CN116681154A

CN116681154A - Photovoltaic power calculation method based on EMD-AO-DELM

Info

Publication number: CN116681154A
Application number: CN202310511494.9A
Authority: CN
Inventors: 曹哲; 赵葵银; 林国汉
Original assignee: Hunan Institute of Engineering
Current assignee: Hunan Institute of Engineering
Priority date: 2023-05-08
Filing date: 2023-05-08
Publication date: 2023-09-01

Abstract

The invention discloses a photovoltaic power calculation method based on EMD-AO-DELM, which comprises the following steps of S1: the method comprises the steps of obtaining meteorological factors, wherein the meteorological factors comprise cloud cover, air temperature, air pressure, humidity and total radiation, calculating that the total radiation and the cloud cover are highly positively correlated with photovoltaic power, and the relative humidity and the atmospheric pressure are negatively correlated, so that the cloud cover and the total radiation are selected as DELM initial input data; s2: establishing an AO-DELM calculation model, taking the DELM initial input weight as an initial population position of an AO algorithm, and setting an fitness function as the sum of mean square errors of a training set and a test set; s3: establishing an EMD-AO-DELM calculation model, decomposing a photovoltaic power generation power curve by adopting EMD, so as to decompose different scale fluctuation or trend existing in an original environment signal step by step, and respectively carrying out AO-DELM modeling analysis on the decomposed IMF component; s4: verifying the validity and accuracy of the calculation model; the invention realizes more accurate budget photovoltaic power by inputting fewer parameters.

Description

Photovoltaic power calculation method based on EMD-AO-DELM

Technical Field

The invention belongs to the technical field of photovoltaic power calculation, and particularly relates to a photovoltaic power calculation method based on EMD-AO-DELM.

Background

Photovoltaic power generation is widely used at present as a new energy source for power generation. However, the power output of photovoltaic power generation is affected by a number of factors, including environmental factors such as solar radiation intensity, temperature, humidity, and cloud cover. Variations in these factors can cause fluctuations in the power output of the photovoltaic power system, thereby adversely affecting the safety and stability of the power system, as well as presenting challenges to the photovoltaic grid-tie scheduling process.

The prediction technology for photovoltaic power generation power is mainly divided into a direct prediction method and an indirect prediction method. The photovoltaic historical data information is trained and learned, and future power prediction is carried out through a prediction algorithm. The latter adopts a step-by-step prediction mode, and can be divided into two parts of future solar radiation and future power prediction. The comparison document discloses a photovoltaic power station short-term power prediction based on EMD and ELM, and provides a combined power prediction method based on Empirical Mode Decomposition (EMD) and an Extreme Learning Machine (ELM), but the environmental influence factors of the output power of the photovoltaic power station cannot be fully considered. The prior art provides an improved BP neural network photovoltaic prediction method based on a firework algorithm (FWA), and the method has good accuracy for short-term photovoltaic data prediction. However, the BP neural network has the problems of low convergence speed and easy local extremum. The prior art discloses a photovoltaic power prediction method based on CNN-BiLSTM, and provides a prediction algorithm combining a Convolutional Neural Network (CNN) and a bidirectional long and short time memory network (BILSTM), but the method is too single in test sample type and fails to test prediction accuracy in different seasons and weather.

The photovoltaic power generation efficiency has a plurality of influencing factors, and is mainly divided into subjective factors and objective factors. The subjective factors comprise parameters of the plate type of the photovoltaic array, inclination angles and directions of the photovoltaic plates and the like; the objective factors include uncontrollable meteorological factors such as air temperature, humidity, cloud cover, precipitation, illumination radiation and the like, and often play a decisive role. Too many input sample factors can reduce prediction accuracy and make the prediction model complex and redundant.

Disclosure of Invention

The invention aims to provide an EMD-AO-DELM-based photovoltaic power calculation method for solving the problems that excessive input sample factors of the existing photovoltaic power prediction method can reduce prediction accuracy, make a prediction model complex, redundant and the like.

