CN115758137A - Photovoltaic power prediction method and detection method based on differential evolution algorithm - Google Patents

Abstract

The invention provides a photovoltaic power prediction method and a photovoltaic power detection method based on a differential evolution algorithm. The prediction method comprises the steps of constructing a photovoltaic power prediction model by utilizing a support vector machine, coding training parameters of the support vector machine into individuals in a population, designing a differential evolution algorithm based on adaptive reverse variation to optimize the training parameters of the support vector machine, firstly calculating an adaptive fusion factor in the optimization process, then selecting excellent individuals from the population, fusing reverse information of the excellent individuals and information of the optimal individuals in the population according to the adaptive fusion factor to obtain basic individuals, executing reverse variation operation by utilizing the basic individuals to generate variant individuals, executing cross operation to generate test individuals, and executing selection operation to eliminate the population. The method utilizes the reverse information of the excellent individuals and the information of the optimal individuals in the population to enhance the optimization performance of the differential evolution algorithm, and can improve the accuracy of photovoltaic power prediction.

Description

Photovoltaic power prediction method and detection method based on differential evolution algorithm

Technical Field

The invention relates to the technical field of machine learning, in particular to a photovoltaic power prediction method based on a differential evolution algorithm.

Background

The photovoltaic power generation has the advantages of sustainability, environmental protection and the like. Photovoltaic power generation has become an important part of the power grid in China. Since the photovoltaic power generation needs the irradiation of sunlight, the photovoltaic power station can generate power well only when the sunlight is sufficient. Therefore, photovoltaic power generation has large fluctuation. The volatility of photovoltaic power generation presents significant challenges to the control and planning of photovoltaic power generation. In order to better control and plan the photovoltaic power generation, technicians must effectively predict the output power of the photovoltaic power plant. However, the output power of a photovoltaic power plant is affected by many factors, such as geographical location, photovoltaic power generation equipment, and climate. Therefore, photovoltaic power prediction is a very challenging technical problem.

In order to predict photovoltaic power more accurately, researchers propose a photovoltaic power prediction method based on machine learning [ Zhu Yongjiang, tian Jun ] application of a least square support vector machine in photovoltaic power prediction [ J ]. Power grid technology, 2011,35 (07): 54-59]. Support vector machines are a widely used machine learning technique that has met with some success in many practical engineering applications. However, the training parameter setting of the support vector machine can greatly affect the prediction accuracy of the support vector machine, and in practical engineering application, a technician often cannot obtain satisfactory prediction accuracy due to unreasonable training parameter setting of the support vector machine, which limits the application of the support vector machine to a certain extent. Therefore, when the support vector machine is applied to photovoltaic power prediction, the problem of insufficient prediction accuracy is easy to occur.

Disclosure of Invention

Based on the method, the photovoltaic power prediction method and the detection method based on the differential evolution algorithm are provided, the defect that the prediction precision is insufficient when the traditional support vector machine is applied to photovoltaic power prediction is overcome to a certain extent, and the accuracy of photovoltaic power prediction can be improved.

The invention provides a photovoltaic power prediction method based on a differential evolution algorithm, which comprises the following steps of:

step 1, collecting a photovoltaic power data set from a photovoltaic power station operation and maintenance system;

step 2, preprocessing the collected photovoltaic power data set;

step 3, dividing the preprocessed photovoltaic power data set into a training data set and a testing data set;

step 4, inputting a population size GPZ and a maximum search algebra gMaxT;

step 5, setting a current search algebra t =0;

in the step (6), the step (B), random generation population Pop = { GB = ₁ ,GB ₂ ,...,GB _bi ,...,GB _GPZ In which GB _bi Represents the second individual in the population, and individual GB _bi ＝[GB _bi，1 GB _bi,2 GB _bi，3 ]Three training parameters, namely GB, of the support vector machine are stored _bi，1 Represents the penalty coefficient C, GB _bi,2 Representing the coefficients of the kernel function σ, and GB _bi,3 A coefficient epsilon representing an insensitive loss function; individual subscript bi =1,2, ·, GPZ;

step 7, calculating the adaptive values of all individuals in the population;

step 8, finding out the individual with the minimum adaptation value in the population and recording as the optimal individual MinB;

step 9, setting inventory search step length HF _bi = rand (0,1), where rand denotes a random real number generation function;

step 10, setting inventory crossing rate HR _bi ＝rand(0,1)；

Step 11, calculating the current search step CF according to the formula (1) _bi ：

