A method and device for predicting asymmetric interval of wind power generation power

Abstract

The present invention relates to the technical field of new energy power prediction, and discloses a method and device for predicting an asymmetric interval of wind power generation power, including: constructing a double-layer optimization model for predicting an asymmetric interval of wind power generation power based on quantile regression theory; using strong duality conditions to replace the complementary relaxation conditions of KKT conditions, obtaining KKT conditions based on the lower convex optimization problem, and converting the double-layer optimization model into a single-layer non-convex optimization model; after constructing an improved velocity vector update formula and an improved position vector update formula, using an improved wind drive algorithm to solve the single-layer non-convex optimization model, obtaining the model optimal solution including the optimal quantile level offset and the upper end point quantile level and the lower end point quantile level of the asymmetric interval of wind power generation power, and obtaining the asymmetric interval of wind power generation power. The present invention obtains a prediction interval with greater sharpness, reduces the investment cost of power dispatching, and improves the reliability of the prediction of the asymmetric interval of wind power generation power.

Description

Technical Field

[0001] The present invention relates to the technical field of new energy power prediction, and in particular to a method and device for predicting an asymmetric interval of wind power generation power. Background Art

[0002] With the development of social economy, new energy represented by wind power and photovoltaics has gradually become the main direction of future energy development. Due to the randomness of wind speed and light intensity in nature, the output of wind farms and photovoltaic power stations presents strong uncertainty. With the continuous increase in the penetration rate of new energy such as wind and light in the power system, the uncertainty of their output will bring severe challenges to the safe and stable operation of the power system. Therefore, it is necessary to use probabilistic prediction methods to accurately quantify the uncertainty of renewable energy power generation prediction, so as to improve the quality of large-scale grid connection of renewable energy and improve the large-scale consumption level of renewable energy. As one of the means of probabilistic prediction, the interval prediction method can give the power value range that may be observed for renewable energy power generation according to a certain preset probability.

[0003] However, in reality, the probability distribution of the uncertainty in renewable energy power generation forecasting is generally asymmetric, and most of the existing interval forecasting methods study the symmetrical central prediction interval, that is, most of the existing renewable energy power generation interval probability forecasting methods cannot obtain a smaller prediction interval while ensuring reliability. The larger reliable prediction interval width leads to a larger cost in order to solve the uncertainty of renewable energy power generation when making power system decisions, and the existing optimization algorithms cannot be directly applied to the solution process of power generation interval forecasting. Summary of the invention

[0004] To this end, the technical problem to be solved by the present invention is to overcome the problem in the prior art that the asymmetric probability distribution of renewable energy power generation prediction uncertainty is not taken into account, resulting in the inability to obtain a smaller asymmetric interval of wind power generation power while ensuring reliability, which in turn causes the high investment cost of the power system and poor power supply stability.

[0005] In order to solve the above technical problems, the present invention provides a method for predicting asymmetric intervals of wind power generation, comprising:

[0006] Based on quantile regression theory and extreme learning machine, the positive residual matrix and negative residual matrix of wind power generation measurement value and extreme learning machine output matrix are obtained. Taking the asymmetric weighted sum of the positive residual matrix and the negative residual matrix as the minimum optimization target, a quantile estimator for predicting the asymmetric interval of wind power generation is constructed as the lower optimization model.

[0007] Taking the minimum overall width of the asymmetric interval of wind power generation predicted by the quantile estimator as the optimization goal, an upper optimization model is constructed;

[0008] After replacing the complementary relaxation condition in the Karush-Kuhn-Tucker condition with the strong duality condition to obtain the Karush-Kuhn-Tucker condition based on the lower-layer convex optimization problem, the two-layer optimization model consisting of the lower-layer optimization model and the upper-layer optimization model is converted into a single-layer non-convex optimization model;

[0009] In the initial velocity vector update formula of the wind drive algorithm, the tournament selection mechanism is used to dynamically and randomly select the coordinate origin of each particle to obtain the new origin of the particle; the particle velocity is updated based on the new origin of the particle to obtain the improved velocity vector update formula;

[0010] In the initial position vector update formula of the wind drive algorithm, a variable control parameter is introduced as a coefficient to obtain an improved position vector update formula;

[0011] The single-layer non-convex optimization model is solved by using an improved wind-driven algorithm based on an improved velocity vector update formula and an improved position vector update formula to obtain an optimal solution of the single-layer non-convex optimization model; the optimal solution includes an optimal quantile level offset and an upper endpoint quantile level and a lower endpoint quantile level of an asymmetric interval of wind power generation power;

[0012] The sum of the upper endpoint quantile level and the optimal quantile level offset is taken as the upper endpoint of the interval, and the sum of the lower endpoint quantile level and the optimal quantile level offset is taken as the lower endpoint of the interval, and the asymmetric interval of wind power generation is obtained.

[0013] Preferably, the lower layer optimization model is expressed as:

[0014] ;

[0015] in, is the number of scheduling periods; is the quantile level, and They are the upper endpoint quantile level and the lower endpoint quantile level of the asymmetric interval of wind power generation power respectively; Represents the positive residual matrix of the wind power measurement value and the extreme learning machine output matrix, Represents the negative residual matrix of the wind power generation measurement value and the extreme learning machine output matrix; is the input matrix of the extreme learning machine; is the measured value of wind power generation; is the parameter matrix to be optimized of the extreme learning machine, and They are respectively the upper endpoint parameter matrix to be optimized and the lower endpoint parameter matrix to be optimized output by the extreme learning machine; is the output matrix of the extreme learning machine, including the upper endpoint output matrix of the asymmetric interval of wind power generation and the lower endpoint output matrix .

[0016] Preferably, the upper layer optimization model is expressed as:

[0017] ;

[0018] in, is the number of scheduling periods; is the output matrix of the extreme learning machine, including the upper endpoint output matrix of the asymmetric interval of wind power generation and the lower endpoint output matrix ; is the quantile level, and They are the upper endpoint quantile level and the lower endpoint quantile level of the asymmetric interval of wind power generation power respectively; is the quantile level shift variable; is the input matrix of the extreme learning machine; is the parameter matrix to be optimized of the extreme learning machine, and are the upper endpoint parameter matrix to be optimized and the lower endpoint parameter matrix to be optimized output by the extreme learning machine respectively; the nominal coverage is , and They respectively represent the upper endpoint quantile level and the lower endpoint quantile level of the wind power generation center prediction interval.

[0019] Preferably, the single-layer non-convex optimization model is expressed as:

[0020] ;

[0021] in, is the number of scheduling periods; is the quantile level, and They are the upper endpoint quantile level and the lower endpoint quantile level of the asymmetric interval of wind power generation power respectively; is the input matrix of the extreme learning machine; is the parameter matrix to be optimized of the extreme learning machine, and are the upper endpoint parameter matrix to be optimized and the lower endpoint parameter matrix to be optimized of the extreme learning machine respectively; the nominal coverage is , and They represent the upper endpoint quantile level and the lower endpoint quantile level of the wind power generation center prediction interval respectively; is the quantile level shift variable; Represents the positive residual matrix of the wind power measurement value and the extreme learning machine output matrix, Represents the negative residual matrix of the wind power generation measurement value and the extreme learning machine output matrix; is the measured value of wind power generation; is the output matrix of the extreme learning machine, including the upper endpoint output matrix of the asymmetric interval of wind power generation and the lower endpoint output matrix ;

[0022] Among them, the Karush-Kuhn-Tucker conditions based on the underlying convex optimization problem include:

[0023] The feasible condition of the original problem is expressed as: ;

[0024] The dual feasible condition is expressed as: ;

[0025] The first-order stationary value condition is expressed as: ;

[0026] The strong duality condition is expressed as: ;

[0027] represents the hidden layer output matrix of the extreme learning machine, , , , and are all Lagrange multipliers.