The invention realizes the above purpose through the following technical scheme: the method comprises the following steps:

s1: the method comprises the steps of obtaining meteorological factors, wherein the meteorological factors comprise cloud cover, air temperature, air pressure, humidity and total radiation, calculating that the total radiation and the cloud cover are highly positively correlated with photovoltaic power, and the relative humidity and the atmospheric pressure are negatively correlated, so that the cloud cover and the total radiation are selected as DELM initial input data;

s2: establishing an AO-DELM calculation model, taking the DELM initial input weight as an initial population position of an AO algorithm, and setting an fitness function as the sum of mean square errors of a training set and a test set;

s3: establishing an EMD-AO-DELM calculation model, decomposing a photovoltaic power generation power curve by adopting the EMD, so as to decompose different scale fluctuation or trend existing in an original environment signal step by step, respectively carrying out AO-DELM modeling analysis on decomposed IMF components, and then carrying out superposition summation on prediction results of the IMF components to obtain a final prediction value;

s4: and verifying the validity and accuracy of the calculation model.

Further, the step S1 includes: s11: introducing Pearson correlation coefficient through the formulaCalculating a linear relationship between distance variables for measuring whether two data sets are on a line, wherein r>1 represents positive correlation between the two, r<1 represents that the two are in negative correlation; xi and yi represent two respectivelyThe value of factor i; />And->Each representing an average of 2 factors.

Further, the step S2 includes: the calculation formula is fitness=mse (train) +mse (test).

Further, the step S2 includes: s21: data cleaning is carried out, and abnormal values caused by errors in sampling some historical photovoltaic power data are proposed; s22: normalizing the cleaned sample data; s23: initializing AO algorithm parameters, including population scale, maximum iteration times T, and exploring and developing parameters alpha and delta; s24, initializing a population position X, initial population fitness and optimal individuals; s25: the method comprises the steps of sequentially carrying out an expansion exploration phase, a reduction exploration phase, an expansion development phase and a reduction development phase, and continuously updating the population positions; s26: calculating the fitness of the updated population to obtain the current optimal individual position and fitness, comparing the current optimal individual with the optimal individual fitness found until the t generation, and reserving the optimal individual position; s27: judging whether the maximum iteration times or solving conditions are reached, if so, outputting an optimal value, and if not, returning to the step S25; s28: and inputting the final optimized weight value result into the DELM model.

Further, the step S3 includes: s31: decomposing the photovoltaic historical data by adopting EMD to obtain a group of IMF components; s32: respectively establishing an AO-DELM model for each IMF component, and predicting each component; s33: and superposing the prediction results of the subsequences and verifying the accuracy of model prediction.

Further, the step S4 includes: adopts both MAPE and RMSE as error indexes, and the MAPE calculation formula is as followsThe calculation formula of RMSE is +.>Where yoi is the i-th true value in the sample, ypi is the i-th predicted value in the sample, and the smaller the two values, the higher the accuracy.

The beneficial effects are that: the invention has reasonable design, simple and stable structure and strong practicability, and has the following beneficial effects:

1. among photovoltaic power generation influencing factors, the total illumination radiation and cloud quantity are positively correlated with photovoltaic power, and a key effect is played on a final prediction result; the air pressure and the humidity are inversely related to the photovoltaic power, and are not suitable to be used as input data in actual power prediction;

2. aiming at the characteristics of volatility and randomness of photovoltaic power, EMD decomposition is carried out on historical photovoltaic power data, all components are mutually independent, prediction is carried out respectively, and finally superposition and summation are carried out, and experiments prove that the prediction effect is better after the EMD decomposition method is adopted;

3. the prediction performance of the method in four quarters of one year is superior to that of the AO-DELM and DELM models, wherein the prediction accuracy is highest in two quarters of S2 and S3, the prediction accuracy is related to local weather conditions, the weather is stable, and the prediction accuracy of the quarters with high sunny occupation ratio is higher.

Drawings

FIG. 1 is a schematic view of an ELM structure of the present invention;

FIG. 2 is a schematic view of an ELM-AE structure of the present invention;

FIG. 3 is a schematic illustration of the DELM structure of the present invention;

FIG. 4 is a flow chart of the AO-DELM structure of the present invention;

FIG. 5 is a schematic diagram of a first quarter power EMD decomposition sequence according to the present invention;

FIG. 6 is a flow chart of the EMD and AO-DELM combination prediction of the present invention;

FIG. 7 is a graph showing the power predictions for different quarters according to the present invention;

FIG. 8 is a graph of the prediction error of each algorithm for the first quarter of the present invention;

fig. 9 is a flow chart of the present invention.

Detailed Description

The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments.

Embodiment one:

empirical Mode Decomposition (EMD) is a method of signal decomposition based on local features of signals.

The method absorbs the advantages of multi-resolution of wavelet transformation, overcomes the difficulty of selecting wavelet basis and determining decomposition scale in wavelet transformation, is more suitable for the analysis of nonlinear non-stationary signals, and is a self-adaptive signal decomposition method. EMD assumes that any complex signal is composed of a simple characteristic modal function (IMF). And each IMF component is independent of the other. EMD can decompose time series data of different scales or trends, in steps into its constituent parts, and generate a series of data sequences with the same scale features, by which non-stationary nonlinear data is converted into stationary linear data. The decomposed sequence has a higher regularity than the original data sequence. This greatly helps identify hidden relationships and can improve the accuracy of predictions.

The specific decomposition steps are as follows:

for the initial time sequence x (t), taking all maximum values and minimum value points of the initial time sequence x (t), connecting all maximum values as an upper envelope, connecting all minimum value points as a lower envelope, and recording m (t) as the average value of the upper envelope and the lower envelope. Subtracting the original sequence x (t) from the mean value m (t) to obtain a first component h1 (t) =x (t) -m (t);

taking h1 (t) as an initial time sequence, recording m1 (t) as the average value of the upper envelope curve and the lower envelope curve of h1 (t), and repeating the step (1) to obtain a second component h2 (t);

repeating the steps for n times until hn (t) is an eigenmode function or the residual component rn (t) presents monotonicity, and terminating the decomposition process;

to this end, the initial time sequence x (t) may be represented by the sum of n eigenmode components hi (t) and one residual component rn (t),the formula is:

the hawk algorithm principle: the mathematical model is briefly described as follows:

step 1: expansion exploration

In this method, the Aquila bird flies high above the ground, and a hunting space is widely explored, and once the Aquila bird determines the area of the prey, a vertical dive is adopted. The mathematical representation of this behavior is written as:

where Xbast (T) represents the best position obtained so far, XM (T) represents the average position of all Aquica birds in the current iteration, and T and T are the current iteration and the maximum number of iterations, respectively. N is the population size and rand is a random number between 0 and 1.

Step 2: stage of narrowing exploration

This is the most commonly used method of hunting by Aquila birds. It is launched in a short range glide mode to attack the prey after descending in a selected area and flying around the prey. The location update formula is expressed as:

X(t+1)＝X _best (t)×LF(D)+X _R (t)+(y-x)×rand

where XR (t) represents the random position of the Aquila bird, D is the dimension size, LF (D) represents the Levy flight function, which is expressed as follows:

where s and β are constants equal to 0.01 and 1.5, respectively, and u and v are random numbers between 0 and 1. y and x are used to present the spiral in the search and are calculated as follows:

where r1 refers to the number of search cycles between 1 and 20, D1 is comprised of an integer from 1 to dimension D, and ω is equal to 0.005.

Step 3: expanding the development phase

In the third stage, after the area of the prey is approximately determined, the Aquila bird descends vertically, and a preliminary attack is performed. AO utilizes the selected area to approach and attack the prey. This behavior is represented as follows:

X(t+1)＝(X _best (t)-X _M (t))×α-rand+((U _B -L _B )×rand+L _B )×δ

where α and δ are development tuning parameters fixed to 0.1, and UB and LB are upper and lower bounds.

Step 4: shrinking the development phase

In this method, an Aquila bird chases the prey according to the trajectory of the prey's escape and then attacks the prey on the ground. The mathematical representation of this behavior is as follows:

where X (t) is the current position and QF (t) represents the quality function value for the balanced search strategy. G1 represents the kinetic parameters of hawk in tracking hunting, and is a random number between [ -1,1 ]. G2 represents the flight slope at the time of chase of the prey, decreasing linearly from 2 to 0.

ELM:

The model of the extreme learning machine (Extreme Learning Machine, ELM) consists of three parts, respectively: the input layer, hidden layer and output layer are a typical (Single-hidden Layer Feed-forward Neural Network, SLFN) Single hidden layer feedforward neural network, which has the advantages of high learning speed, strong generalization capability and the like.