Wherein tpr1 is a random real number between [0,1 ];

step 12, calculating the current crossing rate CR according to the formula (2) _bi ：

Wherein tpr2 is a random real number between [0,1 ];

step 13, calculating the adaptive fusion factor cw according to the formula (3):

step 14, randomly selecting three individuals from the population: GB _ri1 ，GB _ri2 And GB _ri3 ；

Step 15, individual GB _ri1 ，GB _ri2 And GB _ri3 Recording the excellent individual with the minimum medium adaptive value as LMB;

step 16, performing inverse variation operation according to the formula (4) to generate varied individual MU _bi ：

NU _bi ＝(1-cw)×[(PL+PU)-LMB]+cw×MinB+CF _bi ×(GB _bi -GB _ri2 ) (4)

Wherein PL is the lower search boundary of the population, and PU is the upper search boundary of the population;

step 17, generating variant individual NU by performing variant operation according to formula (5) _bi ：

NU _bi ＝(1-cw)×LMB+cw×MinB+CF _bi ×(GB _bi -GB _ri2 ) (5)

Step 18, generating the test individual TV by performing the cross operation according to the formula (6) _bi ：

Wherein, MU _bi,hj Representing variant individuals MU _bi The hj training parameter of the support vector machine stored in the storage device; GB _bi,hj Representing individual GB _bi The hj training parameter of the support vector machine stored in the storage device; TV (television) _bi,hj Representing test individuals TV _bi The hj training parameter of the support vector machine stored in the storage device; dimension subscript hj =1,2,3;

step 19, generating individual XV according to the cross operation executed by formula (7) _bi ：

Wherein, NU _bi,hj Representing variant individual NU _bi The hj training parameter of the support vector machine stored in the storage device; XV _bi,hj Representing test individuals XV _bi The hj training parameter of the support vector machine stored in the storage device;

step 20, calculating the TV of the test individual _bi And test individuals XV _bi An adaptation value of;

step 21, if the individual TV is tested _bi Is less than the individual XV _bi The adapted value of (1) then sets the successful individual EV _bi ＝TV _bi Otherwise, setting a successful individual EV _bi ＝XV _bi ；

Step 22, if the individual EV is successful _bi Adapted value of less than individual GB _bi Go to step 23, otherwise go to step 26;

step 23, utilizing successful individual EV _bi Replacing individual GB in a population _bi ；

Step 24, set inventory search step size HF _bi ＝CF _bi ；

Step 25, setting inventory crossing rate HR _bi ＝CR _bi ；

Step 26, finding out the individual with the minimum adaptation value in the population and recording the individual as the optimal individual MinB;

step 27, setting a current search algebra t = t +1;

step 28, if the current search algebra t is less than gMaxT, go to step 11, otherwise go to step 29;

and 29, extracting three training parameters of the support vector machine from the optimal individual MinB, and then training the support vector machine for predicting the photovoltaic power on a training data set by using the three training parameters of the support vector machine to realize photovoltaic power prediction.

The photovoltaic power prediction method based on the differential evolution algorithm utilizes the support vector machine to construct a photovoltaic power prediction model, and designs the differential evolution algorithm based on the adaptive reverse variation to optimize the training parameters of the support vector machine. In the differential evolution algorithm of the adaptive reverse variation, an adaptive fusion factor is firstly calculated, then a good individual is selected from a population, and the reverse information of the good individual and the information of the optimal individual in the population are fused based on the adaptive fusion factor.

