CN115952906A - Short-term photovoltaic power prediction method, system, equipment and medium based on LSGAN-GRU - Google Patents
Short-term photovoltaic power prediction method, system, equipment and medium based on LSGAN-GRU Download PDFInfo
- Publication number
- CN115952906A CN115952906A CN202211714868.9A CN202211714868A CN115952906A CN 115952906 A CN115952906 A CN 115952906A CN 202211714868 A CN202211714868 A CN 202211714868A CN 115952906 A CN115952906 A CN 115952906A
- Authority
- CN
- China
- Prior art keywords
- model
- gru
- photovoltaic power
- lsgan
- weather
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
A short-term photovoltaic power prediction method, system, device and medium based on LSGAN-GRU, the method comprising: collecting historical power generation data of a photovoltaic power station, and selecting weather variables with high correlation as input variables; classifying weather types based on the SOM self-organizing mapping neural network; screening 1-type sudden change weather sample data according to the weather type; constructing an MAD-LSGANs model based on the minimum second support of a multi-generator; constructing a GRU model consisting of a cross network and a depth network; determining the hyperparameter of the GRU model to obtain an optimal GRU model; systems, devices and media for implementing a short-term photovoltaic power prediction method based on an LSGAN-GRU network; the method realizes the prediction of the photovoltaic power of the sudden change weather by constructing the LSGAN-GRU model, and has the characteristics of improving the accuracy of the prediction result, reducing the prediction deviation and improving the convergence rate and accuracy.
Description
Technical Field
The invention relates to the field of photovoltaic power prediction methods, in particular to a short-term photovoltaic power prediction method, system, equipment and medium based on LSGAN-GRU.
Background
The fluctuation characteristic of solar energy is researched, accurate ultra-short-term photovoltaic power prediction is obtained within 1h or shorter time, a power grid department is facilitated to determine necessary standby power generation capacity by determining the shortage of actual power generation according to a predicted value, and the method has important significance for promoting a photovoltaic power station to be integrated into a power system and improving the safety and stability of a power grid. Generally, the photovoltaic power generation power prediction mainly comprises two main categories, namely a physical prediction model and a statistical model; the physical prediction model establishes a model according to the physical cause of photovoltaic power by using the solar irradiance prediction value and combining information such as the inclination angle of the photovoltaic cell panel and the position of the power station; the statistical model analyzes the influence of various external factors on the generated energy by relying on historical power data of the photovoltaic power station, and finally, an input and output mapping model is established to realize the prediction of the photovoltaic power; however, the prediction methods have certain limitations, the physical prediction method needs to rely on detailed power station and historical meteorological data information, deviation is easy to generate in the calculation process, and the prediction result is not ideal; the statistical model method is simpler and faster in modeling, but part of models are difficult to converge, most statistical models do not respectively predict the photovoltaic power in different weathers according to weather types, and the weather types have large influence on the photovoltaic power generation power, so that the statistical model method has the defects of large error of prediction results, inaccurate prediction and difficulty in convergence of the models.
The invention relates to a photovoltaic power generation power prediction method, a photovoltaic power generation power prediction device and photovoltaic power generation power prediction equipment integrating multiple weather conditions, and the patent number of CN110717623B, and provides a photovoltaic power generation power prediction method, wherein the prediction method comprises the following steps: respectively inputting weather variable data corresponding to power to be predicted into power predictors under different weather conditions to obtain a preset number of predicted power values; calculating to obtain photovoltaic power generation power fusing various weather conditions according to the obtained predicted power value and the weight value of each power predictor; wherein the power predictor is trained by the following method: and acquiring a training data set, dividing the training data set into a preset number of sample groups, and training untrained predictors by using each sample group to obtain the power predictors under different weather conditions. Because the influence of sudden change weather on power prediction is not considered, and particularly, the problem that sudden change weather data of most photovoltaic power stations are insufficient at present is solved, the method has the defect that a predictor cannot accurately predict when the photovoltaic power stations are in the scene of sudden change weather.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention aims to provide a short-term photovoltaic power prediction method, a system, equipment and a medium based on LSGAN-GRU, weather variables with high correlation are selected as input variables through Pearson analysis, weather types are classified based on an SOM neural network, MAD-LSGANs are generated by utilizing the least square of a multi-generator, an LSGAN-GRU model is obtained by fusing the MAD-LSGANs and an optimal GRU model, and the sudden change weather photovoltaic power is predicted through the LSGAN-GRU model.