[0028] Preferably, the improved velocity vector update formula is expressed as:

[0029] ;

[0030] in, Indicates The particle velocity at iteration , represents the friction coefficient, Indicates The particle velocity at iteration , is the gravitational constant, represents the new origin of the particle obtained using the tournament selection mechanism, Indicates The particle position at iteration , is the universal gas constant, Indicates temperature, Indicates the sorting number of the current particle among all particles. represents the current optimal particle, represents the randomly selected particle Random particle positions for iterations; is a constant, and its value is , To preset the initial coefficient, Represents a random number uniformly generated within a range. Represents a metric based on the standard deviation of particles in the current population.

[0031] Preferably, the metric based on the standard deviation of the current population particles is expressed as:

[0032] ;

[0033] in, Indicates Particle No. The particle position at iteration , Expressed as , Indicates the population size.

[0034] Preferably, the improved position vector update formula is expressed as:

[0035] ;

[0036] in, Indicates The particle position at iteration , Indicates The particle position at iteration , Indicates The particle velocity at iteration , represents the step length, Indicates a variable control parameter.

[0037] Preferably, after updating the particle position using the improved position vector update formula, the method further includes migrating the position of each particle using a particle position migration formula; the particle position migration formula includes:

[0038] ;

[0039] in, Indicates Iteration No. The particle Dimensional variables; Indicates Iteration No. The particle Dimensional variables; represents the current optimal particle, represents the current optimal particle Dimensional variables; represents the worst particle, The worst particle Dimensional variables; Indicates the preset error limit; Represents the negative learning rate, which is a positive increasing constant and ranges from [0.002, 0.01].

[0040] Preferably, the improved wind drive algorithm based on the improved velocity vector update formula and the improved position vector update formula is used to solve the single-layer non-convex optimization model to obtain the optimal solution of the model, including:

[0041] S1: Based on the average interval width of the asymmetric interval of wind power generation, the optimal objective function is constructed, which is expressed as:

[0042] ; Indicates the average interval width of the asymmetric interval of wind power generation, , represents the population size, Indicates the population The upper end point of the asymmetric interval of wind power generation corresponding to each particle, Indicates the population The lower endpoint of the asymmetric interval of wind power generation corresponding to each particle;

[0043] S2: Initialize the improved wind drive algorithm: preset the first continuous search times threshold for the optimal particle without change and the second continuous search times threshold for the optimal particle without improvement, and initialize the position and speed of the particle;

[0044] S3: Update the velocity and position of the particle using the improved velocity vector update formula and the improved position vector update formula;

[0045] S4: Use the particle position migration formula to migrate the position of each particle;

[0046] S5: Determine whether the number of continuous searches without change of the optimal particle reaches a first continuous search number threshold;

[0047] S6: If the number of consecutive searches without change of the optimal particle reaches the first consecutive search number threshold, a new particle is randomly selected from the optimal particles searched in the history as the current optimal particle, and the process returns to step S3;

[0048] S7: If the number of continuous searches without changes in the optimal particle does not reach the first continuous search number threshold, determine whether the number of continuous searches without improvements in the optimal particle reaches the second continuous search number threshold;

[0049] S8: If the number of continuous searches without improvement of the optimal particle does not reach the second continuous search number threshold, return to step S3;

[0050] S9: If the number of continuous searches without improvement of the optimal particle reaches the second continuous search number threshold, the iteration is terminated, and the current quantile horizontal offset is obtained as the optimal quantile horizontal offset.

[0051] The present invention also provides a device for predicting an asymmetric interval of wind power generation power, comprising:

[0052] A single-layer non-convex optimization model construction module is used to obtain the positive residual matrix and negative residual matrix of the wind power generation measurement value and the extreme learning machine output matrix based on the quantile regression theory and the extreme learning machine, and to construct a quantile estimator for predicting the asymmetric interval of wind power generation as a lower-layer optimization model with the minimization of the asymmetric sum of the positive residual matrix and the negative residual matrix as the optimization target; to construct an upper-layer optimization model with the minimization of the overall width of the asymmetric interval of wind power generation predicted by the quantile estimator as the optimization target; to replace the complementary relaxation condition in the Karush-Kuhn-Tucker condition with the strong duality condition to obtain the Karush-Kuhn-Tucker condition based on the lower-layer convex optimization problem, and then convert the double-layer optimization model composed of the lower-layer optimization model and the upper-layer optimization model into a single-layer non-convex optimization model;

[0053] The wind-driven algorithm improvement module is used to dynamically and randomly select the coordinate origin of each particle in the initial velocity vector update formula of the wind-driven algorithm by using the tournament selection mechanism to obtain the new origin of the particle; update the particle velocity based on the new origin of the particle to obtain the improved velocity vector update formula; introduce the variable control parameter as a coefficient in the initial position vector update formula of the wind-driven algorithm to obtain the improved position vector update formula;

[0054] A solution prediction module is used to solve the single-layer non-convex optimization model using an improved wind-driven algorithm based on an improved velocity vector update formula and an improved position vector update formula to obtain the optimal solution of the single-layer non-convex optimization model; the optimal solution includes an optimal quantile level offset and an upper endpoint quantile level and a lower endpoint quantile level of an asymmetric interval of wind power generation; the sum of the upper endpoint quantile level and the optimal quantile level offset is used as the upper endpoint of the interval, and the sum of the lower endpoint quantile level and the optimal quantile level offset is used as the lower endpoint of the interval to obtain the asymmetric interval of wind power generation power.

[0055] The above technical solution of the present invention has the following beneficial effects compared with the prior art:

[0056] The method for predicting an asymmetric interval of wind power generation power described in the present invention constructs a double-layer optimization model based on the asymmetry of wind power generation power prediction, and converts the double-layer optimization model into a single-layer non-convex optimization model after replacing the complementary relaxation conditions in the Karush-Kuhn-Tucker conditions with strong dual conditions, thereby simplifying the problem structure, reducing the computational burden, and improving the solution efficiency. The improved wind-driven algorithm can be directly used to solve the original double-layer optimization model, so that the optimal quantile horizontal offset obtained by the solution is more accurate, and the reliability of the solution is improved, thereby ensuring the high reliability and smaller width of the asymmetric interval of wind power generation power. The improved wind-driven algorithm of the present invention adopts a tournament selection mechanism to dynamically and randomly select the coordinate origin of each particle, thereby preventing the velocity vector and the position vector from being attracted to the absolute origin of the coordinate system when updating, thereby improving particle diversity; and utilizing an improved position vector update formula to update the particle position, thereby avoiding the convergence to zero that may be caused by using the velocity vector, avoiding stagnation, and causing the search process to be trapped in a local optimal solution, thereby improving the search speed and accuracy, making it possible to obtain the shortest reliable asymmetric prediction interval of wind power generation with greater sharpness, thereby improving the reliability of the prediction of the asymmetric interval of wind power generation, so that the power system can adjust the power supply based on the accurate asymmetric interval of wind power generation, thereby reducing the cost invested by the power system in solving the uncertainty of wind power generation when making decisions, and improving the power supply stability of the power system.

