CN112734073A

CN112734073A - Photovoltaic power generation short-term prediction method based on long and short-term memory network

Info

Publication number: CN112734073A
Application number: CN201911031539.2A
Authority: CN
Inventors: 时珉; 王珏; 张沛; 尹瑞; 王铁强; 王一峰; 王彦棡; 田剑刚; 王晓光
Original assignee: Tianjin Hongyuan Huineng Technology Co ltd; State Grid Corp of China SGCC; Computer Network Information Center of CAS; State Grid Hebei Electric Power Co Ltd
Current assignee: Tianjin Hongyuan Huineng Technology Co ltd; State Grid Corp of China SGCC; Computer Network Information Center of CAS; State Grid Hebei Electric Power Co Ltd
Priority date: 2019-10-28
Filing date: 2019-10-28
Publication date: 2021-04-30

Abstract

The invention relates to a photovoltaic power generation short-term prediction method based on a long-term and short-term memory network, which comprises the steps of determining weather data characteristics highly related to power generation power by using Pearson correlation coefficients, establishing a nonlinear relation between input weather data characteristics and output power generation power by using an LSTM network model, capturing the correlation of time series, dividing an original input data set aiming at seasons and weather types, improving the prediction precision of the model, and carrying out linear fitting processing on LSTM model prediction results based on the seasons and the weather types to further improve the accuracy of model prediction.

Description

Photovoltaic power generation short-term prediction method based on long and short-term memory network

Technical Field

The invention relates to the field of photovoltaic power generation, in particular to a photovoltaic power generation power short-term prediction method based on a long-term and short-term memory network.

Background

Photovoltaic power generation is a technology for directly converting light energy into electric energy by utilizing the photovoltaic effect of a semiconductor interface. The photovoltaic power generation has the advantages of cleanness, environmental protection, land saving, low investment cost and the like, and the accurate prediction of the photovoltaic power generation power can help a power grid dispatching department to adjust a dispatching plan in time, so that the running safety, economy and stability of a power grid are improved. The short-term prediction time scale of the photovoltaic power generation power is 1 day to 3 days. The current methods for predicting photovoltaic power generation power are divided into physical methods and statistical methods. The physical method is mainly used for predicting the photovoltaic power generation power based on physical equations such as a photovoltaic module operation equation and a solar radiation transfer equation, and the statistical method is used for fitting the power characteristics based on actual operation data of a power station.

Long-short term memory networks (LSTM) are a class of recurrent neural networks used to process time series data. The core design of the LSTM model is to realize long-term and short-term memory by using the cell state, thereby solving the problems of gradient extinction and gradient explosion in the recurrent neural network. LSTM contain many cells with self-attachment that are capable of storing the temporal state of the model. When the input information enters the LSTM network, the Cell state Cell determines whether the information is useful or not. There are three gates in a Cell, namely an input gate, a forgetting gate and an output gate. The input gate control information is activated and input to the memory unit. The left gate controls the retention state of cell history information, and the output gate control unit activates the output flow to the subsequent cell, so that information which depends on the cell for a long time can be well learned.

Disclosure of Invention

The invention aims to provide a photovoltaic power generation power short-term prediction method based on a long and short-term memory network, so that the accuracy of photovoltaic power generation power prediction is improved.

The technical scheme adopted by the invention for solving the technical problems is that,

in a first aspect, a method for short-term photovoltaic power generation power prediction is provided, and the method includes:

calculating the correlation between each characteristic of weather forecast data and photovoltaic power generation power of a photovoltaic power station according to each characteristic of the weather forecast data of the photovoltaic power station and the photovoltaic power generation power of the photovoltaic power station;

selecting a plurality of weather forecast data features with high correlation to form an input vector of a photovoltaic power generation power short-term prediction model according to the correlation calculation result;

dividing a data set formed by the input vectors according to the lunar calendar information and the weather type to obtain a plurality of subdata sets based on seasons and the weather type;

obtaining a predicted value of the photovoltaic power generation power according to the season and weather type subdata sets and a photovoltaic power generation power short-term prediction model; the photovoltaic power generation short-term prediction model is obtained through training.

Preferably, the calculating of the correlation between each feature of the weather forecast data and the photovoltaic power generation power includes calculating the correlation between each feature of the weather forecast data and the photovoltaic power generation power using a Pearson correlation coefficient formula.