The model structure of the ELM is shown in fig. 1, where the input layer contains q nodes, the hidden layer contains n nodes, the output layer contains e nodes, the hidden layer activation function is g (x), and common functions include Sigmoid, hard-lim, sin, and the like.

Assuming that the samples are xi e rn×rq, yi e rn×re (i=1, 2,.., N), where the output of the hidden layer is equation (10), the relationship between the hidden layer output matrix and ELM network output can be represented by equation (11).

h＝g(ax+b)

h(xi)V＝yi，i＝1,2,...,N

Wherein, the liquid crystal display device comprises a liquid crystal display device,

wherein a is _i ＝[a _i1 ,a _i2 ,…,a _in ] ^T Is the weight connecting the ith input node and hidden layer, bj is the threshold of the jth hidden node, v _j ＝[v _j1 ,v _j2 ,…v _jn ] ^T Is the weight connecting the j-th hidden node and the output layer. H is the hidden layer output matrix of the neural network. Randomly selecting an input weight aij and a threshold value bj of an implicit layer; the output weight V may be obtained by solving a system of equations.

Obtaining output weights using ELM can be divided into three steps.

Randomly selecting a value between 0 and 1 to set the input weight aij and the threshold bj of the hidden layer;

calculating an hidden layer output matrix H; calculating an output weight v=h ⁺ Y。

Where h+ represents the generalized inverse of the output matrix H.

Unlike conventional gradient-based feedforward neural network algorithms, the extreme learning machine network hidden layer randomly generates input weights and thresholds during training. Therefore, the generalized inverse matrix theory can only be adopted to calculate the output weight. ELM, however, is a single hidden layer structure that lacks the ability to capture the valid features of data in the face of large and high-dimensional input data. Therefore, more students adopt the DELM algorithm as a derivative algorithm of the ELM, and the problem that the extreme learning machine with only one hidden layer cannot capture the effective characteristics of the data is solved.

Extreme learning machine-automatic encoder (ELM-AE):

an automatic encoder (ELM-AE) is an artificial neural network module, is commonly used in the field of deep learning, and is a structure for learning samples in an unsupervised manner. It is mainly characterized by that the output and input results of network are identical. The model of ELM-AE is like ELM and consists of three parts, an input layer, an hidden layer and an output layer. The model structure is shown in figure 2, the weight and the threshold value of the hidden layer node are randomly generated in the training process of the constructed ELM-AE, and the model structure has orthogonality, so that the generalization capability of the ELM-AE is optimized to a certain extent. In order to further improve the generalization capability and robustness of the model, regularization parameters are introduced in the process of solving the weight coefficients. The objective function is set as:

assuming that given N different samples, xi e rn×rq (i=1, 2, N), the output of the ELM-AE hidden layer can be expressed as the equation h=g (ax+b), and then the mathematical relationship between the output matrix of the hidden layer and the output of the output layer can be expressed asWhere i=1, 2,..n, for an equal dimension ELM-AE representation, the calculation method of the output weight V is: v=h ^- ¹ X, where H is the ELM-AE hidden layer output matrix and X is the ELM-AE input and output matrix.

Deep Extreme Learning Machines (DELMs) increase the expressive power of the network by building a multi-layer network structure by overlaying an extreme learning machine-automatic encoder (ELM-AE). Is a new structure of the combination of an extreme learning machine and an automatic encoder.

DELM uses ELM-AE to train the model layer by layer. The numerical relationship between the output of the i-layer hidden layer and the output of the (i-1) layer hidden layer can be represented by the following equation: h _i ＝g((v _i ) ^T H _i-1 )

DELM (deep extreme learning machine):

the ELM-AE is used to construct the base unit of the deep extreme learning machine DELM, and then the output weight of the ELM-AE is used to initialize the entire DELM. The DELM concept is to learn advanced features of the original data by minimizing reconstruction errors to make the output infinitely close to the original input, and by iterative training layer by layer.

ELM-AE maps the input to an hidden layer feature vector at the encoder, reconstructing the original input from the feature vector at the decoder. From a structural point of view, DELM is equivalent to connecting multiple ELMs. Compared with ELM, the DELM can more comprehensively capture sample characteristics and improve the accuracy of processing high-dimensional input. The DELM performs unsupervised training and learning layer by layer through ELM-AE, and finally connects to the last output layer for supervised training. The parameters of the system do not need to be adjusted simultaneously. The structure of the DELM network is shown in FIG. 3. The input weights of each hidden layer of the DELM are initialized through the ELM-AE, and hierarchical unsupervised training is performed. During this entire process, the DELM does not need to be back-trimmed.