Drawings

FIG. 1 shows an adaptive fusion factor according to the present invention.

Detailed Description

To facilitate an understanding of the invention, the invention will now be described more fully with reference to the accompanying drawings. Preferred embodiments of the present invention are shown in the drawings. This invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

Referring to fig. 1, the present invention provides a photovoltaic power prediction method based on a differential evolution algorithm, including the following steps:

step 1, collecting a photovoltaic power data set from a photovoltaic power station operation and maintenance system;

step 2, preprocessing the collected photovoltaic power data set;

step 3, dividing the preprocessed photovoltaic power data set into a training data set and a testing data set;

step 4, inputting the size GPZ of the population and the maximum search algebra gMaxT;

step 5, setting a current search algebra t =0;

step 6, randomly generating a population Pop = { GB = ₁ ,GB ₂ ,...,GB _bi ,...,GB _GPZ In which GB _bi Represents the second individual in the population, and individual GB _bi ＝[GB _bi，1 GB _bi,2 GB _bi，3 ]Three training parameters, namely GB, of the support vector machine are stored _bi，1 Represents the penalty coefficient C, GB _bi,2 Representing the coefficients of the kernel function σ, and GB _bi,3 A coefficient epsilon representing an insensitive loss function; individual subscript bi =1,2, ·, GPZ;

step 7, calculating the adaptive values of all individuals in the population;

step 8, finding out the individual with the minimum adaptation value in the population and recording as the optimal individual MinB;

step 9, setting inventory searching step length HF _bi = rand (0,1), where rand denotes a random real number generation function;

step 10, setting inventory crossing rate HR _bi ＝rand(0,1)；

Step 11, calculating the current search step CF according to the formula (1) _bi ：

Wherein tpr1 is a random real number between [0,1 ];

step 12, calculating the current crossing rate CR according to the formula (2) _bi ：

Wherein tpr2 is a random real number between [0,1 ];

step 13, calculating an adaptive fusion factor cw according to the formula (3):

step 14, randomly selecting three individuals from the population: GB _ri1 ，GB _ri2 And GB _ri3 ；

Step 15, individual GB _ri1 ，GB _ri2 And GB _ri3 Recording the excellent individual with the minimum medium adaptive value as LMB;

step 16, performing inverse variation operation according to the formula (4) to generate varied individual MU _bi ：

NU _bi ＝(1-cw)×[(PL+PU)-LMB]+cw×MinB+CF _bi ×(GB _bi -GB _ri2 ) (4)

Wherein PL is the lower search boundary of the population, and PU is the upper search boundary of the population;

step 17, generating variant individual NU by performing variant operation according to formula (5) _bi ：

NU _bi ＝(1-cw)×LMB+cw×MinB+CF _bi ×(GB _bi -GB _ri2 ) (5)

Step 18, generating the test individual TV by performing the cross operation according to the formula (6) _bi ：

Wherein, MU _bi,hj Representing variant individuals MU _bi The hj training parameter of the support vector machine stored in the storage device; GB _bi,hj Representing individual GB _bi The hj training parameter of the support vector machine stored in the memory; TV (television) _bi,hj Representing test individuals TV _bi The hj training parameter of the support vector machine stored in the storage device; dimension subscript hj =1,2,3;

step 19, generating individual XV according to the cross operation executed by formula (7) _bi ：

Wherein, NU _bi,hj Representing variant individual NU _bi The hj training parameter of the support vector machine stored in the storage device; XV _bi,hj Representing test individuals XV _bi The hj training parameter of the support vector machine stored in the storage device;

step 20, calculating the TV of the test individual _bi And test subject XV _bi An adaptation value of;

step 21, if the individual TV is tested _bi Is less than the individual XV _bi The adapted value of (1) then sets the successful individual EV _bi ＝TV _bi Otherwise, setting a successful individual EV _bi ＝XV _bi ；

Step 22, if the individual EV is successful _bi Adapted value of less than individual GB _bi Go to step 23, otherwise go to step 26;

step 23, utilizing the successful individual EV _bi Replacing individual GB in a population _bi ；

Step 24, set inventory search step size HF _bi ＝CF _bi ；

Step 25, setting inventory crossing rate HR _bi ＝CR _bi ；

Step 26, finding out the individual with the minimum adaptation value in the population and recording the individual as the optimal individual MinB;

step 27, setting a current search algebra t = t +1;

step 28, if the current search algebra t is less than gMaxT, go to step 11, otherwise go to step 29;

and 29, extracting three training parameters of the support vector machine from the optimal individual MinB, and then training the support vector machine for predicting the photovoltaic power on a training data set by using the three training parameters of the support vector machine to realize photovoltaic power prediction.

For example, in one embodiment, the following steps are performed in accordance with the present invention with reference to the accompanying drawings:

step 1, collecting a photovoltaic power data set from a photovoltaic power station operation and maintenance system; the photovoltaic power data set includes, but is not limited to, time, altitude, air temperature, humidity, irradiance, wind direction, wind speed, air pressure, photovoltaic module temperature, and measured photovoltaic power data;

step 2, preprocessing the collected photovoltaic power data set;

step 3, dividing the preprocessed photovoltaic power data set into a training data set and a testing data set;

step 4, inputting a population size GPZ =50, and inputting a maximum search algebra gMaxT =2000;

step 5, setting a current search algebra t =0;

step 6, randomly generating a population Pop = { GB = ₁ ,GB ₂ ,...,GB _bi ,...,GB _GPZ In which GB _bi Represents the second individual in the population, and individual GB _bi ＝[GB _bi，1 GB _bi,2 GB _bi，3 ]Three training parameters, namely GB, of the support vector machine are stored _bi，1 Represents the penalty coefficient C, GB _bi,2 Representing the coefficients of the kernel function σ, and GB _bi,3 A coefficient epsilon representing an insensitive loss function; individual subscript bi =1,2, ·, GPZ;

step 7, calculating the adaptive values of all individuals in the population; the calculation process of the adaptive value is as follows: for the second individual GB in the population _bi First from individual GB _bi Three training parameters of the support vector machine are extracted: a penalty coefficient C, a kernel function coefficient sigma and a coefficient epsilon of an insensitive loss function, and then training a support vector machine PVM for predicting the photovoltaic power on a training data set by utilizing three training parameters of the support vector machine _bi And calculating and predicting the PVM of the photovoltaic power _bi Error in the test data set GERr _bi Setting an individual GB _bi Adaptation value of (1) GERr _bi (ii) a The PVM (support vector machine) for predicting photovoltaic power _bi The input variables are air temperature, humidity, irradiance, wind direction, wind speed, air pressure, photovoltaic module temperature and photovoltaic power at the current moment, and the support vector machine PVM for predicting the photovoltaic power _bi The output variable of (a) is the photovoltaic power at the next moment；

Step 8, finding out the individual with the minimum adaptive value in the population and recording the individual as the optimal individual MinB;

step 9, setting inventory search step length HF _bi = rand (0,1), where rand denotes a random real number generation function;

step 10, setting inventory crossing rate HR _bi ＝rand(0,1)；

Step 11, calculating the current search step CF according to the formula (1) _bi ：

Wherein tpr1 is a random real number between [0,1 ];

step 12, calculating the current crossing rate CR according to the formula (2) _bi ：

Wherein tpr2 is a random real number between [0,1 ];

step 13, calculating the adaptive fusion factor cw according to the formula (3):

step 14, randomly selecting three individuals from the population: GB _ri1 ，GB _ri2 And GB _ri3 ；

Step 15, the individual GB _ri1 ，GB _ri2 And GB _ri3 Recording the excellent individual with the minimum medium adaptive value as LMB;

step 16, performing inverse variation operation according to the formula (4) to generate varied individual MU _bi ：

MU _bi ＝(1-cw)×[(PL+PU)-LMB]+cw×MinB+CF _bi ×(GB _bi -GB _ri2 ) (4)