Preferably, the photovoltaic power generation short-term prediction model is a recurrent neural network (LSTM) long-term and short-term memory model.

Preferably, the performing seasonal division on the data set of the input vector according to the lunar calendar information includes performing division according to spring, summer, autumn and winter to obtain a seasonal sub-data set; wherein, the spring from spring to summer, the summer from summer to autumn, the autumn to winter and the winter are the autumn, and the other is the winter.

Specifically, each season subdata set is trained to obtain a season-based LSTM photovoltaic power generation short-term prediction model.

More specifically, the photovoltaic power generation short-term prediction model is a double-layer LSTM model, the input layer is the input vector, the hidden layer performs feature extraction and feature regression, and the output layer is the predicted value of the photovoltaic power generation power.

Preferably, the data set of the input vector is divided according to weather types, and the method is that the data set of the input vector is divided according to the weather types based on the weather forecast information every day, and the weather types are divided into three categories, namely, sunny days, cloudy days/cloudy days, rainy days/snowy days, so as to obtain the weather type sub data set.

Specifically, each weather type subdata set is trained to obtain an LSTM photovoltaic power generation short-term prediction model based on the weather type.

Preferably, the characteristic data set of the photovoltaic power station numerical weather forecast data is subdivided according to the seasons and the weather types by combining the seasons and the weather types to obtain the sub data sets of the seasons and the weather types.

Specifically, each sub data set of the seasons and the weather types is trained to obtain an LSTM photovoltaic power generation short-term prediction model based on the seasons and the weather types.

Preferably, the predicted value of the photovoltaic power generation power is corrected through a fitting model.

In particular, the fitted model is a unary linear model; and the fitting parameters of the fitting model are obtained by automatic learning according to the predicted value and the true value of the photovoltaic power generation power.

In a second aspect, a method for training a photovoltaic power generation short-term prediction model is provided, the method comprising:

training a photovoltaic power generation power short-term prediction model based on the seasons and the weather types according to the season and weather type subdata sets and historical photovoltaic power generation power; and the photovoltaic power generation short-term prediction model obtains a predicted value of the photovoltaic power generation power according to the season and weather type subdata sets.

The photovoltaic power generation short-term prediction method based on the long and short-term memory network has the following advantages that:

1) characteristics in the photovoltaic power station weather data set which are highly related to the generated power can be more accurately determined by using Pearson correlation coefficients.

2) The LSTM network model is used for not only establishing a nonlinear relation between input characteristics and output power, but also capturing the correlation of time series.

3) And the original data sets are divided according to seasons and weather types, so that the model prediction precision is improved.

4) And linear fitting processing is carried out on the LSTM model prediction result based on the season and the weather type, so that the model prediction precision is further improved.

Drawings

Fig. 1 is a flowchart of a short-term photovoltaic power generation power prediction method based on a long-term and short-term memory network according to an embodiment of the present invention;

FIG. 2 is a schematic structural diagram of an LSTM unit according to an embodiment of the present invention;

FIG. 3 is a flow chart of a two-layer LSTM network model training provided by an embodiment of the present invention;

fig. 4 is a Pearson correlation coefficient diagram provided by an embodiment of the present invention;

fig. 5 is a diagram of spring and summer after linear fitting of the model prediction results based on season and weather type according to the embodiment of the present invention;

FIG. 6 is a subdivision flow diagram of one embodiment of the present invention.

The technical solution of the present invention is further described in detail by the accompanying drawings and embodiments.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail with reference to the accompanying drawings, and it is apparent that the described embodiments are only a part of the embodiments of the present invention, not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

Fig. 1 is a flowchart of a short-term photovoltaic power generation power prediction method based on a long-term and short-term memory network according to an embodiment of the present invention. As shown, the method comprises the following steps:

step 110, calculating the correlation between the photovoltaic power station numerical weather forecast data and the photovoltaic power generation power by using a Pearson correlation coefficient formula;

specifically, the Pearson Correlation Coefficient (Pearson Correlation Coefficient) is used to measure the linear relationship between two data set variables. The Pearson correlation coefficient formula is shown below, and the correlation coefficient r describes the degree of linear correlation between the variable X and the variable Y, which has a value range of [ -1,1 ]. The closer the absolute value of the correlation coefficient r is to 1, the stronger the correlation between the two sets of variables. The closer the correlation coefficient r is to 0, the weaker the correlation. And calculating the correlation between each characteristic of the photovoltaic power station numerical weather forecast data and the photovoltaic power generation power by using a Pearson correlation coefficient formula.