[0057] After the particle position is updated by using the improved position vector update formula, the present invention migrates the independent variable of the current particle based on the worst particle searched so far, makes full use of the information accumulated in the search history to guide the update optimization process, improves the convergence speed, makes the particle position update more accurate, solves the single-layer non-convex optimization model to obtain the optimal quantile level offset and the upper end point quantile level and the lower end point quantile level of the asymmetric interval of wind power generation more accurately, obtains the asymmetric interval of wind power generation power with a smaller width while ensuring reliability, and the accurate asymmetric interval of wind power generation power enables the power system to accurately perform power dispatch, thereby improving the power supply safety and stability of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to make the content of the present invention more clearly understood, the present invention is further described in detail below according to specific embodiments of the present invention in conjunction with the accompanying drawings, wherein:

[0059] Figure 1 It is a flow chart of the steps of the method for predicting the asymmetric interval of wind power generation provided by the present invention;

[0060] Figure 2 It is an update and iteration flow chart of the improved wind drive algorithm provided by the present invention;

[0061] Figure 3 It is a schematic diagram of the wind power generation power prediction interval under the 90% nominal coverage rate provided by the present invention. DETAILED DESCRIPTION

[0062] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention.

[0063] Reference Figure 1 As shown, the flow chart of the method for predicting the asymmetric interval of wind power generation power of the present invention includes the following specific steps:

[0064] S101: Based on quantile regression theory and extreme learning machine, the positive residual matrix and negative residual matrix of the wind power generation measurement value and the extreme learning machine output matrix are obtained. The asymmetric weighted sum of the positive residual matrix and the negative residual matrix is ​​minimized as the optimization goal. A quantile estimator for predicting the asymmetric interval of wind power generation is constructed as the lower optimization model, which is expressed as:

[0065] ;

[0066] in, is the number of scheduling periods; is the quantile level, and They are the upper endpoint quantile level and the lower endpoint quantile level of the asymmetric interval of wind power generation power respectively; Represents the positive residual matrix of the wind power measurement value and the extreme learning machine output matrix, Represents the negative residual matrix of the wind power generation measurement value and the extreme learning machine output matrix; is the input matrix of the extreme learning machine; is the measured value of wind power generation; is the parameter matrix to be optimized of the extreme learning machine, and They are respectively the upper endpoint parameter matrix to be optimized and the lower endpoint parameter matrix to be optimized output by the extreme learning machine; is the output matrix of the extreme learning machine, including the upper endpoint output matrix of the asymmetric interval of wind power generation and the lower endpoint output matrix .

[0067] S102: Taking the minimum overall width of the asymmetric interval of wind power generation predicted by the quantile estimator as the optimization goal, an upper optimization model is constructed, which is expressed as:

[0068] ;

[0069] in, is the number of scheduling periods; is the output matrix of the extreme learning machine, including the upper endpoint output matrix of the asymmetric interval of wind power generation and the lower endpoint output matrix ; is the quantile level, and They are the upper endpoint quantile level and the lower endpoint quantile level of the asymmetric interval of wind power generation power respectively; is the quantile level shift variable; is the input matrix of the extreme learning machine; is the parameter matrix to be optimized of the extreme learning machine, and are the upper endpoint parameter matrix to be optimized and the lower endpoint parameter matrix to be optimized output by the extreme learning machine respectively; the nominal coverage is , and They respectively represent the upper endpoint quantile level and the lower endpoint quantile level of the wind power generation center prediction interval.

[0070] S103: After replacing the complementary relaxation condition in the Karush-Kuhn-Tucker condition with the strong duality condition to obtain the Karush-Kuhn-Tucker condition based on the lower-layer convex optimization problem, the two-layer optimization model consisting of the lower-layer optimization model and the upper-layer optimization model is converted into a single-layer non-convex optimization model, which is expressed as:

[0071] ;

[0072] in, is the number of scheduling periods; is the quantile level, and They are the upper endpoint quantile level and the lower endpoint quantile level of the asymmetric interval of wind power generation power respectively; is the input matrix of the extreme learning machine; is the parameter matrix to be optimized of the extreme learning machine, and are the upper endpoint parameter matrix to be optimized and the lower endpoint parameter matrix to be optimized of the extreme learning machine respectively; the nominal coverage is , and They represent the upper endpoint quantile level and the lower endpoint quantile level of the wind power generation center prediction interval respectively; is the quantile level shift variable; Represents the positive residual matrix of the wind power measurement value and the extreme learning machine output matrix, Represents the negative residual matrix of the wind power generation measurement value and the extreme learning machine output matrix; is the measured value of wind power generation; is the output matrix of the extreme learning machine, including the upper endpoint output matrix of the asymmetric interval of wind power generation and the lower endpoint output matrix ;

[0073] Among them, the Karush-Kuhn-Tucker conditions based on the underlying convex optimization problem include:

[0074] The feasible condition of the original problem is expressed as: ;

[0075] The dual feasible condition is expressed as: ;

[0076] The first-order stationary value condition is expressed as: ;

[0077] The strong duality condition is expressed as: ;

[0078] represents the hidden layer output matrix of the extreme learning machine, , , , and are all Lagrange multipliers.

[0079] S104: In the initial velocity vector update formula of the wind drive algorithm, the coordinate origin of each particle is dynamically and randomly selected using the tournament selection mechanism to obtain a new origin of the particle; the particle velocity is updated based on the new origin of the particle to obtain an improved velocity vector update formula, which is expressed as:

[0080] ;

[0081] in, Indicates The particle velocity at iteration , represents the friction coefficient, Indicates The particle velocity at iteration , is the gravitational constant, represents the new origin of the particle obtained using the tournament selection mechanism, Indicates The particle position at iteration , is the universal gas constant, Indicates temperature, Indicates the sorting number of the current particle among all particles. represents the current optimal particle, represents the randomly selected particle Random particle positions for iterations; is a constant, and its value is , To preset the initial coefficient, Represents a random number uniformly generated within a range. Represents a metric based on the standard deviation of the current population particles, expressed as: ; Indicates Particle No. The particle position at iteration , Expressed as , Indicates the population size.

[0082] S105: In the initial position vector update formula of the wind drive algorithm, a variable control parameter is introduced as a coefficient to obtain an improved position vector update formula, which is expressed as:

[0083] ;

[0084] in, Indicates The particle position at iteration , Indicates The particle position at iteration , Indicates The particle velocity at iteration , represents the step length, Indicates a variable control parameter.

[0085] S106: Using an improved wind-driven algorithm based on an improved velocity vector update formula and an improved position vector update formula, the single-layer non-convex optimization model is solved to obtain an optimal solution of the single-layer non-convex optimization model; the optimal solution includes an optimal quantile level offset and an upper endpoint quantile level and a lower endpoint quantile level of an asymmetric interval of wind power generation; the sum of the upper endpoint quantile level and the optimal quantile level offset is taken as the upper endpoint of the interval, and the sum of the lower endpoint quantile level and the optimal quantile level offset is taken as the lower endpoint of the interval, to obtain an asymmetric interval of wind power generation power.

[0086] Specifically, in the embodiment of the present invention, after the particle position is updated by using the improved position vector update formula, the position of each particle is migrated by using a particle position migration formula; the particle position migration formula includes:

[0087] ;

[0088] in, Indicates Iteration No. The particle Dimensional variables; Indicates Iteration No. The particle Dimensional variables; represents the current optimal particle, represents the current optimal particle Dimensional variables; represents the worst particle, The worst particle Dimensional variables; Indicates the preset error limit; Represents the negative learning rate, which is a positive increasing constant and ranges from [0.002, 0.01].

[0089] After the particle position is updated by using the improved position vector update formula, the present invention migrates the independent variable of the current particle based on the worst particle searched so far, makes full use of the information accumulated in the search history to guide the update optimization process, improves the convergence speed, makes the particle position update more accurate, solves the single-layer non-convex optimization model to obtain the optimal quantile level offset and the upper end point quantile level and the lower end point quantile level of the asymmetric interval of wind power generation more accurately, obtains the asymmetric interval of wind power generation power with a smaller width while ensuring reliability, and the accurate asymmetric interval of wind power generation power enables the power system to accurately perform power dispatch, thereby improving the power supply safety and stability of the power system.

[0090] Reference Figure 2 As shown in FIG. 1 , it is an iterative flow chart of the improved wind-driven algorithm update; in the embodiment of the present invention, in step S106, the improved wind-driven algorithm based on the improved velocity vector update formula and the improved position vector update formula is used to solve the single-layer non-convex optimization model to obtain the optimal solution of the model, including:

[0091] S1: Based on the average interval width of the asymmetric interval of wind power generation, the optimal objective function is constructed, which is expressed as:

[0092] ; Indicates the average interval width of the asymmetric interval of wind power generation, , represents the population size, Indicates the population The upper end point of the asymmetric interval of wind power generation corresponding to each particle, Indicates the population The lower endpoint of the asymmetric interval of wind power generation corresponding to each particle;

[0093] S2: Initialize the improved wind drive algorithm: preset the first continuous search times threshold for the optimal particle without change and the second continuous search times threshold for the optimal particle without improvement, and initialize the position and speed of the particle;

[0094] S3: Update the velocity and position of the particle using the improved velocity vector update formula and the improved position vector update formula;

[0095] S4: Use the particle position migration formula to migrate the position of each particle;

[0096] S5: Determine whether the number of continuous searches without change of the optimal particle reaches a first continuous search number threshold;

[0097] S6: If the number of consecutive searches without change of the optimal particle reaches the first consecutive search number threshold, a new particle is randomly selected from the optimal particles searched in the history as the current optimal particle, and the process returns to step S3;

[0098] S7: If the number of continuous searches without changes in the optimal particle does not reach the first continuous search number threshold, determine whether the number of continuous searches without improvements in the optimal particle reaches the second continuous search number threshold;

[0099] S8: If the number of continuous searches without improvement of the optimal particle does not reach the second continuous search number threshold, return to step S3;

[0100] S9: If the number of continuous searches without improvement of the optimal particle reaches the second continuous search number threshold, the iteration is terminated, and the current quantile horizontal offset is obtained as the optimal quantile horizontal offset.

[0101] The method for predicting an asymmetric interval of wind power generation power described in the present invention constructs a double-layer optimization model based on the asymmetry of wind power generation power prediction, and converts the double-layer optimization model into a single-layer non-convex optimization model after replacing the complementary relaxation conditions in the Karush-Kuhn-Tucker conditions with strong dual conditions, thereby simplifying the problem structure, reducing the computational burden, and improving the solution efficiency. The improved wind-driven algorithm can be directly used to solve the original double-layer optimization model, so that the optimal quantile horizontal offset obtained by the solution is more accurate, and the reliability of the solution is improved, thereby ensuring the high reliability and smaller width of the asymmetric interval of wind power generation power. The improved wind-driven algorithm of the present invention adopts a tournament selection mechanism to dynamically and randomly select the coordinate origin of each particle, thereby preventing the velocity vector and the position vector from being attracted to the absolute origin of the coordinate system when updating, thereby improving particle diversity; and utilizing an improved position vector update formula to update the particle position, thereby avoiding the convergence to zero that may be caused by using the velocity vector, avoiding stagnation, and causing the search process to be trapped in a local optimal solution, thereby improving the search speed and accuracy, making it possible to obtain the shortest reliable asymmetric prediction interval of wind power generation with greater sharpness, thereby improving the reliability of the prediction of the asymmetric interval of wind power generation, so that the power system can adjust the power supply based on the accurate asymmetric interval of wind power generation, thereby reducing the cost invested by the power system in solving the uncertainty of wind power generation when making decisions, and improving the power supply stability of the power system.

[0102] Based on the above embodiment, in an embodiment of the present invention, the wind power generation power asymmetric interval prediction method proposed in the present invention is used to perform interval prediction, and the specific process includes:

[0103] S201: Based on the adaptive quantile regression theory, a two-layer optimization model for asymmetric interval prediction of wind power generation is constructed;

[0104] S202: solving the double-layer optimization model using an improved wind drive algorithm;

[0105] S203: Based on the wind power data actually measured at a certain wind farm within a preset time period, a case analysis is carried out, and the wind power generation power asymmetric interval prediction method provided by the present invention is used to obtain the wind power generation power prediction interval under the nominal coverage rate, and the test set is used for verification.

[0106] Specifically, the upper and lower endpoints of the wind power generation prediction interval can be represented by a pair of quantiles at a given quantile level. Therefore, based on the quantile regression theory, the present invention uses an extreme learning machine to construct a quantile estimator for the upper and lower endpoints of the wind power generation prediction interval. In step S201, the construction of a two-layer optimization model for asymmetric interval prediction of wind power generation includes:

[0107] S201-1: Lower optimization model of the two-layer optimization model for asymmetric interval prediction of wind power generation;

[0108] According to the quantile regression theory, the quantile estimators of the upper and lower endpoints of the wind power generation prediction interval can be expressed by an asymmetric weighted sum minimization optimization model of the positive and negative residuals of the wind power measurement value and the extreme learning machine output value;

[0109] The optimization objective of the lower model is to minimize the asymmetric weighted sum of the positive and negative residuals of the upper and lower endpoints of the prediction interval of the wind power measurement value and the extreme learning machine output. Its expression is expressed as follows:

[0110] ;

[0111] in, is the number of scheduling periods; is the quantile level, and They are the upper endpoint quantile level and the lower endpoint quantile level of the asymmetric interval of wind power generation power respectively; is the input matrix of the extreme learning machine; is the parameter matrix to be optimized of the extreme learning machine, and They are respectively the upper endpoint parameter matrix to be optimized and the lower endpoint parameter matrix to be optimized output by the extreme learning machine; is the measured value of wind power generation; is the output matrix of the extreme learning machine, including the upper endpoint output matrix of the asymmetric interval of wind power generation and the lower endpoint output matrix ; and They are the positive residual and negative residual matrices of the wind power measurement value and the output matrix of the extreme learning machine, respectively.

[0112] S201-2: Upper optimization model of the two-layer optimization model for asymmetric interval prediction of wind power generation;

[0113] Assuming that the nominal coverage of the wind power generation prediction interval is 100 (1-β)%, the quantile levels corresponding to the upper and lower endpoints of the central prediction interval of wind power generation should be β / 2 and 1-(β / 2), respectively. However, since the probability distribution of the inaccuracy of renewable energy power generation prediction is generally asymmetric, the width of the central prediction interval will be relatively large; therefore, it is necessary to optimize the quantile levels corresponding to the upper and lower endpoints of the central prediction interval of wind power generation. The upper model takes the minimum overall width of the wind power generation prediction interval as the optimization goal, and its expression is expressed as:

[0114] ;

[0115] in, It is the quantile level offset variable, which represents the offset of the quantile level corresponding to the upper and lower endpoints of the asymmetric interval of wind power generation relative to the central prediction interval.

[0116] S201-3: Single-layer equivalent optimization model for asymmetric interval prediction of wind power generation;

[0117] The lower optimization model for wind power generation asymmetric interval prediction is a linear programming problem, and the optimal solution of the lower optimization problem can be expressed by KKT conditions. The expression of the optimal solution of the lower optimization model for wind power generation asymmetric interval prediction transformed by KKT conditions includes:

[0118] The feasibility conditions of the original problem in KKT conditions are: ;

[0119] The dual feasible condition in KKT condition is: ;

[0120] The first-order stationary value condition in the KKT condition is: ;

[0121] Since the complementary relaxation conditions in the KKT conditions contain a large number of bilinear terms, they can be replaced by strong dual conditions to eliminate a large number of bilinear terms in the original KKT conditions; the strong dual conditions are expressed as: ;

[0122] in, , , , and All are Lagrange multipliers; is the hidden layer output matrix of the extreme learning machine.

[0123] By adopting KKT conditions to perform equivalent transformation on the lower optimization model of wind power generation asymmetric interval prediction, the original two-layer optimization model of wind power generation asymmetric interval prediction can be equivalently transformed into a single-layer non-convex optimization model; the single-layer non-convex equivalent optimization model of wind power generation asymmetric interval prediction is expressed as:

[0124] ;

[0125] S201-4: The probability interval prediction method is evaluated comprehensively from the perspectives of reliability and sharpness.

[0126] ① Reliability represents the correctness of the prediction results, emphasizing that the prediction results should be as consistent as possible with the observed results, which can be expressed by the average coverage probability deviation; the smaller the average coverage probability deviation, the higher the reliability; the calculation expression of the average coverage probability deviation is:

[0127] ;

[0128] in, is the sample size; is the lower endpoint of the prediction interval; is the upper endpoint of the prediction interval; is an indicator function. When the expression in the brackets is satisfied, the output value of the indicator function is 1; otherwise, the output value of the indicator function is 0.

[0129] ② Sharpness represents the compactness of the prediction interval, which can be expressed by the average interval width; the smaller the average interval width, the higher the sharpness. The calculation expression of the average interval width is:

[0130] .

[0131] Specifically, in step S202, since the single-layer equivalent optimization model for asymmetric interval prediction of wind power generation power constructed by the present invention is non-convex, the constructed model cannot be directly solved by commercial solvers such as cplex or gurobi; therefore, an embodiment of the present invention uses an improved wind-driven algorithm to solve the single-layer equivalent optimization model for asymmetric interval prediction of wind power generation power.

[0132] The inspiration of the wind-driven optimization algorithm WDO comes from the climate model; in nature, wind blows from high-pressure areas to low-pressure areas at different speeds to balance the imbalance of air pressure; according to Newton's second law of motion, the initial update process of the velocity vector and position vector in the wind-driven optimization algorithm is expressed as:

[0133] The initial velocity vector update formula is expressed as:

[0134] ;

[0135] The initial position vector update formula is expressed as: ;

[0136] in, is the friction coefficient, is the gravitational constant, is the universal gas constant, is the temperature, is a constant, is the ranking of the current particle among all particles, is the best particle found so far. is the step length.

[0137] Compared with other heuristic algorithms, the wind-driven optimization algorithm (WDO) has a clear principle and simple numerical implementation; however, due to the possible loss of diversity of particles and the difficulty in balancing exploration and development searches, WDO is prone to premature convergence when solving complex global optimization problems. To solve the above problems, the embodiment of the present invention improves the wind-driven optimization algorithm and proposes an improved wind-driven algorithm to solve a single-layer non-convex optimization model for asymmetric interval prediction of wind power generation.

[0138] Specifically, the process of obtaining the improved wind drive algorithm includes:

[0139] ①Position and velocity update strategy;

[0140] According to the laws of physics, the second term in the above velocity vector update formula represents attraction, which attracts particles to the absolute origin of the coordinate system. Because each particle is always attracted to the absolute origin under the influence of attraction, the diversity of particles deteriorates. To solve this problem, the improved wind drive algorithm in this embodiment adopts a tournament selection mechanism to dynamically and randomly select the coordinate origin of each particle. In the tournament selection mechanism, two feasible particles are selected from all particles in the current population, and the better one is used as the origin of the current population. The velocity vector update formula and the position vector update formula are modified to obtain:

[0141] The improved velocity vector update formula is expressed as:

[0142] ;

[0143] Improve the position vector update formula, expressed as:

[0144] ;

[0145] in, is the new origin of the particle obtained using the tournament selection mechanism and is the control parameter that decreases linearly from the maximum to the minimum value during the iteration process.

[0146] ②Optimal particle selection mechanism;

[0147] In order to further balance the global and refined search, the best particle currently searched in the revised velocity vector update formula is , if there is no change after a given number of consecutive search cycles Nre, it will be randomly reinitialized.

[0148] If after N consecutive iterations, the best solution so far has not been further improved, a new best particle is randomly selected from the best particles searched in the past ;

[0149] For a problem with N optimization variables, the dimension of the independent variables is N. The selection of intervals for independent variables of different dimensions has a significant impact on engineering optimization research or inverse problems. The last term in the modified velocity vector update formula shows that inappropriate interval selection may lead to extremely small or extremely large velocities. To ensure that the interval size is appropriate, the intervals of all independent variables are normalized to [0 1].

[0150] ③Adaptive mutation operation;

[0151] The embodiment of the present invention introduces disturbance to the velocity vector based on the last term in the initial velocity vector update formula to ensure the diversity of the velocity. Obviously, this term plays the role of a mutation operator in the optimization process. However, in the search phase of complex optimization problems, the velocity vector may converge to zero, resulting in the so-called stagnation phenomenon, causing the search process to be limited to a local optimal solution. To solve this problem, the embodiment of the present invention uses the position vector instead of the velocity vector. In addition, in order to maintain a good balance between global and refined searches, if the diversity of particles in the current population is high, the mutation rate is small, and vice versa. To achieve this goal, the embodiment of the present invention introduces a metric based on the standard deviation of the particles in the current population, expressed as:

[0152] ;

[0153] in, N Indicates the population size.

[0154] In the embodiment of the present invention, the velocity vector is randomly corrected using this metric, and the above corrected velocity vector update formula is further modified to obtain an improved velocity vector update formula, which is expressed as:

[0155] ;

[0156] ;

[0157] in, is a uniformly generated random number in the range [0 1].

[0158] ④ Migration strategy based on the worst particles;

[0159] In order to make full use of the information accumulated in the search history to guide the optimization process and improve the convergence speed, the embodiment of the present invention migrates the current particle based on the worst particle searched so far. After the position of particle j is updated using the position vector after the first correction, the improved position vector update formula is further used to migrate particle j's d as an independent variable; the particle position migration formula is expressed as:

[0160] ;

[0161] in, It is the worst particle ever found. d dimensional variables, It is a positive increasing constant, called negative learning rate, and its value range is [0.002 0.01].

[0162] Based on the above improved velocity vector update formula and improved position vector update formula, the improved wind drive algorithm is run. The specific iterative process includes:

[0163] Initialize algorithm parameters;

[0164] Based on the improved velocity vector update formula, the improved position vector update formula and the particle position migration formula, the velocity and position of the current particle are updated;

[0165] Determine the optimal particle Whether there is a change after a given number of consecutive search cycles (Nre):

[0166] If there is no change, a new particle is randomly selected from the best particles searched in the history. Randomly select a new ;

[0167] If there is no change, determine whether the optimal objective function has been improved after Ns consecutive iterations:

[0168] If there is no improvement, the algorithm terminates and outputs the optimization result, i.e., the optimal quantile level offset and the shortest reliable prediction interval of renewable energy power generation with greater sharpness;

[0169] If there is improvement, return to update the speed and position of the current particle until it is determined that the optimal objective function has not been improved after Ns consecutive iterations, and output the optimization result.

[0170] Specifically, based on the above embodiment, in this embodiment, based on the wind power data with a sampling time of 10 minutes obtained by actual measurement of a wind farm, the wind power generation power asymmetric interval prediction method provided by the present invention is used to perform case analysis and calculate the analysis results. In this embodiment, the acquired wind power data is divided into a training set and a test set, the training set is used to train the extreme learning machine model, and the test set is used to verify the effectiveness of the prediction method provided by the present invention.

[0171] In this embodiment, the nominal coverage rate is set to 90%, and the quantile horizontal offset obtained by using the asymmetric interval prediction method for wind power generation provided by the present invention is , that is, the quantile levels corresponding to the upper and lower endpoints of the shortest reliable prediction interval of wind power generation are 0.9632 and 0.0632 respectively.

[0172] The proposed method is verified based on all wind power data in the test set, and the wind power generation power prediction interval diagram under 90% nominal coverage is finally obtained as shown in the figure. Figure 3 As shown in the figure, the average coverage probability deviation of the prediction interval obtained using all wind power data in the test set is -0.0075, the empirical coverage is basically consistent with the nominal coverage, and has high reliability; the average interval width of the prediction interval is 4.1824, which has high sharpness.

[0173] according to Figure 3 It can be seen that the asymmetric interval prediction method for wind power generation based on the improved wind drive algorithm proposed in the present invention can obtain the shortest reliable prediction interval of wind power generation with a small average interval width, most of the real wind power in the test set data can fall within the shortest reliable prediction interval, the empirical coverage rate is basically consistent with the nominal coverage rate, and it has high reliability, and the shortest reliable prediction interval finally obtained has high sharpness. The effectiveness of the asymmetric interval prediction method for wind power generation based on the improved wind drive algorithm proposed in the present invention has been verified.

[0174] The present invention proposes a method for predicting asymmetric intervals of wind power generation based on an improved wind drive algorithm. Compared with existing methods, the sharpness of the reliable prediction interval of renewable energy generation is larger, which can effectively reduce the cost of the power system to solve the uncertainty of renewable energy generation when making decisions and improve the stability of the power system.

[0175] Based on the above embodiment, the embodiment of the present invention further provides a device for predicting an asymmetric interval of wind power generation power, comprising:

[0176] The single-layer non-convex optimization model construction module 100 is used to obtain the positive residual matrix and the negative residual matrix of the wind power generation measurement value and the extreme learning machine output matrix based on the quantile regression theory and the extreme learning machine, and to construct a quantile estimator for predicting the asymmetric interval of wind power generation as a lower-layer optimization model with the minimization of the asymmetric sum of the positive residual matrix and the negative residual matrix as the optimization target; to construct an upper-layer optimization model with the minimization of the overall width of the asymmetric interval of wind power generation predicted by the quantile estimator as the optimization target; to replace the complementary relaxation condition in the Karush-Kuhn-Tucker condition with the strong duality condition, and after obtaining the Karush-Kuhn-Tucker condition based on the lower-layer convex optimization problem, convert the double-layer optimization model composed of the lower-layer optimization model and the upper-layer optimization model into a single-layer non-convex optimization model;

[0177] The wind-driven algorithm improvement module 200 is used to dynamically and randomly select the coordinate origin of each particle in the initial velocity vector update formula of the wind-driven algorithm by using the tournament selection mechanism to obtain a new origin of the particle; update the particle velocity based on the new origin of the particle to obtain an improved velocity vector update formula; introduce a variable control parameter as a coefficient in the initial position vector update formula of the wind-driven algorithm to obtain an improved position vector update formula;

[0178] The solution prediction module 300 is used to solve the single-layer non-convex optimization model using an improved wind-driven algorithm based on an improved velocity vector update formula and an improved position vector update formula to obtain the optimal solution of the single-layer non-convex optimization model; the optimal solution includes an optimal quantile level offset and an upper endpoint quantile level and a lower endpoint quantile level of an asymmetric interval of wind power generation; the sum of the upper endpoint quantile level and the optimal quantile level offset is used as the upper endpoint of the interval, and the sum of the lower endpoint quantile level and the optimal quantile level offset is used as the lower endpoint of the interval to obtain the asymmetric interval of wind power generation.

[0179] The wind power generation power asymmetric interval prediction device of this embodiment is used to implement the aforementioned wind power generation power asymmetric interval prediction method. Therefore, the specific implementation method of the wind power generation power asymmetric interval prediction device can be seen in the embodiment part of the wind power generation power asymmetric interval prediction method in the previous text, for example, the single-layer non-convex optimization model construction module 100 is used to implement steps S101, S102 and S103 in the above-mentioned wind power generation power asymmetric interval prediction method; the wind drive algorithm improvement module 200 is used to implement steps S104 and S105 in the above-mentioned wind power generation power asymmetric interval prediction method; the solution prediction module 300 is used to implement step S106 in the above-mentioned wind power generation power asymmetric interval prediction method. Therefore, its specific implementation method can refer to the description of the corresponding various parts of the embodiment, which will not be repeated here.

