CN115965132A - Power prediction method for distributed photovoltaic digital twin system based on GA-BP neural network - Google Patents

Abstract

The invention discloses a distributed photovoltaic digital twin system power prediction method based on a GA-BP neural network, which comprises the following steps: s1, constructing a digital twin system for distributed photovoltaic power generation prediction; s2, quantifying a photovoltaic operation state and analyzing photovoltaic output fluctuation based on the actual power generation amount of the distributed photovoltaic power generation system; and S3, building a photovoltaic power generation power prediction model based on the GA-BP neural network in the digital twin system based on the quantitative analysis result, and performing power prediction. The distributed photovoltaic power generation digital twin system is constructed, so that a stable environment can be provided for photovoltaic cluster access capability evaluation and power accurate prediction; a genetic algorithm is adopted to optimize the weight and the threshold on the basis of the BP photovoltaic power generation power prediction model, so that the accuracy of power prediction can be greatly improved.

Description

Power prediction method of distributed photovoltaic digital twin system based on GA-BP neural network

technical field

The invention belongs to the technical field of distributed photovoltaic power prediction, and in particular relates to a power prediction method of a distributed photovoltaic digital twin system based on a GA-BP neural network.

Background technique

Today, the world has gradually entered the era of renewable energy. China has proposed a "dual-carbon" energy development plan and a future power grid construction goal of "creating a new power system with new energy as the core". At present, China's distributed solar power grid tends to expand campuses and regions, but large-scale, clustered distributed photovoltaic power grids are becoming an important development route. On the other hand, a large number of intermittent and unpredictable photovoltaic power sources are scattered into the grid, which greatly increases the complexity of the grid and the difficulty of regulation. In order to complete regulation, the output power change block of distributed photovoltaic power generation must be combined with panoramic data to achieve effective prediction and control. How to ensure that distributed photovoltaic power generation can be connected to the power grid in an orderly, safe, reliable, adaptable and efficient manner has become a basic scientific issue in the field of energy and power grids.

At present, some scholars have carried out related research on the power prediction of distributed photovoltaics, and have achieved certain research results. The first method takes the aggregated energy system as the research object, and adopts a scenario-based approach to study the impact of the number and type of inputs on the performance of power generation prediction by comprehensively considering the total capacity of photovoltaic and wind turbine power generation in a country. However, this method does not study the respective characteristics of photovoltaic and wind power generation. The second method is an excellent hybrid neural network ultra-short-term photovoltaic power prediction method based on cloud image feature extraction, as a solution to the serious power minute variation problem that may be caused by the unpredictability and instability of photovoltaic power output. These variations may be unexpected and erratic results of photovoltaic power generation output. However, this approach establishes a direct mapping between the historical data it receives and the forecast data it produces. This mapping does not take into account the temporal relationship between data series. Another approach is to divide the short-term photovoltaic power generation forecasting method into two processes: using the K-means algorithm to group different kinds of weather, and using EEDM to deconstruct the photovoltaic output power from the data generated by the clustering. This method uses the MFA-Elman neural network. This strategy does not guarantee that accurately predicted needs will continue to be met in a reliable manner as time, environmental and operating conditions continue to evolve.

Contents of the invention

In view of the above shortcomings in the prior art, the distributed photovoltaic digital twin system power prediction method based on GA-BP neural network provided by the present invention solves the problems of low accuracy, long time consumption and large error of the traditional distributed photovoltaic power prediction method.

In order to achieve the purpose of the above invention, the technical solution adopted in the present invention is: a distributed photovoltaic digital twin system power prediction method based on GA-BP neural network, including the following steps:

S1. Construct a digital twin system for distributed photovoltaic power generation prediction;

S2. Based on the actual power generation of the distributed photovoltaic power generation system, quantify the photovoltaic operation status and analyze the fluctuation of photovoltaic output;

S3. Based on the quantitative analysis results, build a photovoltaic power prediction model based on the GA-BP neural network in the digital twin system, and perform power prediction.

Further, the digital twin system for distributed photovoltaic power generation prediction in step S1 includes a physical layer, a perception layer, a data transmission layer, a data processing layer, and a decision-making layer;

The physical layer is used to obtain its power data, operating data, operating environment parameters and photovoltaic array installation data from photovoltaic power generation facilities, and transmit them to the perception layer;

The sensing layer includes weather stations and sensors positioned on the photovoltaic array for collecting weather data;

The data transmission layer stores data in a hybrid manner of local distributed storage and centralized cloud storage, and is used to realize data transmission based on a wireless network;

The data processing layer is used to calculate the initial value of the photovoltaic power generation prediction according to the collected data, and use it as the input of the photovoltaic power generation prediction model;

The decision-making layer is used to evaluate the prediction results of the photovoltaic power generation prediction model and generate a photovoltaic grid-connected strategy; the decision-making layer is also used to provide operation and maintenance instructions for photovoltaic power generation equipment according to the collected data.

Further, in the step S2, the photovoltaic operating state is quantified as photovoltaic output P(t) at different times, and its calculation formula is:

In the formula, B( ) is the Beta distribution function, a _L , b _L are the Beta distribution functions under the cloud state, F _0t is the standard light intensity, and P _N is the rated power of photovoltaics.

Further, in the step S2, the method for analyzing fluctuations in photovoltaic output is specifically:

According to the variation characteristics of the power generation of the distributed photovoltaic power generation system in the daily cycle, determine the daily output characteristics of photovoltaic output fluctuations;

The power generation change characteristics of the distributed photovoltaic power generation system satisfy:

In the formula, P(t) is the photovoltaic power generation power at sampling time t, and P _S is the total installed capacity of the photovoltaic power station.

Further, the photovoltaic power generation prediction model in step S3 includes an input layer, a hidden layer and an output layer that are sequentially connected, and the initial connection weight and threshold are changed through the GA algorithm between the input layer, the hidden layer and the output layer;

The input of the photovoltaic power prediction model is the normalized solar radiation intensity, temperature and humidity, and the output is the photovoltaic power generation level.

Further, the method of changing the initial connection weight and threshold through the GA algorithm is specifically as follows:

S31. Randomly start the population, and encode the neurons in the photovoltaic power generation prediction model by means of real number encoding;

S32. Calculate the fitness value of the current population;

S33. According to the fitness value, the real number crossover method is used to update the population of the neurons as a genetic operator;

S34. Steps S32-S33 are repeated until the fitness value of the population is less than a preset threshold;

S35. Use the connection weight and threshold of the current neuron as the initial connection weight and threshold of the photovoltaic power generation prediction model.

Further, in the step S31, the size of the total population is greater than the number of all neurons contained in the photovoltaic power prediction model, and each member of the population has a corresponding connection weight and threshold during encoding .

Further, in the step S32, the fitness function e(x) for calculating the fitness value is:

In the formula, Y _0i expected output value, Y ₁ and Y ₂ are the predicted output values of hidden layer and output layer respectively;

The adaptation function e(x) satisfies the following formula:

In the formula, b is the bounded conservative estimate of the fitness function e(x), and satisfies b≥0, b-e(x)≥0.

Further, in the step S33, the real number crossover method means that the m-th chromosome A _m and the n-th chromosome A _n perform a crossover operation at the i-position, wherein k is a random number in [0,1], and at the same time, the i-th chromosome Mutation operation is performed on the mth gene of chromosome Ai;

Among them, the cross operation is expressed as:

The mutation operation is expressed as:

In the formula, _LT and L _B represent the upper and lower bounds of the gene, respectively, h(u) is a coefficient representing the number of iterations, and λ is a random number in [0,1].

The beneficial effects of the present invention are:

(1) The present invention uses digital twin technology to establish a high-precision, multi-coupled digital twin power grid model for the case of distributed photovoltaic power generation clusters connected to the power grid, and then incorporates the operation history of the power grid into the simulation process of the physical model, thereby realizing the distribution Accurate prediction of photovoltaic power generation;

(2) The present invention constructs a distributed photovoltaic power generation digital twin system that can provide a stable environment for photovoltaic cluster access capability assessment and power accurate prediction;

(3) In the prediction process, the present invention uses a genetic algorithm to optimize the weight and threshold based on the BP photovoltaic power prediction model, which can greatly improve the accuracy of power prediction.

Description of drawings

Fig. 1 is a flow chart of the power prediction method for distributed photovoltaic digital twin system based on GA-BP neural network provided by the present invention.

Fig. 2 is a schematic diagram of the trend of the power output curve of the photovoltaic power plant provided by the present invention.

Fig. 3 is a schematic diagram of sunny weather prediction results provided by the present invention.

Fig. 4 is a schematic diagram of cloudy day prediction results provided by the present invention.

Fig. 5 is a schematic diagram of rainy weather prediction results provided by the present invention.

Detailed ways

The specific embodiments of the present invention are described below so that those skilled in the art can understand the present invention, but it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art, as long as various changes Within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions and creations using the concept of the present invention are included in the protection list.

Example 1:

The embodiment of the present invention provides a distributed photovoltaic digital twin system power prediction method based on GA-BP neural network, as shown in Figure 1, including the following steps:

S1. Construct a digital twin system for distributed photovoltaic power generation prediction;

S2. Based on the actual power generation of the distributed photovoltaic power generation system, quantify the photovoltaic operation status and analyze the fluctuation of photovoltaic output;

S3. Based on the quantitative analysis results, build a photovoltaic power prediction model based on the GA-BP neural network in the digital twin system, and perform power prediction.

In step S1 of the embodiment of the present invention, when building a digital twin system, the underlying infrastructure must be able to support digital and physical components, and exchange data in two ways, so as to realize the digital twin. The digital twin system for distributed photovoltaic power generation prediction in this embodiment includes a physical layer, a perception layer, a data transmission layer, a data processing layer, and a decision-making layer;

The physical layer is used to obtain its power data, operating data, operating environment parameters and photovoltaic array installation data from photovoltaic power generation facilities, and transmit them to the perception layer;

The perception layer includes weather stations and sensors positioned on the photovoltaic array to collect weather data, including solar radiation, temperature, humidity and wind speed data;

The data transmission layer uses a hybrid method of local distributed storage and centralized cloud storage to store data for wireless network-based data transmission; switches and Ethernet form the basis of the network, enabling a large amount of data to move effectively, including equipment operational and weather information;

The data processing layer is used to calculate the initial value of the photovoltaic power generation forecast based on the collected data, and it is used as the input of the photovoltaic power generation prediction model; the data processing layer is the key to realize the photovoltaic power generation forecast, it can be used as the decision As the core of the digital twin of the forecasting system, this layer is crucial for forecasting photovoltaic power generation; in order to generate the final value predicted by the digital twin, the initial value of photovoltaic power forecast should be modified according to the previous meteorological data compensation. ;

The decision-making layer is used to evaluate the prediction results of the photovoltaic power generation prediction model, and generate a photovoltaic grid-connected strategy, which is then sent back to the terminal equipment, and the terminal equipment uses it to manage power grid scheduling; the decision-making layer is also used to The data provides operation and maintenance instructions for photovoltaic power generation equipment to ensure the continuous normal operation of the photovoltaic power generation system.

In step S2 of the embodiment of the present invention, when quantifying the photovoltaic operating state, since the photovoltaic power generation is in a continuous state, in order to accurately estimate, the quantity in the continuous state must first be quantified and then converted into a state quantity. In order to ensure the accuracy of state quantity division, the state is divided according to the concept of equal frequency of data determined by historical data statistics. When it comes to the station area, the historical load statistical data of each station area of the distribution network is classified according to the size. When the load is divided into M states, the corresponding data will also be divided into M intervals, each interval corresponds to a different load state, and the lower limit and upper limit of the load in an interval are determined by the interval of the state. By splitting the state in this way, we can ensure that each state can be accessed, avoiding the blindness caused by state separation. The probability density of the load data in each state interval conforms to the normal distribution, and the specific state parameters of the normal distribution can be determined by calculating the data of each state interval to determine the load of the state.

Since photovoltaic output depends on solar radiation values, which are regulated by cloud cover, random changes in cloud cover lead to variations in photovoltaic output. Therefore, using cloud cover as a unified division of PV output states is a possibility to ensure the continuity of PV state transitions and reduce the need for individual characterization of each PV, with the cloud cover level (CCL) index representing the cloud state changes, as shown in the following formula:

In the formula, g _C (t) is the cloud cover level at any time in a day, F _t is the irradiance at the time, and F _0t is the reference irradiance on sunny days at the time, which can be corrected according to different seasons.

According to different cloud states, the probability density function of photovoltaic output is established as follows:

In the formula: Γ( ) is the gamma function; x _P =F _t /F _0t ; a and b can be calculated by the average value λ and standard deviation η of the ratio of the historical illumination data to the sunny reference irradiance, where a=b( 1-η)/η, b=a(1-λ)/λ.

In the above formula, several cloud layers are separated, and the irradiance statistics of each cloud state are calculated. Finally, by integrating these information with the baseline irradiance of sunny days, the photovoltaic output at different times is calculated. Therefore, in step S2 in this embodiment, the photovoltaic operating state is quantified as photovoltaic output P(t) at different times , whose calculation formula is:

In the formula, B( ) is the Beta distribution function, a _L , b _L are the Beta distribution functions under the cloud state, F _0t is the standard light intensity, and P _N is the rated power of photovoltaics.

In step S2 of the embodiment of the present invention, the power generation of the photovoltaic power generation system usually changes in a daily cycle. During the day, the power generation is at a peak, but it decreases to almost nothing at night. It is possible that there will be negative electricity throughout the night. The main reason is that the photovoltaic power station must draw electricity from the main grid at night. As shown in Figure 2, normal distribution can be used to represent the daily characteristics of photovoltaic power generation, and this distribution can fit the power output curve trend of photovoltaic power plants at specific locations; therefore, in step S2 of this embodiment, according to distributed photovoltaic power generation The change characteristics of the power generation of the system in the daily cycle determine the daily output characteristics of photovoltaic output fluctuations;

The power generation change characteristics of the distributed photovoltaic power generation system satisfy:

In the formula, P(t) is the photovoltaic power generation power at sampling time t, and P _S is the total installed capacity of the photovoltaic power station.

In the embodiment of the present invention, the complex BP neural network is more prevalent in the field of solar power generation forecasting. If the initial value and threshold of the typical BP neural network are not fully adjusted, the model will have a phenomenon of delayed convergence and will not recover to Tendency to a local optimum. As a direct result, the accuracy of photovoltaic generation forecasts does not meet the basic characteristics. In the embodiment of the present invention, genetic algorithm (GA) is used to determine the optimal weight and threshold performance. This method is based on the BP photovoltaic power generation prediction model; if the number of variables forming the input layer of the BP neural network prediction model is too large, it will almost certainly This leads to an increase in model complexity and a delay in convergence, but this has no effect on the accuracy of predictions. Therefore, the photovoltaic power generation prediction model in step S3 of this embodiment includes an input layer, a hidden layer, and an output layer that are sequentially connected, and the initial connection weight and threshold are changed through the GA algorithm between the input layer, the hidden layer, and the output layer;

In this embodiment, the number of neurons in the hidden layer will affect the ability of the BP neural network prediction model to process data and information. If the number of neurons in the hidden layer is too large, the calculation of the model and solution will become complicated , if there are too few, the data processing will not be precise enough to meet the standard. As a result of applying Kolmogorov's hidden layer neuron selection criteria to this scheme, the total number of hidden layer neurons in this embodiment is eight.

The input of the photovoltaic power generation prediction model in this embodiment is the normalized solar radiation intensity, temperature and humidity, and its output is the photovoltaic power generation level; wherein, the normalized formula is:

In the formula, Y represents the processed data, X represents the original data, and X _min and X _max represent the maximum and minimum values of the data, respectively.

In the embodiment of the present invention, the method of changing the initial connection weight and threshold through the GA algorithm is specifically as follows:

S31. Randomly start the population, and encode the neurons in the photovoltaic power generation prediction model by means of real number encoding;

S32. Calculate the fitness value of the current population;

S33. According to the fitness value, the real number crossover method is used to update the population of the neurons as a genetic operator;

S34. Steps S32-S33 are repeated until the fitness value of the population is less than a preset threshold;

S35. Use the connection weight and threshold of the current neuron as the initial connection weight and threshold of the photovoltaic power generation prediction model.

In step S31 of this embodiment, the size of the total population is greater than the number of all neurons included in the photovoltaic power prediction model, and when encoding, each member of the population has a corresponding connection weight and threshold .

In step S32 of this embodiment, determine who in the group has the highest degree of fitness by calculating the fitness value, and then determine its connection weight and threshold; in this embodiment, the fitness function e(x) for calculating the fitness value is :

e(x)＝|Y _0i -Y _i |,i∈{1,2}

In the formula, Y _0i expected output value, Y ₁ and Y ₂ are the predicted output values of hidden layer and output layer respectively;

The adaptation function e(x) satisfies the following formula:

In the formula, b is the bounded conservative estimate of the fitness function e(x), and satisfies b≥0, b-e(x)≥0.

In step S33 of this embodiment, when neurons are used as genetic operators, the genetic operator is a key component of the genetic algorithm, which simulates natural selection and survival of the fittest in biological evolution. This stage is required to fully implement the program. In this case, a tournament selection procedure is employed, where the original population of neurons is randomly selected to compete in a series of "tournaments", and the offspring population is then composed of the neurons with the greatest fitness value among the winners of each tournament composition. In order to combine the genes of any two chromosomes, the real number crossover method is used in this embodiment to update the population. The real number crossover method refers to the crossover operation between the mth chromosome A _m and the nth chromosome A _n at the i position, where k is A random number of [0,1], while performing a mutation operation on the m-th gene of the i-th chromosome A _i ;

Among them, the cross operation is expressed as:

The mutation operation is expressed as:

In the formula, _LT and L _B represent the upper and lower bounds of the gene, respectively, h(u) is a coefficient representing the number of iterations, and λ is a random number in [0,1].

In step S34 of this embodiment, during the iterative execution of steps S32-S33, the next-generation population is generated through a genetic algorithm, and then checked for errors. If the total network error meets the accuracy requirement, the method is complete; otherwise, the population is used as the parent population, and the previous process is repeated until the error meets the accuracy requirement, which is expressed as e < ε;

Example 2:

This embodiment provides a simulation experiment comparison between the prediction method in embodiment 1 and the other two methods under the same conditions;

The photovoltaic data in this example are selected from the open source website of the University of Queensland and the photovoltaic power station of a renewable energy start-up company in Chongqing. The source domain is selected from the center of the University of Queensland, plus the three-year historical data of the ST Lucia site (2018.01-2021.12), The target domain is selected from the QBP site (2018.02-2019.02), the AB site (2019.01-2020.02), and the photovoltaic power station (2012.01). Since the solar cell does not generate electricity at night, only the data between 7:00 and 19:00 every day is of interest in this embodiment. Jinnen's open source website and database has photovoltaic output power and related climate data, including solar irradiance, temperature, relative humidity, wind speed and wind direction, both with a time resolution of 1 minute, 70% of the data as a training set, 30 % of the data is used as a test set, and the photovoltaic power generation is affected by solar irradiance, temperature, relative humidity and wind speed.

The weather data of the test day is used as the input of the prediction model to obtain the predicted value of the GA-BP neural network; then, the corresponding weather data is uploaded to the historical database for comparative analysis; the actual value and Prediction value and error compensation to obtain the final digital twin prediction value; finally, compare the prediction value with the actual value of photovoltaic power generation to obtain the final digital twin prediction value. The forecast results of sunny days, cloudy days and rainy days are shown in Figures 3, 4 and 5, respectively.

It can be seen from Fig. 3-Fig. 5 that the distributed photovoltaic digital twin system power prediction method based on GA-BP neural network proposed by the present invention is consistent with the actual output power trend in cloudy and sunny days; The trend of the forecast curve is basically consistent with the actual output power, but there is a certain range of deviation fluctuations in some forecast periods, which is due to the short period of forecast. Thus, a negative compensating effect arises. However, the weather variation in rainy days can be kept stable within a certain range most of the time, and the historical data has a positive compensatory effect. Obviously, the proposed method can accurately predict the photovoltaic power generation under three different weather conditions.

In order to reflect the superiority of the distributed photovoltaic prediction method of the present invention, the method of the present invention is compared with existing two methods (the ultra-short-term photovoltaic power prediction method M1 of the hybrid neural network based on cloud feature extraction and the use of K-means algorithm for different types Grouping the weather, and using EEDM to deconstruct the photovoltaic output power MFA-Elman neural network prediction method M2) from the data generated by clustering for comparative analysis, using different methods to predict the power of distributed photovoltaic power generation, and the prediction results are as follows As shown in Table 1;

Table 1: Statistics on the prediction results of distributed photovoltaic power generation by different methods

It can be seen from Table 1 that under three different weather environments, the time-consuming, accuracy and error standards of the distributed photovoltaic power prediction method proposed by the present invention are better than the other two comparative methods. The accuracy rate of the proposed algorithm can reach up to 95.24%, the minimum time consumption is 6.53s, the minimum RMSE is 5.32%, and the minimum MAE is 48.53W, which is greatly improved compared with the other two comparison algorithms. This is because the distributed photovoltaic digital twin system can realize all-round perception and network connection of photovoltaic power generation systems, providing a stable environment for power generation prediction. In addition, based on the BP photovoltaic power prediction model, the genetic algorithm is used to optimize the weight and threshold, which greatly improves the accuracy of power prediction. On the basis of the model, genetic algorithm is used to optimize the weight and threshold, which can greatly improve the accuracy of power prediction.

Claims (9)

1. A distributed photovoltaic digital twin system power prediction method based on a GA-BP neural network is characterized by comprising the following steps:

s1, constructing a digital twin system for distributed photovoltaic power generation prediction;

s2, quantifying a photovoltaic operation state and analyzing photovoltaic output fluctuation based on the actual power generation amount of the distributed photovoltaic power generation system;

and S3, building a photovoltaic power generation power prediction model based on the GA-BP neural network in the digital twin system based on the quantitative analysis result, and performing power prediction.

2. The GA-BP neural network-based distributed photovoltaic digital twin system power prediction method as claimed in claim 1, wherein the distributed photovoltaic power generation predicted digital twin system in step S1 comprises a physical layer, a sensing layer, a data transmission layer, a data processing layer and a decision layer;

the physical layer is used for acquiring electric power data, operation environment parameters and photovoltaic array installation data from a photovoltaic power generation facility and transmitting the electric power data, the operation environment parameters and the photovoltaic array installation data to the sensing layer;

the sensing layer comprises a weather station and a sensor which are positioned on the photovoltaic array and used for collecting weather data;

the data transmission layer stores data in a mixed mode of local distributed storage and centralized cloud storage and is used for realizing data transmission based on a wireless network;

the data processing layer is used for calculating an initial value of photovoltaic power generation prediction according to the collected data and taking the initial value as the input of a photovoltaic power generation power prediction model;

the decision layer is used for evaluating the prediction result of the photovoltaic power generation power prediction model and generating a photovoltaic grid-connected strategy; the decision layer is further used for providing operation and maintenance instructions of the photovoltaic power generation equipment according to the collected data.

3. The method for predicting the power of a distributed photovoltaic digital twin system based on a GA-BP neural network as claimed in claim 1, wherein in the step S2, the photovoltaic operation state is quantized into photovoltaic output P (t) at different times, and the calculation formula is as follows:

wherein B (-) is a Beta distribution function, a _L ,b _L Respectively Beta distribution function in cloud layer state, F _0t Is the standard illumination intensity, P _N Is the photovoltaic rated power.

4. The distributed photovoltaic digital twin system power prediction method based on the GA-BP neural network according to claim 1, wherein the photovoltaic output fluctuation analysis method in the step S2 specifically comprises:

determining the solar output characteristic of photovoltaic output fluctuation according to the change characteristic of the generated energy of the distributed photovoltaic power generation system in each day period;

the generated energy change characteristics of the distributed photovoltaic power generation system meet the following requirements:

wherein P (t) is the photovoltaic power generation power at the sampling time t, P _S The total installed capacity of the photovoltaic power station.

5. The distributed photovoltaic digital twin system power prediction method based on GA-BP neural network of claim 1, wherein the photovoltaic power generation power prediction model of step S3 comprises an input layer, a hidden layer and an output layer which are connected in sequence, and the initial connection weight and threshold are changed among the input layer, the hidden layer and the output layer through GA algorithm;

the input of the photovoltaic power generation power prediction model is the solar radiation intensity, the temperature and the humidity which are subjected to normalization processing, and the output of the photovoltaic power generation power prediction model is the photovoltaic power generation level.

6. The distributed photovoltaic digital twin system power prediction method based on the GA-BP neural network as claimed in claim 5, wherein the method for changing the initial connection weight and the threshold value by the GA algorithm specifically comprises:

s31, randomly initiating a population, and coding neurons in a photovoltaic power generation power prediction model in a real number coding mode;

s32, calculating the fitness value of the current population;

s33, according to the fitness value, a real number intersection method is adopted, and the neuron is used as a genetic operator to update the population;

s34, repeating the steps S32-S33 until the fitness value of the population is smaller than a preset threshold value;

and S35, taking the connection weight and the threshold of the current neuron as the initial connection weight and the threshold of the photovoltaic power generation power prediction model.

7. The method of claim 6, wherein in step S31, the size of the total population is larger than the total number of neurons in the photovoltaic power generation prediction model, and each member in the population has a corresponding connection weight and a threshold when encoding.

8. A distributed photovoltaic digital twin system power prediction method based on GA-BP neural network as claimed in claim 6, wherein in step S32, the fitness function e (x) for calculating the fitness value is:

e(x)＝|Y _0i -Y _i |,i∈{1,2}

in the formula, Y _0i Desired output value, Y ₁ And Y ₂ Prediction output values of a hidden layer and an output layer respectively;

the adaptation function e (x) satisfies the following equation:

in the formula, b is a conservative estimation value of the boundary of the adaptive function e (x), and b is more than or equal to 0,b-e (x) and more than or equal to 0.

9. The distributed photovoltaic digital twin system power prediction method based on GA-BP neural network of claim 8,wherein the real number cross method in step S33 is the mth chromosome A _m To the nth chromosome A _n Performing a crossover operation at position i, where k is [0,1]Simultaneously to the ith chromosome A _i Mutation operation is carried out on the mth gene;

wherein the interleaving operation is represented as:

the mutation operation is represented as:

in the formula, L _T And L _B Respectively representing the upper and lower bounds of the gene, h (u) is a coefficient representing the number of iterations, and lambda is [0,1 ]]The random number of (2).

Images