In the above equation, cov (X, Y) represents the covariance of two sets of vectors X and Y, and σ X, σ Y represent the standard deviation of vectors X, Y, respectively.

Respectively, are the mean values of X and Y.

Step 120, selecting the numerical weather forecast data features with high correlation to form an input vector of the LSTM model based on the correlation coefficient calculated by the Pearson correlation coefficient formula in the step 110;

step 130, dividing the input vector data set according to the season of the lunar calendar and the weather type information of the weather forecast to obtain subdata sets divided according to the season and the weather type;

in one embodiment, in order to highlight the influence of seasonal changes on photovoltaic power generation power, the data sets are divided into seasons according to the lunar calendar information, wherein spring to summer are spring days, summer to autumn is summer, autumn to winter is autumn days, and the rest is winter, each subdata set is trained, and a season-based LSTM photovoltaic power generation short-term prediction model is established.

In another embodiment, to verify the effect of weather type on photovoltaic power generation, the raw data set is divided into sunny, cloudy, rainy, and snowy data sets by weather type based on daily weather forecast information for a weather forecast network. The similarity between cloudy days and between rainy days and snowy days is higher, so the weather types are divided into three categories, namely sunny days, cloudy/cloudy days and rainy/snowy days. And training each subdata set to form an LSTM photovoltaic power generation short-term prediction model based on the weather type.

In another embodiment, in order to further improve the model and improve the photovoltaic power generation power prediction accuracy, two factors of seasons and weather types are combined, the original data set is subdivided based on the seasons and the weather types, each subdata set is trained, and an LSTM photovoltaic power generation power short-term prediction model based on the seasons and the weather types is established.

Step 140, training a photovoltaic power generation power short-term prediction model based on seasons and weather types according to the season and weather type subdata sets and historical photovoltaic power generation power; the photovoltaic power generation short-term prediction model obtains a predicted value of the photovoltaic power generation power according to the season and the weather type subdata set.

Photovoltaic power generation is a continuous process. The photovoltaic power generation power at each moment depends not only on the input characteristics at the present moment but also on the input characteristics at the past moment. LSTM is a recurrent neural network used to process time series data. The LSTM network model consists of an input layer, a hidden layer and an output layer, the hidden layer is no longer a normal neural unit, but an LSTM unit with unique memory cells and 3 "gates" to control the state of the memory cells. The 3 "gates" are respectively: an input gate, a forgetting gate and an output gate. LSTM relies on unique memory cells and "gate" structures to solve the problem of short-term RNN memory, and can remember more distant information. The LSTM cell structure is shown in fig. 2.

The calculation formula between the variables in fig. 2 is as follows:

f_t＝σ(W_f*(h_t-1，x_t)+b_f)

i_t＝σ(W_i*(h_t-1，x_t)+b_i)

C_t＝tanh(W_C*(h_t-1，x_t)+b_C)

C_t＝f_t*C_t-1+i_t*C_t

o_t＝σ(W_o*(h_t-1，x_t)+b_o)

h_t＝o_t*tanh(C_t)

y_t＝σ(W*h_t)

wherein: f. of_t、i_t、o_t、C_tVector values of the states of the forgetting gate, the input gate, the output gate and the memory cell, W_f、W_i、W_o、W_CWeight coefficients of forgetting gate, input gate, output gate, memory cell, b_f、b_i、b_o、b_CThe offset vectors of a forgetting gate, an input gate, an output gate and a memory cell are provided, sigma is an activation function, generally a sigmoid function, and tanh is a hyperbolic tangent function. Data x at the current time t_tAfter input, state c of the previous cell_t-1Will be updated to generate the state c of the current time_t。h_tSeen as a short-term state, c_tAnd the information is regarded as a long-term state, so that the information which depends on the long term can be well learned. y is_tIs the output of the current time.