Wherein PL is the lower search boundary of the population, and PU is the upper search boundary of the population;

step 17, generating variant individual NU by performing variant operation according to formula (5) _bi ：

MU _bi ＝(1-cw)×LMB+cw×MinB+CF _bi ×(GB _bi -GB _ri2 ) (5)

Step 18, generating the test individual TV by performing the cross operation according to the formula (6) _bi ：

Wherein, MU _bi,hj Representing variant individuals MU _bi The hj training parameter of the support vector machine stored in the storage device; GB _bi,hj To represent Individual GB _bi The hj training parameter of the support vector machine stored in the storage device; TV (television) _bi,hj Representing test individuals TV _bi The hj training parameter of the support vector machine stored in the storage device; dimension subscript hj =1,2,3;

step 19, generating individual XV according to the cross operation executed by formula (7) _bi ：

Wherein, NU _bi,hj Representing variant individual NU _bi The hj training parameter of the support vector machine stored in the storage device; XV _bi,hj Representing test individuals XV _bi The hj training parameter of the support vector machine stored in the storage device;

step 20, calculating the TV of the test individual _bi And test subject XV _bi An adaptation value of;

step 21, if the individual TV is tested _bi Is less than the individual XV _bi The adapted value of (1) then sets the successful individual EV _bi ＝TV _bi Otherwise, setting a successful individual EV _bi ＝XV _bi ；

Step 22, if the individual EV is successful _bi Adapted value of less than individual GB _bi Go to step 23, otherwise go to step 26;

step 23, utilizing the successful individual EV _bi Replacing individual GB in a population _bi ；

Step 24, set inventory search step size HF _bi ＝CF _bi ；

Step 25, setting inventory crossing rate HR _bi ＝CR _bi ；

Step 26, finding out the individual with the minimum fitness value in the population and recording the individual as the optimal individual MinB;

step 27, setting a current search algebra t = t +1;

step 28, if the current search algebra t is less than gMaxT, go to step 11, otherwise go to step 29;

and 29, extracting three training parameters of the support vector machine from the optimal individual MinB, and then training the support vector machine for predicting the photovoltaic power on a training data set by using the three training parameters of the support vector machine, namely realizing the photovoltaic power prediction.

The photovoltaic power prediction method based on the differential evolution algorithm utilizes the support vector machine to construct a photovoltaic power prediction model, encodes training parameters of the support vector machine into individuals in a population, and designs the differential evolution algorithm based on the adaptive reverse variation to optimize the training parameters of the support vector machine. In the differential evolution algorithm of the adaptive reverse variation, an adaptive fusion factor is firstly calculated, then a good individual is selected from a population, the reverse information of the good individual and the information of the optimal individual in the population are fused based on the adaptive fusion factor, namely, the reverse information of the good individual and the information of the optimal individual in the population are fused according to the adaptive fusion factor to obtain a basic individual, the basic individual is utilized to execute the reverse variation operation to generate a variant individual, a cross operation is executed to generate a test individual, and then a selection operation is executed to win or lose the population. The method comprehensively utilizes the reverse information of the excellent individuals and the information of the optimal individuals in the population to enhance the optimization performance of the differential evolution algorithm, and can improve the accuracy of photovoltaic power prediction.

The above examples only show the embodiments of the present invention, and the description thereof is specific and detailed, but not to be construed as limiting the scope of the present invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the inventive concept, which falls within the scope of the present invention. Therefore, the protection scope of the present patent shall be subject to the appended claims.

Claims (1)

1. A photovoltaic power prediction method based on a differential evolution algorithm is characterized by comprising the following steps:

step 1, collecting a photovoltaic power data set from a photovoltaic power station operation and maintenance system;

step 2, preprocessing the collected photovoltaic power data set;

step 3, dividing the preprocessed photovoltaic power data set into a training data set and a testing data set;

step 4, inputting the size GPZ of the population and the maximum search algebra gMaxT;

step 5, setting a current search algebra t =0;

step 6, randomly generating a population Pop = { GB = ₁ ,GB ₂ ,...,GB _bi ,...,GB _GPZ In which GB _bi Represents the second individual in the population, and individual GB _bi ＝[GB _bi，1 GB _bi,2 GB _bi，3 ]Three training parameters, namely GB, of the support vector machine are stored _bi，1 Represents the penalty coefficient C, GB _bi,2 Representing the coefficients of the kernel function σ, and GB _bi,3 A coefficient epsilon representing an insensitive loss function; individual subscripts bi =1,2., GPZ;

step 7, calculating the adaptive values of all individuals in the population;

step 8, finding out the individual with the minimum adaptation value in the population and recording as the optimal individual MinB;

step 9, setting inventory search step length HF _bi = rand (0,1), where rand denotes a random real number generation function;

step 10, setting inventory crossing rate HR _bi ＝rand(0,1)；

Step 11, calculating according to the formula (1)Current search step CF _bi ：

Wherein tpr1 is a random real number between [0,1 ];

step 12, calculating the current crossing rate CR according to the formula (2) _bi ：

Wherein tpr2 is a random real number between [0,1 ];

step 13, calculating the adaptive fusion factor cw according to the formula (3):

step 14, randomly selecting three individuals from the population: GB _ri1 ，GB _ri2 And GB _ri3 ；

Step 15, individual GB _ri1 ，GB _ri2 And GB _ri3 Recording the excellent individual with the minimum medium adaptive value as LMB;

step 16, performing inverse variation operation according to the formula (4) to generate varied individual MU _bi ：

NU _bi ＝(1-cw)×[(PL+PU)-LMB]+cw×MinB+CF _bi ×(GB _bi -GB _ri2 ) (4)

Wherein PL is the lower search boundary of the population, and PU is the upper search boundary of the population;

step 17, generating variant individual NU by performing variant operation according to formula (5) _bi ：

NU _bi ＝(1-cw)×LMB+cw×MinB+CF _bi ×(GB _bi -GB _ri2 ) (5)

Step 18, generating the test individual TV by performing the crossover operation according to the formula (6) _bi ：

Wherein, MU _bi,hj Representing variant individuals MU _bi The hj training parameter of the support vector machine stored in the storage device; GB _bi,hj Representing individual GB _bi The hj training parameter of the support vector machine stored in the storage device; TV (television) _bi,hj Representing test individuals TV _bi The hj training parameter of the support vector machine stored in the storage device; dimension subscript hj =1,2,3;

step 19, generating individual XV according to the cross operation executed by formula (7) _bi ：

Wherein, NU _bi,hj Representing variant individual NU _bi The hj training parameter of the support vector machine stored in the storage device; XV _bi,hj Representing test individuals XV _bi The hj training parameter of the support vector machine stored in the storage device;

step 20, calculating the TV of the test individual _bi And test subject XV _bi An adaptation value of;

step 21, if the individual TV is tested _bi Is less than the individual XV _bi The adapted value of (1) then sets the successful individual EV _bi ＝TV _bi Otherwise, setting a successful individual EV _bi ＝XV _bi ；

Step 22, if the individual EV is successful _bi Adapted value of less than individual GB _bi Go to step 23, otherwise go to step 26;

step 23, utilizing the successful individual EV _bi Replacing individual GB in a population _bi ；

Step 24, setting inventory searching step length HF _bi ＝CF _bi ；

Step 25, setting inventory crossing rate HR _bi ＝CR _bi ；

Step 26, finding out the individual with the minimum adaptation value in the population and recording the individual as the optimal individual MinB;

step 27, setting a current search algebra t = t +1;

step 28, if the current search algebra t is less than gMaxT, go to step 11, otherwise go to step 29;

and 29, extracting three training parameters of the support vector machine from the optimal individual MinB, and then training the support vector machine for predicting the photovoltaic power on a training data set by using the three training parameters of the support vector machine to realize photovoltaic power prediction.

Images