In order to achieve the purpose, the invention adopts the technical scheme that:
a short-term photovoltaic power prediction method based on LSGAN-GRU comprises the following steps:
step 1: acquiring historical power generation data of a photovoltaic power station, acquiring the correlation between weather variable forecast data and photovoltaic power data based on Pearson correlation analysis, and selecting weather variables with high correlation as input variables;
step 2: classifying the weather types according to the irradiation intensity based on the SOM self-organizing mapping neural network;
and 3, step 3: screening 1-type sudden change weather sample data for the photovoltaic data sequence under each weather type according to the weather type;
and 4, step 4: constructing an MAD-LSGANs model based on the least square of a multi-generator;
and 5: constructing a GRU model consisting of a cross network and a depth network;
step 6: determining the hyper-parameters of the GRU model by adopting a backbone differential evolution IBBDE algorithm so as to obtain an optimal GRU model;
and 7: and constructing an LSGAN-GRU model to obtain a photovoltaic short-term power generation power prediction result in sudden change weather.
The specific process of the step 2 is as follows: the SOM self-organizing mapping neural network comprises an input layer and a computing layer, neurons of the computing layer of the SOM self-organizing mapping neural network are classified into different response areas through self-organizing learning, input variables are automatically classified when passing through the computing layer, and clustering results output through the computing layer in a classified mode are sunny days, cloudy days and rainy days.
The specific process of the step 4 is as follows:
4.1, constructing the discrimination model D by adopting a full-connection neural network;
4.2, d one-dimensional random noises with the length of c are input into the generation model G, and a pseudo sample is obtained;
4.3, fixing parameters of the generated model G, training a discrimination model D, training the least square generated MAD-LSGANs model by using a random gradient descent method according to the one-dimensional heat tensor, updating the parameters of the discrimination model D, and thus obtaining a trained discrimination model D';
4.4 fixing the parameters of the trained discrimination model D ' for training the generation model G, and updating the parameters of each generation model by using a random gradient descent method according to the heat tensor output by the discrimination model D ', thereby obtaining a trained generation model G ';
4.5, fixing the parameters of the trained generation model G ', and carrying out optimization training on the trained discrimination model D ' by using the parameters to obtain an optimized training discrimination model D ';
4.6 repeat 4.2-4.5 continuously until the discriminant model marks 1's for all the generated pseudo samples, and finish training, so as to generate the latest sample Q' from the generated model finally optimized and trained " G,(c×d)×(m+1) ；
4.7 set of the latest samples Q " G,(c×d)×(m+1) And training sample E a×(m+1) After merging, a new training sample set X is obtained s×(m+1) ，s＝a+(c×d)。
The specific method of the step 5 comprises the following steps:
5.1 the GRU neural network adopts a four-layer neural network structure;
5.2 determining a fitness function of the particles;
and 5.3, updating the individual extreme value and the global extreme value of the particle, comparing and screening the individual extreme value and the global extreme value with the historical optimal value, judging whether the optimal solution is found or the maximum iteration number is reached, if the optimal solution is met, terminating the iteration, and otherwise, continuing the iteration optimization.
The specific method of the step 6 comprises the following steps:
6.1IBBDE encodes an initial value, and adopts a random initialization mode, wherein the initial value comprises a cross network layer number U, a depth network layer number H, a neuron number theta and a training period tau;
6.2, respectively calculating the fitness values of all individuals in the population to complete population initialization;
using the number of cross network layers U, the number of depth network layers H, the number of neurons theta and a training period tau, using the super parameters of the GRU model as each dimension component of a population individual in an IBBDE algorithm, establishing a population, decoding parameters transmitted by the IBBDE, obtaining corresponding iteration times, the number of network layers and the number of hidden layer nodes, outputting corresponding adaptive values, and realizing population initialization;
6.3IBBDE carries out IBBDE population updating according to the fitness value output by 6.2;
6.4 calculating the fitness value to update the full-field optimal solution;
6.5, judging whether the conditions are met, if not, returning to IBBDE population updating again to continue the steps 6.3-6.4, when the iteration times reach the maximum iteration times, meeting termination, and if so, outputting the super parameters of the optimal GRU model to obtain the optimal GRU model.
The specific process of the step 7 is as follows:
7.1 completing the construction of the LSGAN-GRU model on the basis of the step 3 and the step 6;
7.2 New training sample set X obtained through step 4 s×(m+1) Inputting the prediction model into an LSGAN-GRU model for training, and obtaining a trained prediction model;
7.3 test set R to be subjected to the test obtained in step 3 e×(m+1) And inputting the prediction model after training for prediction, thereby obtaining a photovoltaic short-term power generation power prediction result under sudden change weather.
A LSGAN-GRU based short term photovoltaic power prediction system comprising:
a data acquisition module: the system is used for collecting the required weather data according to the location of the selected photovoltaic power station;
weather type divides module: classifying the weather types of the collected weather data by using an SOM neural network to obtain 3 types of sample data sets of sunny days, cloudy days and rainy days;
a data generation module: screening 1-type mutant weather sample data from the 3-type sample data set, and generating a new sample through MAD-LSGANS;
a parameter optimization module: finding out the optimal GRU model hyperparameters by utilizing an IBBDE optimization algorithm, and determining the optimal GRU model;
a prediction output module: training the constructed optimal GRU model through a training sample set generated based on the MAD-LSGANS, performing photovoltaic power prediction through the trained model, and outputting the short-term photovoltaic power result based on the LSGAN-GRU.
A LSGAN-GRU based short term photovoltaic power prediction apparatus comprising:
a memory for storing a computer program;
a processor for implementing said method of short term LSGAN-GRU based photovoltaic power prediction when executing said computer program.
A computer readable storage medium storing a computer program that when executed by a processor is capable of predicting short-term photovoltaic power based on LSGAN-GRU.
Compared with the prior art, the invention has the beneficial effects that:
1. the technology of the invention has the MAD-LSGANs model constructed based on the least square of a plurality of generators, so that a sample data set can be finely divided, different meteorological factors are selected according to different weather types to establish different models, and the MAD-LSGANs model is generated by means of the least square, so that a new sample is generated, the sample capacity is increased, and the problem of insufficient data is solved.
2. According to the technology, the GRU model hyper-parameters are optimized through the backbone differential evolution IBBDE, each different network has the optimal parameters, so that sudden change weather under different weather conditions can be predicted more accurately, the difference value between the predicted power and the actual power is small, the prediction error value is small, and the accuracy of the prediction of the photovoltaic power station power generation power under the sudden change weather is improved.
3. The method comprises the following steps that the influence of different weather types on prediction accuracy is not considered in the prior art, the problem of insufficient abrupt change weather data samples cannot be processed, and parameters such as hidden layers, the number of neurons and the like are mostly determined according to experience in the prior art; the prediction model part mainly adopts a GRU model consisting of a cross network and a depth network, and the global search capability of a backbone differential evolution IBBDE algorithm is adopted to search the number of network layers of an optimal depth cross neural network, the number of neurons corresponding to each layer and a training period so as to obtain an optimal solution model, so that the situation that the optimal solution falls into a local optimal solution is avoided in the prediction process, the convergence rate and the accuracy in photovoltaic prediction are greatly improved, and the photovoltaic power generation power prediction accuracy of a photovoltaic power station under the sudden change weather is improved.
Drawings
FIG. 1 is a flow chart of the steps of the present invention;
FIG. 2 is a flow chart of the method for obtaining an optimal GRU model by determining hyper-parameters of the GRU model using a backbone differential evolution IBBDE algorithm.
Detailed Description
The invention and the working principle will be described in detail below with reference to the accompanying drawings.
Referring to fig. 1, a short-term photovoltaic power prediction method based on LSGAN-GRU includes the following steps:
step 1: acquiring the nearly 8-year calendar history power generation data of the photovoltaic power station, acquiring the correlation between weather variable forecast data and photovoltaic power data based on Pearson correlation analysis, and selecting weather variables with large correlation as input variables, such as direct radiation, scattered radiation, average temperature, average humidity, average wind speed and the like. The pearson correlation coefficient is calculated as follows:
wherein E is a mathematical expectation, cov represents a covariance, X is weather variable data, and Y is photovoltaic power data;
step 2: the adopted SOM self-organizing mapping neural network comprises an input layer and a computing layer, neurons of the computing layer of the SOM self-organizing mapping neural network are classified into different response areas through self-organizing learning, and the input is automatically classified when passing through the computing layer, wherein the parameters of the neurons are updated as follows:
in the formula, w j (m) represents the mth parameter of neuron j, α is the input, η is the learning rate, the weight coefficient h and the distance d of neuron j j (α) related, distance d j (a) Described is w j (m) and α;
taking the mean value and standard deviation of the ratio of scattering horizontal radiation (DHI) to total horizontal radiation (GHI) as input values of the SOM self-organizing mapping neural network, and classifying and outputting clustering results through a calculation layer to be sunny days, cloudy days and rainy days;
and step 3: screening 1-type sudden change weather sample data for the photovoltaic data sequence under each weather type according to the weather type;
3.1 selecting n samples from each 1 type of mutant weather sample data to form a sample data set, wherein each sample in the sample data set consists of m meteorological features, and after the sample data set is subjected to normalization processing, a meteorological feature matrix with dimension n multiplied by m is obtained
Wherein x is i,j The j weather characteristics of the normalized i sample are shown, i belongs to 1,2, \ 8230, n, j belongs to 1,2, \8230, m;
3.2 collecting photovoltaic power generation power data corresponding to m meteorological features in n samples in the sample data set, and carrying out normalization processing to obtain a power sequence P = [ P ] = 1 ,P 2 ,…P i ,…P n ]，P i To representThe power of the normalized ith sample;