In one embodiment of the invention, a two-layer LSTM model is used to make a point-by-point prediction of photovoltaic power generation. The LSTM model input layer is a feature vector composed of features selected by Pearson correlation coefficients. The model hidden layer can be abstracted into a feature extraction part and a feature regression part, wherein the feature extraction part is completed by an LSTM deep learning method, and the feature regression is completed by a full connection layer.

In another embodiment, a model training flow diagram is shown in FIG. 3, model input data x₁x₂......x₉₆Firstly, after normalization processing, sending the data into a double-layer LSTM network to dynamically capture a time sequence of input data, and then dynamically outputting a predicted value y of the photovoltaic generating power through a full-connection layer with a RuLU activation function₁y₂…y₉₆. The photovoltaic power generation power predicted by the model needs to be subjected to inverse normalization processing, so that the photovoltaic power generation power has practical significance. The photovoltaic power generation short-term prediction model based on the LSTM not only establishes a nonlinear relation between input characteristics and output power, but also captures the correlation of time series.

And 150, performing linear fitting on the predicted value and the true value of the photovoltaic power generation power, obtaining a fitting parameter value of a linear fitting model according to the historical predicted value and the true value, and fitting the predicted value of the photovoltaic power generation power output by the short-term photovoltaic power generation power prediction model by using the fitting model to obtain the fitted predicted value which is closer to the actual photovoltaic power generation power.

Specifically, because a certain error exists between the predicted value and the true value of the photovoltaic power generation predicted by the LSTM model, the predicted value and the true value are subjected to linear fitting. Because the predicted value and the true value are in a linear relation, a unary linear model is used for fitting the predicted value and the true value, and the formula is as follows:

Y＝a*y+b

and Y is the actual operating power of the photovoltaic power generation, and Y is a predicted value predicted by the LSTM model. The unary linear model can automatically learn fitting parameters a and b between the real power and the predicted power. And performing linear fitting on the value Y predicted by the LSTM model in the future and the already-obtained parameters a and b to obtain a theoretical power value closer to the real power.

The method has the following effects:

the effect of the method of the invention is further illustrated below using the actual data of one embodiment. Taking Ningxia Shizui photovoltaic power station as an example, the data set comprises photovoltaic power information and meteorological data of 15 minutes by year from 1/2017 to 12/31/2017. 96 time points of data per day, for a total of 35040 data.

(1) Data feature selection

Correlation between each characteristic of the photovoltaic power station numerical weather forecast data and photovoltaic power generation power, that is, correlation between solar short wave radiation, solar long wave radiation, surface air pressure, precipitation, surface wind speed, surface wind direction, solar altitude angle, solar azimuth angle and photovoltaic power generation power, is calculated using the Pearson correlation coefficient formula, as shown in fig. 4.

Based on the correlation coefficient calculated by the Pearson correlation coefficient formula, selecting the numerical weather forecast data characteristics with the absolute value larger than 0.2 to form the input vector of the LSTM model, namely the solar short wave radiation, the solar altitude angle and the environment humidity.

(2) Prediction model accuracy

Use ofTwo indicators, root Mean square error rmse (root Mean Squared error) and Mean Absolute error mae (Mean Absolute error), evaluate the accuracy of the prediction model. Wherein N is the number of samples in the test set, y_iAs actual value of power, Y_iIs a predicted value.

Aiming at the characteristic vectors, a season-based LSTM photovoltaic power generation short-term prediction model, a weather-type-based LSTM photovoltaic power generation short-term prediction model and a season-and weather-type-based LSTM photovoltaic power generation short-term prediction model are respectively established (as shown in FIG. 6). The results of the model experiments are shown below.

Table.1 LSTM photovoltaic power generation short-term prediction model result based on full data set

Table.2 LSTM photovoltaic power generation short-term prediction model result based on seasons

Season	RMSE(MW)	MAE(MW)
			Spring	11.20	5.45
Summer	11.44	5.68
			Autumn	12.10	6.02
In winter	11.16	4.99

Table.3 weather-type-based LSTM photovoltaic power generation short-term prediction model result

Weather type	RMSE(MW)	MAE(MW)
			In sunny days	8.37	4.16
Cloudy/cloudy day	9.20	4.27
			Rainy/snowy day	9.97	5.46

Table.4 LSTM photovoltaic power generation short-term prediction model result based on seasons and weather types

Therefore, the LSTM short-term photovoltaic power prediction model based on the seasons and the weather types can effectively improve the photovoltaic power generation power prediction accuracy.

(3) Performing linear fitting on the prediction result

And performing linear fitting on the model prediction result based on the season and the weather type, and further improving the model prediction precision.

The formula for the linear fit is: y is 0.96 x Y +0.07

The linear fit of the spring summer results is shown in fig. 5.

The photovoltaic power generation short-term prediction method based on the long and short-term memory network provided by the embodiment of the invention determines weather data characteristics highly related to power generation power by using a Pearson correlation coefficient; the LSTM network model is used for not only establishing a nonlinear relation between input characteristics and output power, but also capturing the correlation of a time sequence; original data sets are divided according to seasons and weather types, and model prediction accuracy is improved; and linear fitting processing is carried out on the LSTM model prediction result based on the season and the weather type, so that the model prediction precision is further improved. By applying the method provided by the embodiment of the invention, the influence factors of the predicted power of the photovoltaic power station, which need to be considered, can be determined, and the feature vector is formed by selecting the features with higher correlation; reasonably modeling the characteristic vector, and properly splitting the data set to obtain higher precision; post-processing the predicted value obtained by the model to obtain a theoretical power value closer to the true value

The foregoing description has been directed to specific embodiments of this disclosure. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing may also be possible or may be advantageous.

Those of skill would further appreciate that the various illustrative components and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both, and that the various illustrative components and steps have been described above generally in terms of their functionality in order to clearly illustrate this interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.

The steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied in hardware, a software module executed by a processor, or a combination of the two. A software module may reside in Random Access Memory (RAM), memory, Read Only Memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

The above-mentioned embodiments are intended to illustrate the objects, technical solutions and advantages of the present invention in further detail, and it should be understood that the above-mentioned embodiments are merely exemplary embodiments of the present invention, and are not intended to limit the scope of the present invention, and any modifications, equivalent substitutions, improvements and the like made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims

1. A method for short-term prediction of photovoltaic power generation, the method comprising:

2. The method of claim 1, wherein the calculating the correlation between the respective characteristics of the weather forecast data and the photovoltaic power generation power comprises calculating the correlation between the respective characteristics of the weather forecast data and the photovoltaic power generation power using a Pearson correlation coefficient formula.

3. The method of claim 1, wherein the photovoltaic power generation short-term prediction model is a recurrent neural network (LSTM) long-term short-term memory model.

4. The method of claim 3, wherein the photovoltaic power generation short-term prediction model is a two-layer LSTM model, an input layer is the input vector, a hidden layer performs feature extraction and feature regression, and an output layer is a predicted value of photovoltaic power generation.

5. The method of claim 1, wherein the seasonal division of the data set of the input vector according to the lunar calendar information comprises dividing by spring, summer, fall, and winter to obtain a set of seasonal sub-data; wherein, the spring from spring to summer, the summer from summer to autumn, the autumn to winter and the winter are the autumn, and the other is the winter.

6. The method of claim 4, wherein each of the seasonal sub-data sets is trained to derive a seasonal-based LSTM photovoltaic power generation short-term prediction model.

7. The method of claim 1, wherein the dataset of input vectors is partitioned according to weather type by dividing the dataset of input vectors according to weather type based on weather forecast information every day, and the weather type is partitioned into three categories, namely sunny days, cloudy/cloudy days, rainy days, and snowy days, to obtain the weather type sub dataset.

8. The method of claim 6, wherein each of the weather-type sub-datasets is trained to derive a weather-type based LSTM photovoltaic power generation short-term prediction model.

9. The method of claim 1, wherein the feature data sets of the photovoltaic plant numerical weather forecast data are subdivided by season and weather type in combination with the season and the weather type to obtain sub-sets of season and weather type data.

10. The method of claim 8, wherein each of the sub-sets of season and weather type data is trained to derive a season and weather type based LSTM photovoltaic power generation short-term prediction model.

11. The method according to claim 1, characterized in that the predicted value of the photovoltaic power generation power is corrected by fitting a model.

12. The method of claim 10, wherein the fitting model is a unary linear model; and the fitting parameters of the fitting model are obtained by automatic learning according to the predicted value and the true value of the photovoltaic power generation power.

13. A method for training a photovoltaic power generation short-term prediction model is characterized by comprising the following steps: