministic scenario sets. The modeling formulas for wind power output, photovoltaic output and conventional load power are as follows. The actual predicted values ​​of wind power output and photovoltaic output are expressed as the sum of the predicted value and the random prediction error value:

[0104]

[0105] in, are the predicted actual values ​​of wind power output, photovoltaic output and conventional load, are the predicted values ​​of wind power output, photovoltaic output and conventional load, Δξ w,t , Δξ pv,t , Δξ L,t They are the prediction errors of wind power output, photovoltaic output and conventional load respectively.

[0106] The predicted value can be obtained through the existing DG and load forecasting methods with higher accuracy.

[0107] Step 4. According to the source-load uncertainty model generated in step 3, a deterministic output operation scenario can be generated by sampling to characterize its randomness. Therefore, the present invention selects Latin hypercube sampling (LHS) to simulate the load and wind and solar output generation. Compared with methods such as Monte Carlo simulation, LHS can ensure that the sampling points include all sampling areas based on stratified sampling, and the results will be more representative. Taking the load operation scenario as an example, the steps are explained. The steps of Latin hypercube sampling are as follows:

[0108] a) Set the sampling times to M, and assume that the actual value scenario set of daily load forecast for Q days is P = [P1, P2, …, PQ].

[0109] b) Integrate the load distribution probability density function to obtain the cumulative distribution function, and set the value range of the distribution function

[0110] [0,1] is divided into M equal sub-intervals.

[0111] c) Draw a random value in each interval according to uniform distribution, draw M times in total, and then use the inverse function of CDF to obtain the corresponding daily load forecast actual value PQ.

[0112] d) Randomly extract the actual values ​​of the daily load forecast selected in step (c) from each interval and combine them into the original scenario set P = [P1, P2, ..., PM] of the load operation.

[0113] The sampling values ​​of the LHS sampling random variable are:

[0114]

[0115] Where: is a random variable X m The hth sampling value of m is the distribution function; U h =(U+h-1) / H, which represents the random number of the hth subinterval after dividing the interval [0,1] into equal parts, U∈[0,1].

[0116] Step 5: Based on the large number of original source-load operation scenarios generated in step 4, the improved hierarchical K-means clustering algorithm is used to extract and reduce the original source-load operation scenarios. First, the maximum distance method is used to obtain the initial cluster value, and then the optimal cluster center is obtained based on the hierarchical K-means algorithm to obtain the occurrence probability and output curve of each typical scenario. Taking the load operation scenario as an example, the steps are explained as follows:

[0117] a) Assume that a total of M days of effective load operation data are obtained, and the original scenario set of load operation is P = [P1, P2, ..., P M ], the sth scenario load data vector is Ps = [p s,1 ,p s,2 ,…,p s,T ], the initial number of clusters is K1.

[0118] b) Select K1 initial cluster centers by the maximum distance method. First, the two scenes with the largest distance in the original scene set are regarded as the initial cluster centers. The distance d is calculated as:

[0119]

[0120] Among them, P s is the initial cluster center of the sth class, P j is the initial cluster center of the jth class, p s,t 、p j,t is the load value at each moment of these two scenarios;

[0121] Secondly, among the remaining M-2 running scenarios, the scenario with the largest product of distances to the two initial cluster centers is selected as the third cluster center; and so on until K1 cluster centers are selected.

[0122] c) Implement K-means clustering, that is, all scenes belong to the nearest cluster center. Suppose the number of iterations λ=1, and obtain the λth clustering measure function value J(λ), which is calculated by the following formula.

[0123]

[0124] Among them, M s is the number of scenes in the sth category, is the jth data vector in the sth category; C s is the cluster center of the sth class;

[0125] Where: M s is the number of scenes in the sth category, is the jth data vector in the sth category; C s is the cluster center of the sth class.

[0126] Perform the next level of clustering. First, select the class with the largest radius among all clusters. The calculation formula of the class radius r is:

[0127] r s =max||P j -C s ||,j=1,2,...,M s

[0128] Among them, r s is the radius of the cluster center of the sth cluster, P j is the cluster center of the jth class; secondly, the two scenes with the largest distance in this class are selected as new cluster centers; finally, clustering is performed again using the cluster centers, and λ=λ+1 is set to obtain the J(λ+1) value of the λ+1th time.

[0129] d) Define δ=(J (λ) -J (λ+1) ) / J (λ) , when δ>δ0, return to step 4, where δ0 is a given threshold, which can be set according to the clustering measure function value change curve; otherwise, output the clustering result.

[0130] e) Obtain the occurrence probability and cluster center of each typical scenario during load reduction operation;

[0131] Step 6: After scene extraction and reduction in step 5, a source-load uncertainty model based on multi-scenario technology is established.

[0132] The source-load operation scenario generation and extraction method considering uncertainty proposed in the present invention can effectively reduce the uncertainty on both sides of the source and load of the distribution network, and realize the accurate extraction of the source-load operation scenario before the day. Figure 3 and Figure 4 By comparison, the operating scenarios obtained by the traditional K-means clustering algorithm are relatively concentrated and fail to effectively characterize the source-load operating conditions at each level. The results obtained by the method proposed in the present invention have good stratification characteristics, relatively evenly covering all levels from small to large, effectively retaining some extreme scenarios, and the differences between different scenarios also better reflect the accuracy of the prediction and effectively reduce the impact of the prediction error. A typical source-load operating scenario that effectively takes into account the impact of uncertainty is obtained, which can then accurately evaluate the operating level of the regional distribution network.

[0133] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0134] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. An optimization control method based on a source-load operation scenario model considering uncertainty, characterized in that: The following steps are involved: Step S1: construct an initial sampling scenario set for wind and solar power output and load forecast; Step S2: combining the forecast error distribution of wind and solar power output and the forecast error distribution of load, and calculating the corresponding errors of the initial sampling scenario set of wind and solar power output and load forecast by moment estimation method; Step S3: According to the wind / solar output and load forecast initial sampling scene set in step 1 and the error value calculated in step 2, the actual initial sampling scene set of wind / solar output and load is calculated; Step S4: using the Latin hypercube sampling method to sample the actual initial sampling scenario set of wind and solar power output and load, to obtain a wind power output sampling scenario set, a photovoltaic output sampling scenario set and a load sampling scenario set; Step S5: Use an improved hierarchical K-means clustering algorithm to extract and reduce the wind power output sampling scenario set, the photovoltaic output sampling scenario set and the load sampling scenario set, respectively, and construct a source-load uncertainty model of multi-scenario technology based on the extracted and reduced scenario sets, and optimize and regulate the power grid system based on the constructed source-load uncertainty model.

2. According to claim 1, an optimization control method based on a source-load operation scenario model taking into account uncertainty is characterized in that: The step S1 specifically includes: combining the historical data of distributed power sources and loads in the past, calculating the predicted value of conventional load power by a neural network method, calculating the predicted values ​​of wind power output and photovoltaic output by a data statistical prediction method, and constructing an initial sampling scenario set for wind and photovoltaic output and load prediction.

3. The optimization control method based on the uncertain source-load operation scenario model according to claim 1 is characterized in that: The prediction errors of wind and solar power output all obey the TLS distribution, and the distribution formula is: Among them, v t is the shape parameter of the wind-solar TLS distribution during period t, σ DG,t is the variance of the wind and solar TLS distribution during period t, μ t is the location parameter of the wind-solar TLS distribution during period t, and Γ is the gamma function.

4. The optimization control method based on the source-load operation scenario model considering uncertainty according to claim 1 is characterized in that: The load distribution probability density function is: in, is the predicted value of load at time t, is the mean of all load forecast values ​​at time t, σ L,t is the variance of the conventional load during period t.

5. The optimization control method based on the source-load operation scenario model considering uncertainty according to claim 1 is characterized in that: The calculation formula of step S3 is: in, are the predicted actual values ​​of wind power output, photovoltaic output and conventional load, are the predicted values ​​of wind power output, photovoltaic output and conventional load, Δξ w,t , Δξ pv,t , Δξ L,t They are the prediction errors of wind power output, photovoltaic output and conventional load respectively.

6. The optimization control method based on the uncertain source-load operation scenario model according to claim 1 is characterized in that: The actual initial sampling scenario set of wind and solar power output and load includes predicted actual values ​​of multiple daily wind power outputs, daily photovoltaic outputs and daily conventional loads.

7. The optimization control method based on the uncertain source-load operation scenario model according to claim 1 is characterized in that: In step S4, the Latin hypercube sampling method is used to sample the actual initial sampling scene set of wind and solar power output and load, which specifically includes the following steps: Step S41: Integrate the load distribution probability density function to obtain the cumulative distribution function Y k =F m (P Q ), the value range of the distribution function [0,1] is divided into M sub-intervals, each of which has the same probability and the same length, where the daily load forecast actual value scenario set P = [P1, P2, ..., P Q ]; Step S42: Draw a random value Y in each sub-interval according to uniform distribution k , a total of M times, and then the inverse function F of CDF -1 m (Y k ) to obtain the actual value of the corresponding daily load forecast P Q ; Step S43: Randomly extract the actual values ​​of the daily load forecast selected in step S42 from each interval and form a load sampling scenario set P = [P1, P2, ..., P M ].

8. The optimization control method based on the source-load operation scenario model considering uncertainty according to claim 7 is characterized in that: The construction method of the wind power output sampling scenario set and the photovoltaic output sampling scenario set is the same as that of the load sampling scenario set.

9. The optimization control method based on the uncertain source-load operation scenario model according to claim 1 is characterized in that: In step S5, an improved hierarchical K-means clustering algorithm is used to extract and reduce the wind power output sampling scene set, the photovoltaic output sampling scene set and the load sampling scene set, respectively, which specifically includes the following steps: Step S51: The two scenes with the largest distance in the original scene set are regarded as the initial cluster centers. The distance calculation formula is: Among them, P s is the initial cluster center of the sth class, P j is the initial cluster center of the jth class, p S,t 、p j,t is the load value at each moment of these two scenarios; Step S52: In the remaining scene set, the scene with the largest product of distances to the two initial cluster centers is selected as the third cluster center, and so on until K cluster centers are selected, where K is the initial number of clusters; Step S53: assign the remaining scenes to the nearest cluster center, set the number of iterations λ = 1, and obtain the λth clustering measure function value J ( λ ) , the calculation formula is: Among them, M s is the number of scenes in the sth category, is the jth data vector in the sth category; C s is the cluster center of the sth class; Step S54: Perform the next level of clustering: select the class with the largest radius among all clusters. The class radius r is calculated as: r s =max||P j -C s ||,j=1,2,...,M s Among them, r s is the radius of the cluster center of the sth cluster, P j is the cluster center of the jth class; Secondly, the two scenes with the largest distance in this class are selected as new cluster centers; finally, cluster again using the cluster centers and set λ = λ + 1 to obtain the λ + 1th J (λ+1) value; Step S55: define δ=(H (λ) -J (λ+1) ) / J (λ) , determine whether δ is greater than δ0. If so, execute step S54. If less than or equal to, output the occurrence probability and cluster center of each typical scene, and use the cluster center as the typical day scene set.

10. The optimization control method based on the source-load operation scenario model considering uncertainty according to claim 9 is characterized in that: The δ0 is a preset threshold value, which is set according to the clustering measurement function value change curve.

Images