[0180] The method and device for predicting an asymmetric interval of wind power generation power described in the present invention construct a double-layer optimization model based on the asymmetry of wind power generation power prediction, and convert the double-layer optimization model into a single-layer non-convex optimization model after replacing the complementary relaxation conditions in the Karush-Kuhn-Tucker conditions with strong dual conditions, thereby simplifying the problem structure, reducing the computational burden, and improving the solution efficiency. The improved wind-driven algorithm can be directly used to solve the original double-layer optimization model, so that the optimal quantile horizontal offset obtained by the solution is more accurate, and the reliability of the solution is improved, thereby ensuring the high reliability and small width of the asymmetric interval of wind power generation power. The improved wind-driven algorithm of the present invention adopts a tournament selection mechanism to dynamically and randomly select the coordinate origin of each particle, thereby preventing the velocity vector and the position vector from being attracted to the absolute origin of the coordinate system when updating, thereby improving particle diversity; and utilizing an improved position vector update formula to update the particle position, thereby avoiding the convergence to zero that may be caused by using the velocity vector, avoiding stagnation, and causing the search process to be trapped in a local optimal solution, thereby improving the search speed and accuracy, making it possible to obtain the shortest reliable asymmetric prediction interval of wind power generation with greater sharpness, thereby improving the reliability of the prediction of the asymmetric interval of wind power generation, so that the power system can adjust the power supply based on the accurate asymmetric interval of wind power generation, thereby reducing the cost invested by the power system in solving the uncertainty of wind power generation when making decisions, and improving the power supply stability of the power system.

[0181] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0182] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0183] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0184] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0185] Obviously, the above embodiments are merely examples for the purpose of clear explanation and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the present invention.

Claims

1. A method for predicting asymmetric interval of wind power generation, characterized in that: include: Based on quantile regression theory and extreme learning machine, the positive residual matrix and negative residual matrix of wind power generation measurement value and extreme learning machine output matrix are obtained. Taking the asymmetric weighted sum of the positive residual matrix and the negative residual matrix as the minimum optimization target, a quantile estimator for predicting the asymmetric interval of wind power generation is constructed as the lower optimization model. Taking the minimum overall width of the asymmetric interval of wind power generation predicted by the quantile estimator as the optimization goal, an upper optimization model is constructed; After the complementary relaxation condition in the Karush-Kuhn-Tucker condition is replaced by the strong duality condition and the Karush-Kuhn-Tucker condition based on the lower-layer convex optimization problem is obtained, the two-layer optimization model consisting of the lower-layer optimization model and the upper-layer optimization model is converted into a single-layer non-convex optimization model, which is expressed as: ； in, is the number of scheduling periods; is the quantile level, and They are the upper endpoint quantile level and the lower endpoint quantile level of the asymmetric interval of wind power generation power respectively; is the input matrix of the extreme learning machine; is the parameter matrix to be optimized of the extreme learning machine, and are the upper endpoint parameter matrix to be optimized and the lower endpoint parameter matrix to be optimized of the extreme learning machine respectively; the nominal coverage is , and They represent the upper endpoint quantile level and the lower endpoint quantile level of the wind power generation center prediction interval respectively; is the quantile level shift variable; Represents the positive residual matrix of the wind power measurement value and the extreme learning machine output matrix, Represents the negative residual matrix of the wind power generation measurement value and the extreme learning machine output matrix; is the measured value of wind power generation; is the output matrix of the extreme learning machine, including the upper endpoint output matrix of the asymmetric interval of wind power generation and the lower endpoint output matrix ; Among them, the Karush-Kuhn-Tucker conditions based on the underlying convex optimization problem include: The feasible condition of the original problem is expressed as: ; The dual feasible condition is expressed as: ; The first-order stationary value condition is expressed as: ; The strong duality condition is expressed as: ; represents the hidden layer output matrix of the extreme learning machine, , , , and All are Lagrange multipliers; In the initial velocity vector update formula of the wind drive algorithm, the tournament selection mechanism is used to dynamically and randomly select the coordinate origin of each particle to obtain the new origin of the particle; the particle velocity is updated based on the new origin of the particle to obtain the improved velocity vector update formula, which is expressed as: ； in, Indicates The particle velocity at iteration , represents the friction coefficient, Indicates The particle velocity at iteration , is the gravitational constant, represents the new origin of the particle obtained using the tournament selection mechanism, Indicates The particle position at iteration , is the universal gas constant, Indicates temperature, Indicates the sorting number of the current particle among all particles. represents the current optimal particle, represents the randomly selected particle Random particle positions for iterations; is a constant, and its value is , To preset the initial coefficient, Represents a random number uniformly generated within a range. represents a metric based on the standard deviation of particles in the current population; In the initial position vector update formula of the wind drive algorithm, the variable control parameter is introduced as a coefficient to obtain the improved position vector update formula, which is expressed as: ； in, Indicates The particle position at iteration , Indicates The particle position at iteration , Indicates The particle velocity at iteration , represents the step length, Indicates variable control parameters; The improved wind-driven algorithm based on the improved velocity vector update formula and the improved position vector update formula is used to solve the single-layer non-convex optimization model and obtain the optimal solution of the single-layer non-convex optimization model, including: S1: Based on the average interval width of the asymmetric interval of wind power generation, the optimal objective function is constructed, which is expressed as: ; Indicates the average interval width of the asymmetric interval of wind power generation, , represents the population size, Indicates the population The upper end point of the asymmetric interval of wind power generation corresponding to each particle, Indicates the population The lower endpoint of the asymmetric interval of wind power generation corresponding to each particle; S2: Initialize the improved wind drive algorithm: preset the first continuous search times threshold for the optimal particle without change and the second continuous search times threshold for the optimal particle without improvement, and initialize the position and speed of the particle; S3: Update the velocity and position of the particle using the improved velocity vector update formula and the improved position vector update formula; S4: Use the particle position migration formula to migrate the position of each particle; S5: Determine whether the number of continuous searches without change of the optimal particle reaches a first continuous search number threshold; S6: If the number of consecutive searches without change of the optimal particle reaches the first consecutive search number threshold, a new particle is randomly selected from the optimal particles searched in the history as the current optimal particle, and the process returns to step S3; S7: If the number of continuous searches without changes in the optimal particle does not reach the first continuous search number threshold, determine whether the number of continuous searches without improvements in the optimal particle reaches the second continuous search number threshold; S8: If the number of continuous searches without improvement of the optimal particle does not reach the second continuous search number threshold, return to step S3; S9: If the number of continuous searches without improvement of the optimal particle reaches the second continuous search number threshold, the iteration is terminated, and the current quantile level offset is obtained as the optimal quantile level offset; The optimal solution includes an optimal quantile level offset and an upper endpoint quantile level and a lower endpoint quantile level of an asymmetric interval of wind power generation power; The sum of the upper endpoint quantile level and the optimal quantile level offset is taken as the upper endpoint of the interval, and the sum of the lower endpoint quantile level and the optimal quantile level offset is taken as the lower endpoint of the interval, and the asymmetric interval of wind power generation is obtained.

2. The method for predicting asymmetric interval of wind power generation according to claim 1, characterized in that: The lower layer optimization model is expressed as: ； in, is the number of scheduling periods; is the quantile level, and They are the upper endpoint quantile level and the lower endpoint quantile level of the asymmetric interval of wind power generation power respectively; Represents the positive residual matrix of the wind power measurement value and the extreme learning machine output matrix, Represents the negative residual matrix of the wind power generation measurement value and the extreme learning machine output matrix; is the input matrix of the extreme learning machine; is the measured value of wind power generation; is the parameter matrix to be optimized of the extreme learning machine, and They are respectively the upper endpoint parameter matrix to be optimized and the lower endpoint parameter matrix to be optimized output by the extreme learning machine; is the output matrix of the extreme learning machine, including the upper endpoint output matrix of the asymmetric interval of wind power generation and the lower endpoint output matrix .

3. The method for predicting asymmetric interval of wind power generation according to claim 1, characterized in that: The upper layer optimization model is expressed as: ； in, is the number of scheduling periods; is the output matrix of the extreme learning machine, including the upper endpoint output matrix of the asymmetric interval of wind power generation and the lower endpoint output matrix ; is the quantile level, and They are the upper endpoint quantile level and the lower endpoint quantile level of the asymmetric interval of wind power generation power respectively; is the quantile level shift variable; is the input matrix of the extreme learning machine; is the parameter matrix to be optimized of the extreme learning machine, and are the upper endpoint parameter matrix to be optimized and the lower endpoint parameter matrix to be optimized output by the extreme learning machine respectively; the nominal coverage is , and They respectively represent the upper endpoint quantile level and the lower endpoint quantile level of the wind power generation center prediction interval.

4. The method for predicting asymmetric interval of wind power generation according to claim 1, characterized in that: The metric based on the standard deviation of the current population particles is expressed as: ； in, Indicates Particle No. The particle position at iteration , Expressed as , Indicates the population size.

5. The method for predicting asymmetric interval of wind power generation according to claim 1, characterized in that: After updating the particle position by using the improved position vector updating formula, the position of each particle is migrated by using the particle position migration formula; The particle position migration formula includes: ； in, Indicates Iteration No. The particle Dimensional variables; Indicates Iteration No. The particle Dimensional variables; represents the current optimal particle, represents the current optimal particle Dimensional variables; represents the worst particle, The worst particle Dimensional variables; Indicates the preset error limit; Represents the negative learning rate, which is a positive increasing constant and ranges from [0.002, 0.01].

6. A device for predicting asymmetric intervals of wind power generation, characterized in that: include: A single-layer non-convex optimization model construction module is used to obtain the positive residual matrix and negative residual matrix of the wind power generation measurement value and the extreme learning machine output matrix based on the quantile regression theory and the extreme learning machine, and to construct a quantile estimator for predicting the asymmetric interval of wind power generation as the lower-layer optimization model with the asymmetric weighted sum of the positive residual matrix and the negative residual matrix as the optimization target; Taking the minimum overall width of the asymmetric interval of wind power generation predicted by the quantile estimator as the optimization goal, an upper optimization model is constructed; After the complementary relaxation condition in the Karush-Kuhn-Tucker condition is replaced by the strong duality condition and the Karush-Kuhn-Tucker condition based on the lower-layer convex optimization problem is obtained, the two-layer optimization model consisting of the lower-layer optimization model and the upper-layer optimization model is converted into a single-layer non-convex optimization model, which is expressed as: ； in, is the number of scheduling periods; is the quantile level, and They are the upper endpoint quantile level and the lower endpoint quantile level of the asymmetric interval of wind power generation power respectively; is the input matrix of the extreme learning machine; is the parameter matrix to be optimized of the extreme learning machine, and are the upper endpoint parameter matrix to be optimized and the lower endpoint parameter matrix to be optimized of the extreme learning machine respectively; the nominal coverage is , and They represent the upper endpoint quantile level and the lower endpoint quantile level of the wind power generation center prediction interval respectively; is the quantile level shift variable; Represents the positive residual matrix of the wind power measurement value and the extreme learning machine output matrix, Represents the negative residual matrix of the wind power generation measurement value and the extreme learning machine output matrix; is the measured value of wind power generation; is the output matrix of the extreme learning machine, including the upper endpoint output matrix of the asymmetric interval of wind power generation and the lower endpoint output matrix ; Among them, the Karush-Kuhn-Tucker conditions based on the underlying convex optimization problem include: The feasible condition of the original problem is expressed as: ; The dual feasible condition is expressed as: ; The first-order stationary value condition is expressed as: ; The strong duality condition is expressed as: ; represents the hidden layer output matrix of the extreme learning machine, , , , and All are Lagrange multipliers; The wind-driven algorithm improvement module is used to dynamically and randomly select the coordinate origin of each particle in the initial velocity vector update formula of the wind-driven algorithm using the tournament selection mechanism to obtain the new origin of the particle; based on the new origin of the particle, the particle velocity is updated to obtain the improved velocity vector update formula, which is expressed as: ;in, Indicates The particle velocity at iteration , represents the friction coefficient, Indicates The particle velocity at iteration , is the gravitational constant, represents the new origin of the particle obtained using the tournament selection mechanism, Indicates The particle position at iteration , is the universal gas constant, Indicates temperature, Indicates the sorting number of the current particle among all particles. represents the current optimal particle, represents the randomly selected particle Random particle positions for iterations; is a constant, and its value is , To preset the initial coefficient, Represents a random number uniformly generated within a range. Represents a metric based on the standard deviation of the current population particles; in the initial position vector update formula of the wind drive algorithm, a variable control parameter is introduced as a coefficient to obtain an improved position vector update formula, which is expressed as: ;in, Indicates The particle position at iteration , Indicates The particle position at iteration , Indicates The particle velocity at iteration , represents the step length, Indicates variable control parameters; A solution prediction module is used to solve the single-layer non-convex optimization model by using an improved wind drive algorithm based on an improved velocity vector update formula and an improved position vector update formula to obtain an optimal solution of the single-layer non-convex optimization model, including: S1: Based on the average interval width of the asymmetric interval of wind power generation, the optimal objective function is constructed, which is expressed as: ; Indicates the average interval width of the asymmetric interval of wind power generation, , represents the population size, Indicates the population The upper end point of the asymmetric interval of wind power generation corresponding to each particle, Indicates the population The lower endpoint of the asymmetric interval of wind power generation corresponding to each particle; S2: Initialize the improved wind drive algorithm: preset the first continuous search times threshold for the optimal particle without change and the second continuous search times threshold for the optimal particle without improvement, and initialize the position and speed of the particle; S3: Update the velocity and position of the particle using the improved velocity vector update formula and the improved position vector update formula; S4: Use the particle position migration formula to migrate the position of each particle; S5: Determine whether the number of continuous searches without change of the optimal particle reaches a first continuous search number threshold; S6: If the number of consecutive searches without change of the optimal particle reaches the first consecutive search number threshold, a new particle is randomly selected from the optimal particles searched in the history as the current optimal particle, and the process returns to step S3; S7: If the number of continuous searches without changes in the optimal particle does not reach the first continuous search number threshold, determine whether the number of continuous searches without improvements in the optimal particle reaches the second continuous search number threshold; S8: If the number of continuous searches without improvement of the optimal particle does not reach the second continuous search number threshold, return to step S3; S9: If the number of continuous searches without improvement of the optimal particle reaches the second continuous search number threshold, the iteration is terminated, and the current quantile level offset is obtained as the optimal quantile level offset; The optimal solution includes the optimal quantile level offset and the upper endpoint quantile level and the lower endpoint quantile level of the asymmetric interval of wind power generation power; the sum of the upper endpoint quantile level and the optimal quantile level offset is taken as the upper endpoint of the interval, and the sum of the lower endpoint quantile level and the optimal quantile level offset is taken as the lower endpoint of the interval to obtain the asymmetric interval of wind power generation power.

Images