Assuming that there are Y hidden layers in the model, the weight matrix V1 can be obtained from the input data X according to the ELM-AE theory described above, and then the output matrix H1 of the hidden layers can be obtained. H1 is then taken as the input and target output of the next ELM-AE. By analogy, the layer-by-layer training can obtain an output weight matrix VY of the Y layer and an output matrix HY of the hidden layer. Wherein the output weight of each ELM-AE is used to initialize the entire DELM. In the ELM-AE training process, the input layer weight and the threshold are orthogonal random matrixes generated randomly; meanwhile, the ELM-AE unsupervised training process adopts a least square method to update parameters. In this process, only the output layer weight parameters are updated, while the input layer weights and thresholds remain unchanged, and the random input weights and random threshold effects of each ELM-AE will have an impact on the prediction accuracy of DELM. Since the initial weights play a more critical role in the prediction of the whole model. Thus, the input weights for DELM herein are optimized using AO algorithm.

The global optimization capacity of the hawk optimization algorithm is utilized, and the input weight of the depth extreme learning machine can be found when the training error is small, so that the generalization capacity of the depth extreme learning machine is improved, and the prediction accuracy of the DELM is improved.

Factor selection:

as shown in fig. 4 to 9, the photovoltaic power generation efficiency has a plurality of influencing factors, and is mainly divided into subjective factors and objective factors. The subjective factors comprise parameters of the plate type of the photovoltaic array, inclination angles and directions of the photovoltaic plates and the like; the objective factors include uncontrollable meteorological factors such as air temperature, humidity, cloud cover, precipitation, illumination radiation and the like, and often play a decisive role. Too many input sample factors can reduce prediction accuracy and make the prediction model complex and redundant. To explore the correlation of meteorological factors with photovoltaic power in order to select optimal factors as inputs, pearson correlation coefficients are introduced here, which are used to measure whether two data sets are on a line, which are used to measure the linear relationship between distance variables.

Wherein r is>1 represents positive correlation between the two, r<1 represents that the two are in negative correlation; xi and yi represent the values of the ith of two factors, respectively;and->Each representing an average of 2 factors.

The following table is organized according to Pearson correlation coefficients.

From the table, the total radiation and the cloud amount of illumination are highly positively correlated with the photovoltaic power, and the relative humidity and the atmospheric pressure are negatively correlated, so that the two terms of cloud amount and total radiation are selected as DELM initial input data.

Building an AO-DELM model:

the main ideas of the AO-DELM model are: using the DELM initial input weight as an initial population position of an AO algorithm; and setting the fitness function as the sum of the mean square errors of the training set and the test set, expressed as follows:

fitness＝MSE(train)+MSE(test)

the AO-DELM prediction model flow is as follows:

carrying out data cleaning, removing abnormal values caused by errors in sampling of some historical photovoltaic power data, and carrying out normalization processing on cleaned sample data;

initializing AO algorithm parameters, including population scale, maximum iteration times T, and exploring and developing parameters alpha and delta;

initializing a population position X, an initial population fitness and an optimal individual;

the method comprises the steps of sequentially carrying out an expansion exploration phase, a reduction exploration phase, an expansion development phase and a reduction development phase, and continuously updating the population positions;

calculating the fitness of the updated population to obtain the current optimal individual position and fitness, comparing the current optimal individual with the optimal individual fitness found until the t generation, and reserving the optimal individual position;

judging whether the maximum iteration times or solving conditions are reached, if yes, outputting an optimal value, and if not, returning to the step 5;

inputting the final optimized weight value result into the DELM model, the structure flow chart of which is shown in figure 4

Establishment of EMD-AO-DELM model

The photovoltaic power data are nonlinear and non-stable discrete data, the traditional linear time sequence model method has larger limitation, and the direct predictive modeling of the photovoltaic power data has larger error. Therefore, the EMD is adopted to decompose the photovoltaic power generation power curve, so that different scale fluctuation or trend existing in the original environment signal is decomposed step by step. And respectively carrying out AO-DELM modeling analysis on the decomposed IMF components, and then carrying out superposition summation on the prediction results of the IMF components to obtain a final prediction value.

The method comprises the following specific steps:

decomposing the photovoltaic historical data by adopting EMD to obtain a group of IMF components;

respectively establishing an AO-DELM model for each IMF component, and predicting each component;

superposing the prediction results of the subsequences and verifying the accuracy of model prediction;

the EMD-AO-DELM prediction model is shown in FIG. 6.

The EMD method was used to decompose the photovoltaic history into IMF components of 4 different features and a margin Res, where the IMF sequences for the first quarter photovoltaic power are shown in fig. 5. The IMF component can reflect local characteristics of the original data, better reflect periodic items, random items and trend items of the original data, and accurately reflect characteristics of the original data.

Wherein, IMF1-IMF2 presents an unstable and oscillating curve, belonging to random items; IMF3-IMF4 presents a smooth, frequency-decreasing, periodic trend, belonging to the trend term. Thus, EMD decomposition can highlight the local features of the original photovoltaic power generation power sequence.

Evaluation index:

in order to verify the validity and accuracy of the present predictive model, both MAPE and RMSE are used as error indicators.

Where yoi is the i-th true value (observed) in the sample, ypi is the i-th predicted value (predicted) in the sample, and the smaller the two values, the higher the accuracy.

Embodiment two: the photovoltaic power calculation method based on EMD-AO-DELM in the second embodiment is an improvement based on the above embodiment, and the technical content disclosed in the above embodiment is not repeated, and the disclosure in the above embodiment also belongs to the disclosure in the second embodiment

Referring to fig. 4-9, one embodiment of the present invention is shown: a photovoltaic power calculation method based on EMD-AO-DELM comprises the following steps: s1: cleaning abnormal data: during the actual sampling process of the photovoltaic power station, some abnormal data can be generated. Abnormal data can lead to poor fitting of the prediction model and poor generalization ability. Therefore, it must be cleaned.

The detection of abnormal data adopts the 3 sigma rule principle:where X represents the photovoltaic power initial data,mean value of photovoltaic power initial data is shown. Statistically, the 3 sigma criterion is the percentage within one, two, three standard deviations from the mean in a normal distribution, with more accurate figures being 68.27%, 95.45% and 99.73%.

Data normalization: the same data unit range is different, which affects the fitting speed of the model and is unfavorable for model training. To improve model prediction accuracy, the initial photovoltaic power data is normalized, and the normalized power value is kept at [0, 1]]Between them. Wherein, under the normalization formula:where yi represents normalized data and xi represents the original power data value.

In this embodiment, simulation results: simulation analysis is carried out on the proposed control strategy by adopting Matlab R2022a, and the hidden layer number of the DELM model is set to be 2 in the prediction model established in the text; the number of hidden layer nodes is 5 and 5 respectively; the AO maximum iteration number is 200; the population number is 20; the regularization coefficient is set to infinity. For fairness, the DELM model parameters for comparison are implicitly layer by layer 2; the number of hidden layer nodes is 5 and 5 respectively. The annual data is divided into four quarters on average, and the data for each quarter is calculated as 19:1 are respectively divided into a training set and a testing set, and normalization processing is carried out. Simulation verification is performed on each quarter, wherein the test set takes 8 to 19 daily sunlight time periods with the step length of 1h, and the prediction results of the first to fourth quarters are shown in fig. 7.

In the present embodiment shown in connection with fig. 1-4:

from the view of fig. 7 and the table above, the prediction accuracy is highest in the two quarters of S2 and S3, mainly the land is relatively stable in the sunlight in the two quarters of S2 and S3, and the sunny day is most. But the average solar radiation is lower than the S1 quarter, resulting in a lower overall peak generated power than the S1 quarter. The prediction accuracy of S1 and S2 is relatively poor because the weather of the two seasons of the land is unstable, and the gust cloudy weather is more, particularly on the quarter of S2. The four quarter photovoltaic powers all show typical normal distribution, conform to the actual power situation, and the prediction model is accurate.

In summary, through the simulation diagram and the error analysis of fig. 8, compared with the initial DELM model and the initial AO-DELM model, the EMD-AO-DELM model disclosed herein has significantly improved prediction accuracy, better model stability, and various indexes significantly better than those of the other two algorithms. The method can meet the actual photovoltaic power prediction requirement, and can be matched with photovoltaic grid-connected scheduling work better.

In the embodiment, photovoltaic power generation prediction conditions of four quarters of a year are analyzed, correlation analysis is performed on photovoltaic power generation influence factor analysis, EMD decomposition is performed on photovoltaic historical power generation power, each IMF component is respectively input into an AO-DELM model, and finally, the component results are summed to obtain a prediction result. And then a photovoltaic power prediction model based on EMD-AO-DELM is provided, and the following conclusion is obtained through simulation result analysis:

among photovoltaic power generation influencing factors, the total illumination radiation and cloud quantity are positively correlated with photovoltaic power, and a key effect is played on a final prediction result; air pressure and humidity are inversely related to photovoltaic power, and are not suitable as input data in actual power prediction.

Aiming at the characteristics of volatility and randomness of photovoltaic power, EMD decomposition is carried out on historical photovoltaic power data, all components are mutually independent, prediction is carried out respectively, and finally superposition and summation are carried out. Experiments prove that the prediction effect is better after the EMD decomposition method is adopted.

The methods herein perform better than the AO-DELM and DELM models in four quarters of a year. Of these, the highest prediction accuracy is found in both quarters S2 and S3, which is related to local weather conditions. The weather is stable, and the accuracy of the quarter prediction with high sunny occupation ratio is higher.

It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential characteristics thereof. The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.

Furthermore, it should be understood that although the present disclosure describes embodiments, not every embodiment is provided with a separate embodiment, and that this description is provided for clarity only, and that the disclosure is not limited to the embodiments described in detail below, and that the embodiments described in the examples may be combined as appropriate to form other embodiments that will be apparent to those skilled in the art.

Claims

1. A photovoltaic power calculation method based on EMD-AO-DELM is characterized in that: the method comprises the following steps:

s4: and verifying the validity and accuracy of the calculation model.

2. The EMD-AO-DELM based photovoltaic power calculation method according to claim 1, wherein: the step S1 includes: s11: introducing Pearson correlation coefficient through the formulaCalculating a linear relationship between distance variables for measuring whether two data sets are on a line, wherein r>1 represents positive correlation between the two, r<1 represents that the two are in negative correlation; xi and yi represent the values of the ith of two factors, respectively; />And->Each representing an average of 2 factors.

3. The EMD-AO-DELM based photovoltaic power calculation method according to claim 1, wherein: the step S2 includes: the calculation formula is fitness=mse (train) +mse (test).

4. A method of EMD-AO-DELM based photovoltaic power calculation according to claim 3, wherein: the step S2 includes: s21: data cleaning is carried out, and abnormal values caused by errors in sampling some historical photovoltaic power data are proposed; s22: normalizing the cleaned sample data; s23: initializing AO algorithm parameters, including population scale, maximum iteration times T, and exploring and developing parameters alpha and delta; s24, initializing a population position X, initial population fitness and optimal individuals; s25: the method comprises the steps of sequentially carrying out an expansion exploration phase, a reduction exploration phase, an expansion development phase and a reduction development phase, and continuously updating the population positions; s26: calculating the fitness of the updated population to obtain the current optimal individual position and fitness, comparing the current optimal individual with the optimal individual fitness found until the t generation, and reserving the optimal individual position; s27: judging whether the maximum iteration times or solving conditions are reached, if so, outputting an optimal value, and if not, returning to the step S25; s28: and inputting the final optimized weight value result into the DELM model.

5. The EMD-AO-DELM based photovoltaic power calculation method according to claim 1, wherein: the step S3 comprises the following steps: s31: decomposing the photovoltaic historical data by adopting EMD to obtain a group of IMF components; s32: respectively establishing an AO-DELM model for each IMF component, and predicting each component; s33: and superposing the prediction results of the subsequences and verifying the accuracy of model prediction.

6. The EMD-AO-DELM based photovoltaic power calculation method according to claim 1, wherein: the step S4 includes: adopts both MAPE and RMSE as error indexes, and the MAPE calculation formula is as followsThe calculation formula of RMSE is +.>Where yoi is the i-th true value in the sample, ypi is the i-th predicted value in the sample, and the smaller the two values, the higher the accuracy.