3.3 obtaining the meteorological feature matrix K through the 3.1 n×m Splicing with the power sequence P obtained by 3.2 to obtain a sample matrix
Wherein, P i,m+1 Representing a sample matrix K n×(m+1) The element in the ith row and m +1 column, and P i,m+1 ＝p i ，i∈1,2,…,n，j∈1,2,…,m+1；
3.4 sample matrix K obtained by 3.3 n×(m+1) Division into training samples E a×(m+1) And test specimen R e×(m+1) ,a+e＝n；
And 4, step 4: constructing a multi-generator-based least square generation MAD-LSGANs model for generating new samples;
4.1 respectively constructing D generating models by adopting a fully-connected neural network and constructing the discrimination model D by adopting the fully-connected neural network
G＝[G 1 ,G 2 ,...G g ,...,G d ],
G g Representing the g generation model, g belongs to 1,2, \8230;, d;
4.2 d one-dimensional random noise with length c is input into the generation model G, and a pseudo sample is obtained:
Q G,(c×d)×(m01) ＝[Q 1,(c)×(m+1) ,Q 2,(c)×(m01), …Q g,(c)×(m01) …,Q d,(c)×(m+1) ]；
Q g,(c)×(m+1) representing a dimension generated by the g-th generative model as c x (m + 1) pseudo sample;
4.3 fixing the parameters of the generative model G, which can be used to train a discriminant model D, and applying the training sample E obtained in step 3 a×(m+1 ) And a pseudo sample Q obtained by 4.2 G,(c×d)×(m+1) Inputting the data into a discrimination model D in 4.1 for binary processing, and outputting D one-dimensional heat tensors with the length of c; the sample labeled "1" in the one-dimensional thermal tensor represents a true sample,taking a sample marked as ' 0 ' as a false sample, training the least square generated MAD-LSGANs model by using a random gradient descent method according to the one-dimensional heat tensor, updating the parameters of the discrimination model D in 4.1, and thus obtaining a trained discrimination model D ';
4.4 fixing the parameters of the discriminant model D 'after training, it is possible to train the generated model G, input one-dimensional random noise of length c to the generated model G, and generate new pseudo sample Q' G,(c×d)×(m+1) Training sample E obtained in step 3 a×(m+1) And New pseudo sample Q' G,(c×d)×(m+1) Inputting the data into a 4.3 trained discrimination model D ' to discriminate the truth and falseness of a sample, and updating the parameters of each generation model by using a random gradient descent method according to the heat tensor output by the discrimination model D ', so as to obtain a trained generation model G ';
4.5 the parameters of the generated model G ' after the fixed training can be used for optimizing and training the discriminant model D ' after the training, the one-dimensional random noise with the length c is input into the generated model G ' after the training obtained by 4.4 to obtain a new pseudo sample, and the new pseudo sample and the training sample E obtained by the step 4 are used for optimizing and training a×(m+1) Inputting the data into a trained discrimination model D 'obtained by 4.3 for optimization training to obtain an optimized trained discrimination model D';
4.5 repeat 4.2-4.5 until all the pseudo samples generated by the discriminant model are marked as ' 1 ', the training is finished, and the latest sample Q ' is generated by the generation model finally optimized and trained " G,(c×d)×(m+1) ；
4.6 the latest sample set Q obtained in 4.5 " G,(c×d)×(m+1) And the training sample E obtained by the step 3 a×(m+1) After merging, a new training sample set X is obtained s×(m+1) ，s＝a+(c×d)；
And 5: construction of GRU model
5.1 the number of GRU network parameters is defined as the dimension component of the particle, the GRU network parameters POP = (POP 1, POP2, POP3, POP 4), T are encoded as the position vector X of each dimension of the particle position i,j ＝(X i,1 ,X i,2 ,X i,3 ,X i,4 ) T, pop1-pop4 respectively correspond to the number of neurons of the first hidden layer, the number of neurons of the second hidden layer, the iteration number of GRUs and the learning rate,
X i,j represents the j-th position of the ith particle in 4-dimensional space; pop1, pop2 ∈ [0,300 ∈ ]]，pop3∈[100,1000]，pop4∈[0.001,0.01]Wherein pop1, pop2, pop3 are all integers;
5.2 determining a fitness function of the particles;
5.3 updating the individual extreme value and the global extreme value of the particle, comparing and screening the individual extreme value and the global extreme value with historical optimal values, judging whether an optimal solution is found or the maximum iteration times is reached, if the optimal solution meets the conditions, terminating the iteration, otherwise, continuing the iteration optimization;
referring to fig. 2, step 6: obtaining an optimal GRU model
6.1IBBDE encodes an initial value, and adopts a random initialization mode, wherein the initial value comprises a cross network layer number U, a depth network layer number H, a neuron number theta and a training period tau;
6.2 calculating the fitness values of all individuals in the population respectively to finish population initialization;
using the number of cross network layers U, the number of depth network layers H, the number of neurons theta and the training period tau, and using the super parameters of the GRU model as each dimension component of population individuals in the IBBDE algorithm to establish a population POP = V 1i ,V 2i ,…V Ni Wherein the population scale N =10, i is the number of hyperparameters; decoding parameters transmitted by the IBBDE, obtaining corresponding iteration times, network layer number and hidden layer node number, outputting corresponding adaptive values, and finishing population initialization according to the adaptive values;
6.3IBBDE carries out IBBDE population updating according to the fitness value output by 6.2;
6.4 calculating the fitness value to update the full-field optimal solution;
6.5 judging whether the conditions are met, if not, returning to the IBBDE population updating again to continue the steps (c) to (d), wherein the maximum iteration number It max =20, maximum crossover probability C max =0.9, minimum crossover probability C min =0.1, a variability ratio F =0.1, a termination is satisfied when the number of iterations reaches a maximum number of iterations,and if so, outputting the super parameters of the optimal GRU model to obtain the optimal GRU model.
And 7: constructing an LSGAN-GRU model to obtain a photovoltaic short-term power generation power prediction result
7.1, completing a data generation part by constructing an LSGAN model in step 3, establishing a GRU model based on the optimal hyper-parameter in step 5 and step 6, and finally establishing an LSGAN-GRU model on the basis of the steps;
7.2 New training sample set X obtained through step 4 s×(m+1) Inputting the prediction model into an LSGAN-GRU model for training, and obtaining a trained prediction model;
7.3 test set R to be subjected to the test obtained in step 3 e×(m+1) And inputting the prediction model after training for prediction, thereby obtaining a photovoltaic short-term power generation power prediction result under sudden change weather.
A short-term photovoltaic power prediction system based on LSGAN-GRU comprises a data acquisition module, a weather type division module, a data generation module, a parameter optimization module and a prediction output module; wherein:
a data acquisition module: the system is used for collecting the required weather data according to the location of the selected photovoltaic power station;
a weather type division module: classifying the weather types of the collected weather data by using an SOM neural network to obtain 3 types of sample data sets of sunny days, cloudy days and rainy days;
a data generation module: screening 1-type mutant weather sample data from the 3-type sample data set, and generating a new sample through MAD-LSGANS;
a parameter optimization module: finding out the optimal GRU model hyperparameters by utilizing an IBBDE optimization algorithm, and determining the optimal GRU model;
a prediction output module: training the constructed optimal GRU model through a training sample set generated based on the MAD-LSGANS, predicting the photovoltaic power through the trained model, and outputting the obtained short-term photovoltaic power result based on the LSGAN-GRU.
A LSGAN-GRU based short term photovoltaic power prediction apparatus comprising:
a memory for storing a computer program;
a processor for implementing a short term photovoltaic power prediction method based on an LSGAN-GRU network when executing the computer program.
The processor may be a Central Processing Unit (CPU), other general purpose processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, etc. The general purpose processor may be a microprocessor or the processor may be any conventional processor or the like, the processor being the control center for the one type of LSGAN-GRU based short term photovoltaic power prediction device, the various parts of the short term photovoltaic power prediction device being connected to the entire LSGAN-GRU network using various interfaces and lines.
The processor, when executing the computer program, performs the steps of the LSGAN-GRU based short-term photovoltaic power prediction method described above, such as: selecting the location of a photovoltaic power station to be predicted, and collecting required weather data; classifying the weather types of the collected weather data by using an SOM neural network to obtain 3 types of sample data sets of sunny days, cloudy days and rainy days; screening 1-type mutation weather sample data in the 3-type sample data set, generating a new sample through MAD-LSGANS, finding out an optimal GRU model hyperparameter by utilizing an IBBDE optimization algorithm, and determining an optimal GRU model; training the constructed optimal GRU model through a training sample set generated based on the MAD-LSGANS, predicting the photovoltaic power through the trained model, and outputting the obtained short-term photovoltaic power result based on the LSGAN-GRU.
Alternatively, the processor implements the functions of the modules in the system when executing the computer program, for example: a data acquisition module: the system is used for collecting the required weather data according to the location of the selected photovoltaic power station; weather type divides module: classifying the collected weather data by using an SOM neural network to obtain 3 types of sample data sets in sunny days, cloudy days and rainy days; a data generation module: screening 1-type mutant weather sample data in the 3-type sample data set, and generating a new sample through MAD-LSGANS; a parameter optimization module: finding out an optimal GRU model hyperparameter by utilizing an IBBDE optimization algorithm, and determining an optimal GRU model; a prediction output module: training the constructed optimal GRU model through a training sample set generated based on the MAD-LSGANS, predicting the photovoltaic power through the trained model, and outputting the obtained short-term photovoltaic power result based on the LSGAN-GRU.
Illustratively, the computer program may be partitioned into one or more modules/units that are stored in the memory and executed by the processor to implement the invention. The one or more modules/units may be a series of computer program instruction segments capable of performing preset functions, the instruction segments describing the execution of the computer program in the LSGAN-GRU based short term photovoltaic power prediction apparatus. For example, the computer program may be divided into a data acquisition module, a weather type division module, a data generation module, a parameter optimization module, and a prediction output module, and the specific functions of each module are as follows: a data acquisition module: the system is used for collecting the required weather data according to the location of the selected photovoltaic power station; a weather type division module: classifying the weather types of the collected weather data by using an SOM neural network to obtain 3 types of sample data sets of sunny days, cloudy days and rainy days; a data generation module: screening 1-type mutant weather sample data from the 3-type sample data set, and generating a new sample through MAD-LSGANS; a parameter optimization module: finding out the optimal GRU model hyperparameters by utilizing an IBBDE optimization algorithm, and determining the optimal GRU model; a prediction output module: training the constructed optimal GRU model through a training sample set generated based on the MAD-LSGANS, predicting photovoltaic power through the trained model, and outputting the obtained short-term photovoltaic power result based on the LSGAN-GRU.
The LSGAN-GRU-based short-term photovoltaic power prediction device can be a desktop computer, a notebook computer, a palm computer, a cloud server and other computing devices. The LSGAN-GRU-based short-term photovoltaic power prediction device may include, but is not limited to, a processor, a memory. Those skilled in the art will appreciate that the foregoing is an example of a LSGAN-GRU based short term photovoltaic power prediction device and does not constitute a limitation of a LSGAN-GRU based short term photovoltaic power prediction device and may include more components than those described above, or combine certain components, or different components, for example a LSGAN-GRU based short term photovoltaic power prediction device may also include input output devices, network access devices, buses, etc.
The memory may be used to store the computer programs and/or modules, and the processor may implement the various functions of the LSGAN-GRU based short-term photovoltaic power prediction apparatus by running or executing the computer programs and/or modules stored in the memory, and invoking data stored in the memory.
The memory may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required by at least one function (such as a sound playing function, an image playing function, etc.), and the like; the storage data area may store data (such as audio data, a phonebook, etc.) created according to the use of the cellular phone, and the like. In addition, the memory may include high speed random access memory, and may also include non-volatile memory, such as a hard disk, memory, a plug-in hard disk, a smart card
(smartmedia card, SMC), secure Digital (SD) card, flash memory card (FlashCard), at least one magnetic disk storage device, flash memory device, or other volatile solid state storage device.
The invention also provides a computer readable storage medium storing a computer program which, when executed by a processor, implements the steps of a LSGAN-GRU based short-term photovoltaic power prediction method.
The integrated modules/units of the LSGAN-GRU based short term photovoltaic power prediction system may be stored in a computer readable storage medium if implemented as software functional units and sold or used as stand-alone products.
The invention realizes all or part of the flow of the LSGAN-GRU-based short-term photovoltaic power prediction method, and can also be completed by a computer program instructing related hardware, wherein the computer program can be stored in a computer readable storage medium, and when being executed by a processor, the computer program can realize the steps of the LSGAN-GRU-based short-term photovoltaic power prediction method. Wherein the computer program comprises computer program code, which may be in source code form, object code form, executable file or preset intermediate form, etc.
The computer-readable storage medium may include: any entity or device capable of carrying the computer program code, recording medium, usb disk, removable hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random Access Memory (RAM), electrical carrier wave signals, telecommunications signals, software distribution medium, etc.
It should be noted that the computer readable storage medium may contain content that is subject to appropriate increase or decrease as required by legislation and patent practice in jurisdictions, for example, in some jurisdictions, computer readable storage media that does not include electrical carrier signals and telecommunications signals in accordance with legislation and patent practice.
Claims (9)
1. A short-term photovoltaic power prediction method based on LSGAN-GRU is characterized by comprising the following steps:
step 1: acquiring historical power generation data of a photovoltaic power station, acquiring the correlation between weather variable forecast data and photovoltaic power data based on Pearson correlation analysis, and selecting weather variables with high correlation as input variables;
step 2: classifying the weather types according to the irradiation intensity based on the SOM self-organizing mapping neural network;
and step 3: screening 1-type sudden change weather sample data for the photovoltaic data sequence under each weather type according to the weather type;
and 4, step 4: constructing an MAD-LSGANs model based on the least square of a multi-generator;
and 5: constructing a GRU model consisting of a cross network and a depth network;
step 6: determining the hyper-parameters of the GRU model by adopting a backbone differential evolution (IBBDE) algorithm so as to obtain an optimal GRU model;
and 7: and constructing an LSGAN-GRU model to obtain a photovoltaic short-term power generation power prediction result in sudden change weather.
2. The LSGAN-GRU-based short-term photovoltaic power prediction method as claimed in claim 1, wherein the specific process of step 2 is: the SOM self-organizing mapping neural network comprises an input layer and a computing layer, neurons of the computing layer of the SOM self-organizing mapping neural network are classified into different response areas through self-organizing learning, input variables are automatically classified when passing through the computing layer, and clustering results output through the computing layer in a classified mode are sunny days, cloudy days and rainy days.
3. The LSGAN-GRU-based short-term photovoltaic power prediction method as claimed in claim 1, wherein the specific process of step 4 is:
4.1, constructing the discrimination model D by adopting a full-connection neural network;
4.2, d one-dimensional random noises with the length of c are input into a generation model G, and a pseudo sample is obtained;
4.3 fixing the parameters of the generated model G, training a discrimination model D, training the least square generated MAD-LSGANs model by using a random gradient descent method according to the one-dimensional heat tensor, updating the parameters of the discrimination model D, and thus obtaining a trained discrimination model D';
4.4 fixing the parameters of the trained discrimination model D ' for training the generation model G, and updating the parameters of each generation model by using a random gradient descent method according to the heat tensor output by the discrimination model D ', thereby obtaining a trained generation model G ';
4.5, fixing the parameters of the trained generation model G ', and carrying out optimization training on the trained discrimination model D ' by using the parameters to obtain an optimized training discrimination model D ';
4.6 repeat 4.2-4.5 continuously until the discriminant model marks 1's for all the generated pseudo samples, and finish training, so as to generate the latest sample Q' from the generated model finally optimized and trained " G,(c×d)×(m+1) ；
4.7 set Q of latest samples " G,(c×d)×(m+1) And training sample E a×(m+1) After merging, a new training sample set X is obtained s×(m+1) ，s＝a+(c×d)。
4. The LSGAN-GRU-based short-term photovoltaic power prediction method as claimed in claim 1, wherein the specific method of step 5 is:
5.1 the GRU neural network adopts a four-layer neural network structure;
5.2 determining a fitness function of the particles;
and 5.3, updating the individual extreme value and the global extreme value of the particle, comparing and screening the individual extreme value and the global extreme value with the historical optimal value, judging whether the optimal solution is found or the maximum iteration number is reached, if the optimal solution is met, terminating the iteration, and otherwise, continuing the iteration optimization.
5. The LSGAN-GRU-based short-term photovoltaic power prediction method as claimed in claim 1, wherein the specific method of step 6 is:
6.1IBBDE encodes an initial value in a random initialization mode, wherein the initial value comprises a cross network layer number U, a depth network layer number H, a neuron number theta and a training period tau;
6.2 calculating the fitness values of all individuals in the population respectively to finish population initialization;
using the cross network layer number U, the depth network layer number H, the neuron number theta and the training period tau, using the super parameters of the GRU model as each dimension component of a population individual in the IBBDE algorithm, establishing a population, decoding parameters transmitted by the IBBDE, obtaining corresponding iteration times, the network layer number and the number of hidden layer nodes, outputting corresponding adaptive values, and realizing population initialization;
6.3IBBDE carries out IBBDE population updating according to the fitness value output by 6.2;
6.4 calculating the fitness value to update the full-field optimal solution;
6.5, judging whether the conditions are met, if not, returning to IBBDE population updating again to continue the steps 6.3-6.4, when the iteration times reach the maximum iteration times, meeting termination, and if so, outputting the super parameters of the optimal GRU model to obtain the optimal GRU model.
6. The LSGAN-GRU-based short-term photovoltaic power prediction method as claimed in claim 1, wherein the specific process of step 7 is:
7.1 completing the construction of the LSGAN-GRU model on the basis of the step 3 and the step 6;
7.2 New training sample set X obtained through step 4 s×(m+1) Inputting the prediction model into an LSGAN-GRU model for training, and obtaining a trained prediction model;
7.3 test set R to be subjected to the test obtained in step 3 e×(m+1) And inputting the prediction model after training for prediction, thereby obtaining a photovoltaic short-term power generation power prediction result under sudden change weather.
7. A LSGAN-GRU based short term photovoltaic power prediction system, comprising:
a data acquisition module: the system is used for collecting the required weather data according to the location of the selected photovoltaic power station;
weather type divides module: classifying the weather types of the collected weather data by using an SOM neural network to obtain 3 types of sample data sets of sunny days, cloudy days and rainy days;
a data generation module: screening 1-type mutant weather sample data from the 3-type sample data set, and generating a new sample through MAD-LSGANS;
a parameter optimization module: finding out the optimal GRU model hyperparameters by utilizing an IBBDE optimization algorithm, and determining the optimal GRU model;
a prediction output module: training the constructed optimal GRU model through a training sample set generated based on the MAD-LSGANS, performing photovoltaic power prediction through the trained model, and outputting the short-term photovoltaic power result based on the LSGAN-GRU.
8. A LSGAN-GRU-based short-term photovoltaic power prediction device, comprising:
a memory for storing a computer program;
a processor for implementing a LSGAN-GRU based short-term photovoltaic power prediction method as claimed in any of claims 1-6 when executing the computer program.
9. A computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program which, when executed by a processor, is capable of predicting short-term photovoltaic power based on a LSGAN-GRU network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211714868.9A CN115952906A (en) | 2022-12-29 | 2022-12-29 | Short-term photovoltaic power prediction method, system, equipment and medium based on LSGAN-GRU |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211714868.9A CN115952906A (en) | 2022-12-29 | 2022-12-29 | Short-term photovoltaic power prediction method, system, equipment and medium based on LSGAN-GRU |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115952906A true CN115952906A (en) | 2023-04-11 |
Family
ID=87291127
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211714868.9A Pending CN115952906A (en) | 2022-12-29 | 2022-12-29 | Short-term photovoltaic power prediction method, system, equipment and medium based on LSGAN-GRU |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115952906A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117200199A (en) * | 2023-09-06 | 2023-12-08 | 国网上海市电力公司 | Photovoltaic power prediction method and system based on weather typing |
-
2022
- 2022-12-29 CN CN202211714868.9A patent/CN115952906A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117200199A (en) * | 2023-09-06 | 2023-12-08 | 国网上海市电力公司 | Photovoltaic power prediction method and system based on weather typing |
| CN117200199B (en) * | 2023-09-06 | 2024-04-02 | 国网上海市电力公司 | Photovoltaic power prediction method and system based on weather typing |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11042802B2 (en) | System and method for hierarchically building predictive analytic models on a dataset | |
| JP5888640B2 (en) | Photovoltaic power generation prediction apparatus, solar power generation prediction method, and solar power generation prediction program | |
| Senthil Kumar | Improved prediction of wind speed using machine learning. | |
| CN116596044B (en) | Power generation load prediction model training method and device based on multi-source data | |
| CN111723523B (en) | Estuary surplus water level prediction method based on cascade neural network | |
| CN112100911B (en) | Solar radiation prediction method based on depth BILSTM | |
| CN106600041A (en) | Restricted Boltzmann machine-based photovoltaic power generation short-term power probability prediction method | |
| CN116644970A (en) | Photovoltaic power prediction method based on VMD decomposition and lamination deep learning | |
| CN115392387B (en) | Low-voltage distributed photovoltaic power generation output prediction method | |
| CN114676923A (en) | Method and device for predicting generated power, computer equipment and storage medium | |
| CN115952906A (en) | Short-term photovoltaic power prediction method, system, equipment and medium based on LSGAN-GRU | |
| Cao et al. | Short-term forecasting and uncertainty analysis of photovoltaic power based on the FCM-WOA-BILSTM model | |
| Ragab | Spider Monkey Optimization with Statistical Analysis for Robust Rainfall Prediction. | |
| Kumar | Improved Prediction of Wind Speed using Machine Learning. | |
| CN117728386A (en) | Prediction method for renewable energy power generation capacity of green building | |
| Wang et al. | Time series prediction model of grey wolf optimized echo state network | |
| Alam | Recurrent neural networks in electricity load forecasting | |
| CN115759343A (en) | E-LSTM-based user electric quantity prediction method and device | |
| Feng et al. | Occlusion-perturbed deep learning for probabilistic solar forecasting via sky images | |
| Wu et al. | Short-Term Prediction of Wind Power Considering the Fusion of Multiple Spatial and Temporal Correlation Features | |
| Aalto et al. | Continuous time Gaussian process dynamical models in gene regulatory network inference | |
| CN113902492A (en) | Time-sharing electricity price prediction method and system | |
| Padhan et al. | Power forecasting with minimal loss using lstm and pv model | |
| Zhu et al. | Short-term Load Forecasting Based on Kmeans and FR-DBN Models | |
| Devi et al. | An Advanced Hybrid Meta-Heuristic Model for Solar Power Generation Forecasting via Ensemble Deep Learning. |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |