CN116128148A

CN116128148A - Photovoltaic power station output prediction network training method, prediction method, device and equipment

Info

Publication number: CN116128148A
Application number: CN202310182436.6A
Authority: CN
Inventors: 赵俊炜; 陈凤超; 张鑫; 苏俊妮; 邱泽坚; 何毅鹏; 周立德; 徐睿烽; 刘铮; 饶欢
Original assignee: Guangdong Power Grid Co Ltd; Dongguan Power Supply Bureau of Guangdong Power Grid Co Ltd
Current assignee: Guangdong Power Grid Co Ltd; Dongguan Power Supply Bureau of Guangdong Power Grid Co Ltd
Priority date: 2023-02-28
Filing date: 2023-02-28
Publication date: 2023-05-16

Abstract

The embodiment of the invention discloses a photovoltaic power station output prediction network training method, a prediction method, a device and equipment. The training method comprises the following steps: acquiring historical output data and corresponding meteorological data of a photovoltaic power station; generating input data of a generator in the countermeasure generation network is determined according to the meteorological data, and the generated input data is input into the generator to obtain generated output data corresponding to the historical output data; determining discrimination input data of a discriminator in the countermeasure generation network according to the historical output data, the generated output data and the meteorological data, and inputting the discrimination input data into the discriminator to obtain a discrimination result; updating the discriminator by adopting a discriminator loss function according to the discrimination result, updating a generator by adopting a generator loss function, and performing iterative training until the countermeasure generation network is stable, thus obtaining the photovoltaic power station output prediction network. According to the method, the antagonism generation network is adopted in the photovoltaic power station output prediction, and meteorological data is considered, so that a prediction network can be provided, and the high-precision prediction of the photovoltaic power station output can be realized.

Description

Photovoltaic power station output prediction network training method, prediction method, device and equipment

Technical Field

The invention relates to the technical field of deep learning, in particular to a photovoltaic power station output prediction network training method, a prediction device and equipment.

Background

To achieve the peak carbon neutralization goal, conventional power systems are shifting from fossil energy as the primary energy source to clean energy as the primary energy source. As the proportion of photovoltaic power generation in new energy power generation is gradually increased, the instability of power generation of the photovoltaic power generation brings challenges to the stable operation of a power system. Therefore, output prediction of photovoltaic power plants is crucial.

In the prior art, the output prediction of a photovoltaic power station is performed by adopting a statistical method. Among other common statistical methods are: the Monte Carlo method, copula function (a kind of connection function) sampling, etc. The Monte Carlo method needs to assume that solar irradiance is subject to preset probability distribution, and under the condition, output prediction is carried out by utilizing irradiance and a photovoltaic output power formula. However, in reality the probability distribution of solar irradiance is unknown. Deviations between the assumed preset probability distribution and the unknown probability distribution will directly lead to deviations in the output predictions. In the Copula function sampling method, copula functions have a plurality of different forms, and the selection of the functions can influence the accuracy of the output prediction of the photovoltaic power station.

Disclosure of Invention

The invention provides a photovoltaic power station output prediction network training method, a prediction device and equipment, which are used for providing a prediction network for predicting the output of a photovoltaic power station with high precision, so that the effect that the output prediction result of the photovoltaic power station is closer to the actual output power of the photovoltaic power station is achieved.

According to an aspect of the present invention, there is provided a training method of a photovoltaic power station output prediction network, the method comprising:

acquiring historical output data of a photovoltaic power station, and meteorological data corresponding to the historical output data;

determining generation input data of a generator in an countermeasure generation network according to the meteorological data, and inputting the generation input data into the generator to obtain generation output data corresponding to the historical output data;

determining discrimination input data of a discriminator in the countermeasure generation network according to the historical output data, the corresponding generated output data and the meteorological data, and inputting the discrimination input data into the discriminator to obtain a discrimination result;

and updating the discriminator by adopting a discriminator loss function according to the discrimination result, updating a generator by adopting a generator loss function, and performing iterative training until the countermeasure generation network is stable, thereby obtaining the photovoltaic power station output prediction network.

Optionally, determining the generation input data of the generator in the countermeasure generation network according to the meteorological data includes:

mapping, by an encoder, the meteorological data to an initial meteorological feature;

determining the node quantity of the target meteorological features according to the association relation between the initial meteorological features;

mapping, by an encoder, the meteorological data to target meteorological features corresponding to the number of nodes;

and splicing the target meteorological features and the preset noise features to obtain the generated input data of the generator in the countermeasure generation network.

Optionally, updating the discriminator by using a loss function of the discriminator and updating a generator by using a loss function of the generator according to the discrimination result, and performing iterative training until the countermeasure generation network is stable, to obtain a photovoltaic power station output prediction network, including:

and updating the discriminator by using a discriminator loss function according to the discrimination result, updating a generator and the encoder by using a generator loss function, and performing iterative training until the countermeasure generation network is stable, thereby obtaining the photovoltaic power station output prediction network.

Optionally, determining the discrimination input data of the discriminator in the countermeasure generation network according to the historical output data, the corresponding generated output data, and the meteorological data includes:

Splicing the historical output data and target meteorological features corresponding to the meteorological data to be used as first input data for distinguishing;

splicing the generated output data and the target meteorological features to be used as second input data for distinguishing;

and using the first distinguishing input data and the second distinguishing input data as the distinguishing input data.

Optionally, before updating the discriminator with the discriminator loss function and updating the generator with the generator loss function according to the discrimination result, the method further includes:

determining a loss function of the discriminator according to the distribution distance between the historical output data and the corresponding generated output data and Li Puxi Roots condition;

and determining a generator loss function according to the output result of the discriminator corresponding to the generated output data.

Optionally, the encoder is a fully connected neural network; wherein:

the first layer activation function in the fully-connected neural network of the encoder is a leak ReLU function; and the last layer activation function is the tanh function.

According to another aspect of the present invention, there is provided a photovoltaic power plant output prediction method, the method comprising: the training method of the photovoltaic power station output prediction network provided by any embodiment of the invention is adopted to generate the photovoltaic power station output prediction network;

And inputting the current day output data and the next day weather forecast data of the photovoltaic power station into the photovoltaic power station output prediction network, and predicting to obtain the next day output data of the photovoltaic power station.

According to another aspect of the present invention, there is provided a training apparatus for a photovoltaic power plant output prediction network, the apparatus comprising:

the data acquisition module is used for acquiring historical output data of the photovoltaic power station and meteorological data corresponding to the historical output data;

the generated output data determining module is used for determining generated input data of a generator in an antagonism generation network according to the meteorological data, and inputting the generated input data into the generator to obtain generated output data corresponding to the historical output data;

the judging result determining module is used for determining judging input data of a judging device in the countermeasure generating network according to the historical output data, the corresponding generated output data and the meteorological data, and inputting the judging input data into the judging device to obtain a judging result;

and the photovoltaic power station output prediction network determining module is used for updating the discriminator by adopting a discriminator loss function according to the discrimination result, updating the generator by adopting a generator loss function, and performing iterative training until the countermeasure generation network is stable, so as to obtain the photovoltaic power station output prediction network.

According to another aspect of the present invention, there is provided a photovoltaic power plant output predicting apparatus, comprising:

the photovoltaic power station output prediction network generation module is used for generating a photovoltaic power station output prediction network by adopting the training method of the photovoltaic power station output prediction network provided by any embodiment of the invention;

the next-day output data prediction module is used for inputting the current-day output data of the photovoltaic power station and the next-day weather forecast data into the photovoltaic power station output prediction network, and predicting to obtain the next-day output data of the photovoltaic power station.

According to another aspect of the present invention, there is provided an electronic apparatus including:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein, the liquid crystal display device comprises a liquid crystal display device,

the memory stores a computer program executable by the at least one processor to enable the at least one processor to perform the training method of the photovoltaic power plant output prediction network of any embodiment of the present invention; or a photovoltaic power station output prediction method.

According to another aspect of the present invention, there is provided a computer readable storage medium storing computer instructions for causing a processor to execute a training method of the photovoltaic power plant output prediction network according to any embodiment of the present invention; or a photovoltaic power station output prediction method.

According to the technical scheme, historical output data and corresponding meteorological data of a photovoltaic power station are obtained; generating input data of a generator in the countermeasure generation network is determined according to the meteorological data, and the generated input data is input into the generator to obtain generated output data corresponding to the historical output data; determining discrimination input data of a discriminator in the countermeasure generation network according to the historical output data, the generated output data and the meteorological data, and inputting the discrimination input data into the discriminator to obtain a discrimination result; updating the discriminator by adopting a discriminator loss function according to the discrimination result, updating a generator by adopting a generator loss function, performing iterative training until the antagonism generation network is stable to obtain a photovoltaic power station output prediction network, solving the problem of predicting the photovoltaic power station output, and providing a prediction network to realize high-precision prediction of the photovoltaic power station output.

It should be understood that the description in this section is not intended to identify key or critical features of the embodiments of the invention or to delineate the scope of the invention. Other features of the present invention will become apparent from the description that follows.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a flowchart of a training method of a photovoltaic power plant output prediction network according to a first embodiment of the present invention;

FIG. 2 is a schematic diagram of an countermeasure generation network for training a photovoltaic power plant output prediction network according to a first embodiment of the present invention;

fig. 3 is a flowchart of a training method of a photovoltaic power station output prediction network according to a second embodiment of the present invention;

fig. 4 is a training flowchart of a photovoltaic power station output prediction network according to a second embodiment of the present invention;

fig. 5 is a flowchart of a photovoltaic power station output prediction method according to a third embodiment of the present invention;

fig. 6 is a schematic structural diagram of a training device of a photovoltaic power station output prediction network according to a fourth embodiment of the present invention;

fig. 7 is a schematic structural diagram of a photovoltaic power station output predicting device according to a fifth embodiment of the present invention;

FIG. 8 is a training method for a photovoltaic power plant output prediction network implementing an embodiment of the present invention; or, the structural schematic diagram of the electronic equipment of the photovoltaic power station output prediction method.

Detailed Description

In order that those skilled in the art will better understand the present invention, a technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in which it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without making any inventive effort, shall fall within the scope of the present invention.

It should be noted that the terms "first," "second," and the like in the description and the claims of the present invention and the above figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments of the invention described herein may be implemented in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

Example 1

Fig. 1 is a flowchart of a training method of a photovoltaic power station output power prediction network according to an embodiment of the present invention, where the embodiment may be suitable for a case of predicting output power of a photovoltaic power station, the method may be performed by a training device of the photovoltaic power station output power prediction network, the training device of the photovoltaic power station output power prediction network may be implemented in a form of hardware and/or software, and the training device of the photovoltaic power station output power prediction network may be configured in an electronic device, such as a computer. As shown in fig. 1, the method includes:

Step 110, historical output data of the photovoltaic power station and meteorological data corresponding to the historical output data are obtained.

In the embodiment of the invention, the output prediction of the photovoltaic power station can be understood as the output prediction of the photovoltaic power station. Specifically, the output prediction of the photovoltaic power station can be understood as scene generation of the photovoltaic power station, namely, time sequence scene generation of the generated energy of the photovoltaic power station can convert uncertainty of photovoltaic power generation in the power system into a limited determined time sequence scene according to probability distribution of the uncertainty, and therefore reference is provided for power system scheduling.

The historical output data can be data formed by the daily output power of the photovoltaic power station. The weather data may be actual weather data corresponding to the historical output data, or may be weather forecast data corresponding to the historical output data. More specifically, in the embodiment of the present invention, the historical output data may correspond to actual weather data of the day; alternatively, the historical output data may correspond to weather forecast data for the next day.

By way of example, the meteorological data may be data detected by a detection instrument built into the photovoltaic power plant and related to photovoltaic power generation. The meteorological data may include: one or more of precipitation, temperature, relative humidity, illumination intensity, 10m wind speed, 10m wind direction, 100m wind speed, and 100m wind direction. According to practical researches on the photovoltaic power station, the correlation coefficient of the illumination intensity is found to be the highest among the pearson correlation coefficients of the variables and the power generation amount of the photovoltaic power station. And, the correlation coefficient of the illumination intensity is far higher than that of other meteorological data. Accordingly, in embodiments of the present invention, the meteorological data may select an illumination intensity. When the illumination intensity is not present, solar irradiance may be selected as the meteorological data. By introducing meteorological data into the output prediction of the photovoltaic power station, the accuracy of the prediction can be improved.

And 120, determining generation input data of a generator in the countermeasure generation network according to the meteorological data, and inputting the generation input data into the generator to obtain generation output data corresponding to the historical output data.

In an embodiment of the present invention, the countermeasure generation network may include a generator and a arbiter. The generator may be used to generate output prediction data, i.e. to generate output data, of the photovoltaic power plant. The discriminator is used for discriminating output prediction data generated by the generator and real photovoltaic power station output data so as to guide the discriminator and the generator to update and generate a stable countermeasure generation network.

Wherein generating the input data may be determined from the meteorological data. For example, noise data may be added to the meteorological data to yield generated input data to improve the immunity and robustness of the generator.

In the embodiment of the invention, in order to improve generalization of the countermeasure network and prevent the generator from overfitting when the number of training sets is small, the encoder can be used for extracting the meteorological features of the meteorological data, and then the input data is determined and generated according to the meteorological features.

Fig. 2 is a schematic diagram of an countermeasure generation network for training a photovoltaic power station output prediction network according to a first embodiment of the present invention. In fig. 2, meteorological data w= (w ₁ ,w ₂ ,…,w _n ) Can be used as input data for the encoder CO. The CO output data may be meteorological features h= (h) ₁ ,h ₂ ,...,h _m ). The generated input data of the generator may be meteorological features h= (h) ₁ ,h ₂ ,…,h _m ) And preset noise z= (z) ₁ ,z ₂ ,…,z _p ) In the form of (z) ₁ ,z ₂ ,…,z _p ,h ₁ ,h ₂ ,…,h _m ). The preset noise may be a one-dimensional normal distribution vector within the interval (-1, 1). The generator can output and generate output data

And 130, determining discrimination input data of a discriminator in the countermeasure generation network according to the historical output data, the corresponding generated output data and the meteorological data, and inputting the discrimination input data into the discriminator to obtain a discrimination result.

Wherein the method comprises the steps ofThe discriminator needs to discriminate the generated output data and the historical output data. Thus, the discrimination input data may be determined from historical output data, generated output data, and weather data. Specifically, as shown in FIG. 2, determining the input data may include generating output data

And meteorological features h= (h) ₁ ,h ₂ ,…,h _m ) Is a concatenation of->

Historical output data x= (x) ₁ ,x ₂ ,,x _q ) And meteorological features h= (h) ₁ ,h ₂ ,…,h _m ) Is equal to xh= (x) ₁ ,x ₂ ,,x _q ,h ₁ ,h ₂ ,…,h _m ). The discrimination result output by the discriminator may be a node scalar.

And 140, updating the discriminator by using a discriminator loss function according to the discrimination result, updating a generator by using a generator loss function, and performing iterative training until the countermeasure generation network is stable, thereby obtaining the photovoltaic power station output prediction network.

In contrast generation networks, cross entropy loss functions are often employed. However, in the photovoltaic plant output prediction scenario, the historical output data input to the countermeasure generation network is one-dimensional vector data with high middle and low two sides. The historical output data are highly similar, and the space formed by taking any real value by one-dimensional vectors cannot be paved. The generated output data generated by the generator at the initial stage of training is a one-dimensional vector with random values, and the discriminator can easily distinguish the authenticity of the data and easily reach a stable state along with training. Regardless of what data the generator generates, the arbiter readily recognizes that the data generated by the generator is false, as long as the distribution of the generated data does not ensure that there are long enough segments that are identical to the distribution of the historical output data. Therefore, the loss function of the generator is always high, and meanwhile, the generator cannot be normally trained to achieve Nash equilibrium because the loss function cannot reflect the generation effect of the generator.

Therefore, the cross entropy loss function commonly used in the countermeasure generation network cannot be applied to the training method of the photovoltaic power station output prediction network in the embodiment of the invention. The loss function suitable for the photovoltaic power station output prediction scene needs to be determined through scientific research.

In the embodiment of the invention, the discriminator is updated by adopting a discriminator loss function, and the generator is updated by adopting a generator loss function. Wherein the arbiter loss function may relate to a distribution distance between the historical output data and the corresponding generated output data. The generator loss function may be related to the output of the arbiter. Therefore, the problem that the Nash equilibrium cannot be achieved by the antagonism generation network in the training process can be avoided.

The loss function is adopted to carry out parameter adjustment on the discriminator and the generator, further the next sample data is adopted to carry out iterative training, and the photovoltaic power station output prediction network is obtained when the countermeasure generation network is stable. Wherein, in the structure shown in FIG. 2, an encoder is added, and a loss function L of a discriminator can be adopted first when network training is carried out _D Updating the arbiter D, using the generator loss function L _G When updating the generator G, the generator loss function L is adopted together _G The encoder CO is updated. Therefore, the prediction accuracy of the photovoltaic power station output prediction network can be higher.

Example two

Fig. 3 is a flowchart of a training method of a photovoltaic power station output prediction network according to a second embodiment of the present invention, where the technical solution in this embodiment is further refined, and the technical solution in this embodiment may be combined with each of the alternatives in one or more embodiments. As shown in fig. 3, the method includes:

step 310, historical output data of the photovoltaic power station and meteorological data corresponding to the historical output data are obtained.

Step 320, mapping the meteorological data into initial meteorological features by an encoder; determining the node quantity of the target meteorological features according to the association relation between the initial meteorological features; the meteorological data is mapped by the encoder to target meteorological features corresponding to the number of nodes.

Wherein the encoder may map multi-node weather data into fewer-node weather features. For example, the encoder outputs a weather feature of h= (h ₁ ,h ₂ ,…,h _m ) Wherein the number of encoder output nodes is freely controllable, but desirably as few as possible is possible while being able to incorporate most of the meteorological features. The method aims at distributing different meteorological features in a dense space, so that the meteorological features generated by the encoder on newly input meteorological data in the subsequent use process are close to the features of the historical meteorological data, the generation resistance to network generalization is improved, and the risk of overfitting of the generator under the condition of less training sets can be reduced.

In order to map the meteorological data of multiple nodes into meteorological features of fewer nodes, the number of nodes of the target meteorological features can be determined according to the association relation between the initial meteorological features. For example, the initial meteorological feature of the encoder output is h= (h) ₁ ,h ₂ ,…,h _m ) At this time, the number of output nodes of the encoder is m. If m is too large, it may occur that for different weather data, there is always a linear relationship between some of the values in the weather features. For example, for different meteorological data w, h ₁ Always h ₂ 3 of (3)And 12 times, namely the two numbers can be increased or reduced at the same time. Or, for example, h ₁ Always with 0.5h ₂ +1.4h ₃ Approximately equal at this time h ₁ Has no meaning. Therefore, the number of output nodes of the encoder can be adjusted from m to m-1, and then whether the weather features output by the encoder have the similar phenomena can be judged. Similarly, the number of nodes of the target weather feature can be determined according to the association relation between the initial weather features, so that the weather data is mapped into the target weather feature corresponding to the number of nodes through the encoder.

Since the above method is too labor intensive, in embodiments of the present invention, a self-encoder network may also be trained using meteorological features h to verify that m is too large. If the self-encoder can restore h and the number of the coding layer nodes of the self-encoder is far smaller than m, the number of m needs to be reduced until the number of the coding layer nodes is more than half of that of m.

In an alternative implementation of the embodiment of the present invention, the encoder is a fully connected neural network; wherein: the first layer activation function in the fully-connected neural network of the encoder is a leak ReLU function; and the last layer activation function is the tanh function.

The activation functions of all layers are set according to the actual situation generated by the photovoltaic power generation scene. The tanh function is set at the last layer in order to limit the value of the encoder CO output node between-1 and +1, so that the noise z, which is common with the meteorological features, to the generator input can be chosen to be uniformly distributed between-1 and +1. The input of the neural network should ensure that the parts do not appear, for example, on the order of 1 ten thousand times different, so that a typical neural network will have a data preprocessing stage. In the embodiment of the invention, if the meteorological data and the photovoltaic output data are not wrong, the data preprocessing is not needed, and the training is more concise and convenient. The input to the encoder CO may be solar irradiance. Solar irradiance is a number greater than zero and may be greater than 1. If the sigmod function is selected for the activation function of the previous hidden layer and the input data is much greater than 1, the first layer value varies too much. In the embodiment of the invention, the activation function of the previous hidden function layer selects the leak ReLU function.

In the embodiment of the invention, the selection of the activation function is not unique. In practical application, the first layer activation function is a leak ReLU function, the second layer may be a ReLU function, the third layer may be a sigmod function, the last layer is a tanh function, and the activation function may be changed step by step to make the training effect better. However, not all activation functions are changed to sigmod functions for training.

In order for similar weather data to produce similar weather features, the node weights k and offsets γ in the encoder may be limited during training. In particular, the absolute values of the weights and biases employed by the various nodes of the layers in the fully connected neural network of the encoder may be limited to a set constant range.

Wherein, a neural network has several layers, each layer has several nodes, each node has a bias, and there is a weight between each node. For each node, the output is: y is ₁ ＝tanh(k ₁ x ₁ +k ₂ x ₂ +...+k _n x _n +γ)。(x ₁ ,x ₂ ,...,x _n ) For each node of the upper layer, k= (k ₁ ,k ₂ ,...,k _n ) For the weight between this node and the node of the previous layer, γ is the bias of this node, and tanh is the activation function of this node. All the fully-connected neural networks have the two parameters, and the training process of the neural networks is a process that k and gamma are continuously changed.

By way of example only, and not by way of limitation,

the constant c can be adjusted to any positive number according to the training effect. The constant c is generally in the interval [1,10 ]]And (3) inner part.

Step 330, splicing the target meteorological features and the preset noise features to obtain generated input data of a generator in the countermeasure generation network; and inputting the generated input data into a generator to obtain generated output data corresponding to the historical output data.

Wherein the generator is a fully connected neural network. The activation function of the generator may be a LeakyReLU function. To mitigate the overfitting phenomenon, the model can employ a random deactivation method during training. That is, each node will be deactivated with a certain probability during each training process of the neural network. It can be understood that: there is always a portion of the different node output values that are zero during the generator training process. And during training, the weight updating link does not update the weight and bias of the deactivated nodes in the round of training.

For example, k= (k) of fully connected neural network ₁ ,k ₂ ,...,k _n ) The updating mode of the gamma is as follows: for each k by neural network _i The gradient from the loss function to this variable is found, and the value of this loss function is brought into the gradient, thereby updating k and γ. At the time of updating, the non-deactivated node may be updated, i.e., k= (k) ₁ ,k ₂ ,...,k _n )→k'＝(k ₁ ',k' ₂ ,...,k' _n ) Gamma-gamma'; the inactivated node is not updated, the weight and the bias are unchanged before and after updating, and k= (k) ₁ ,k ₂ ,...,k _n )→k＝(k ₁ ,k ₂ ,...,k _n ) γ→γ, i.e. the last weight and last bias are still maintained. So that the neural network does not rely solely on the local features of the input features. In terms of implementation, dropout functions are only used when constructing a network under the Pytorch framework. The deactivation probability may be set to 0.3.

Step 340, splicing target meteorological features corresponding to the historical output data and the meteorological data to be used as first input data for distinguishing; and splicing the generated output data and the target meteorological features to be used as second input data for distinguishing.

Wherein, the first input data is determined as xh= (x) ₁ ,x ₂ ,,x _q ,h ₁ ,h ₂ ,…,h _m ). Discriminating the second input data as

Step 350, using the first input data and the second input data as the input data; and inputting the discrimination input data into the discriminator to obtain a discrimination result.

The arbiter adopts a fully-connected neural network, the activation function is Tanh, and the input layer is not provided with the activation function. The output of the discriminator is a node scalar, and the function is to generate the real output data distribution and the output data distribution distance of the generator, so as to guide the generator and the discriminator to update the network. Since the distance between the two distributions is difficult to calculate, the average distance between the plurality of samples conforming to the two distributions can be calculated in the actual calculation. I.e. calculating the once-loss function requires a plurality of history data and generator generated data.

For example, definition b _x ＝(b _x1 ,b _x2 ,…,b _xr ) Wherein b _x1 For sample 1 randomly drawn from the historical output data, the format is xh= (x) ₁ ,x ₂ ,,x _q ,h ₁ ,h ₂ ,…,h _m ) The method comprises the steps of carrying out a first treatment on the surface of the From b _x1 To b _xr Each vector has the same format. Similarly, define

A first sample of the generator's generated output data in the form of

The distance between the two distributions can be determined by

And (5) determining. />

Finger discriminator D is at the input +.>

Output value at that time. D is a neural network and can also be seen as a mapping from input to output, i.e. D can be seen as a function.

And step 360, updating the discriminator by using a discriminator loss function according to the discrimination result, updating a generator and an encoder by using a generator loss function, and performing iterative training until the countermeasure generation network is stable, thereby obtaining the photovoltaic power station output prediction network.

In an optional implementation manner of the embodiment of the present invention, before updating the arbiter with the arbiter loss function and updating the generator with the generator loss function according to the discrimination result, the method further includes: determining a loss function of the discriminator according to the distribution distance between the historical output data and the corresponding generated output data and the Li Puxi-z condition; and determining a generator loss function according to the output result of the discriminator corresponding to the generated output data.

In the embodiment of the invention, the arbiter loss function L for updating the arbiter _D May be

Wherein the arbiter D expects a loss function L _D The smaller the better. The first term in the formula indicates that when the identifier hopes that the input quantity is the historical output data, the score is as small as possible, and the generated output data distribution is close to the historical output data distribution. The second term in the formula indicates that when the discriminator hopes to input the generated output data, the numerical value behind the negative sign is as large as possible, and the second term indicates that the discriminator can generate the output data and the historical output data respectively. The third term in the formula indicates that the constraint discriminator satisfies the Li Puxi z condition. I.e. when the two distributions of the arbiter inputs do not change much, the scores that are desired to be generated do not change much. In (1) the->

Is a scalar, & a->

Similarly, the subtraction of the two can take absolute value. In (1) the->

For vectors, the calculation is as follows->

Shown is the mid vector b _xi Wherein i is 1 to r is b _x ＝(b _x1 ,b _x2 ,…,b _xr ) B of (b) _x1 ,b _x2 ,…,b _xr And->

Similarly, the difference between the two can be a two-norm. I denote computation the vector is used for the two norms of the vector, and (2)>

Epsilon in _i,j Is subject to [0,1 ]]Random numbers uniformly distributed in the interval. />

The result of (a) is a scalar. />

The result being a scalar, i.e. can be averaged

Where λ may be a constraint parameter set according to the actual situation of the network training, and may be a scalar, and may be a constant greater than 0 and less than 1. Set to 0, this indicates that the Li Puxi z condition of the network is not limiting.

The generator loss function is

The generator aims at letting the loss function L _G Minimum, L in training process _G The generator G and the encoder CO may be directed to update the parameters.

According to the technical scheme, historical output data of the photovoltaic power station and meteorological data corresponding to the historical output data are obtained; mapping, by an encoder, the meteorological data to an initial meteorological feature; determining the node quantity of the target meteorological features according to the association relation between the initial meteorological features; mapping the meteorological data into target meteorological features corresponding to the number of nodes through an encoder; splicing the target meteorological features and preset noise features to obtain generated input data of a generator in the countermeasure generation network; the generated input data is input into a generator to obtain generated output data corresponding to the historical output data; splicing target meteorological features corresponding to the historical output data and the meteorological data to be used as first input data for distinguishing; splicing the generated output data and the target meteorological features to be used as second input data for distinguishing; the first input data and the second input data are used as the input data; inputting the discrimination input data into a discriminator to obtain a discrimination result; updating a discriminator by adopting a discriminator loss function according to a discrimination result, updating a generator and an encoder by adopting a generator loss function, and performing iterative training until an antagonism generation network is stable, so as to obtain a photovoltaic power station output prediction network, solve the problem of predicting the photovoltaic power station output, provide a prediction network, realize high-precision prediction of the photovoltaic power station output, and avoid overfitting; because the adopted loss function of the discriminator and the loss function of the generator are different from the common loss function in the countermeasure generation network, the success rate of training of the countermeasure generation network can be improved, and the problem that training fails when the distribution of the generated output data of the generator is always different from the distribution of the historical output data is avoided. In addition, the countermeasure generation network shown in fig. 2 in the embodiment of the invention has a simple structure, is easy to train, and has fewer requirements on the number of training sets than other countermeasure networks.

Fig. 4 is a training flowchart of a photovoltaic power station output prediction network according to a second embodiment of the present invention. As shown in fig. 4, photovoltaic power plant generation data x= (x) for one day may be randomly selected from the dataset of historical output data ₁ ,x ₂ ,,x _q ) And meteorological data w= (w) ₁ ,w ₂ ,…,w _n ) Training the discriminator once, and training the generator and the encoder once again; randomly selecting data from the data set, training the discriminator once, and training the generator and the encoder once again; the loop is thus not trained until Nash equilibrium is achieved. Wherein, the liquid crystal display device comprises a liquid crystal display device,

the degree of training of the network is countered for manually set expectations.

Example III

Fig. 5 is a flowchart of a photovoltaic power plant output predicting method according to a third embodiment of the present invention, where the method may be performed by a photovoltaic power plant output predicting device, and the photovoltaic power plant output predicting device may be implemented in hardware and/or software, and the photovoltaic power plant output predicting device may be configured in an electronic device, such as a computer. As shown in fig. 5, the method includes:

step 510, generating a photovoltaic power station output prediction network by adopting the training method of the photovoltaic power station output prediction network provided by any embodiment of the invention.

And step 520, inputting the current day output data of the photovoltaic power station and the next day weather forecast data into a photovoltaic power station output prediction network, and predicting to obtain the next day output data of the photovoltaic power station.

The output data is generated according to the weather forecast data of the next day, so that the output scene of the photovoltaic power station generated by the generator is closer to the actual output power of the sub-solar photovoltaic power station.

According to the technical scheme provided by the embodiment of the invention, the photovoltaic power station output prediction network is generated by adopting the training method of the photovoltaic power station output prediction network provided by any embodiment of the invention; the current day output data and the next day weather forecast data of the photovoltaic power station are input into a photovoltaic power station output prediction network, the next day output data of the photovoltaic power station is obtained through prediction, the problem of output prediction of the photovoltaic power station is solved, and high-precision prediction of the output of the photovoltaic power station can be achieved.

Example IV

Fig. 6 is a schematic structural diagram of a training device for a photovoltaic power station output prediction network according to a fourth embodiment of the present invention. As shown in fig. 6, the apparatus includes: the data acquisition module 610 generates an output data determination module 620, a discrimination result determination module 630 and a photovoltaic power station output prediction network determination module 640. Wherein:

The data acquisition module 610 is configured to acquire historical output data of the photovoltaic power station and weather data corresponding to the historical output data;

the generated output data determining module 620 is configured to determine generated input data of a generator in the countermeasure generation network according to the meteorological data, and input the generated input data into the generator to obtain generated output data corresponding to the historical output data;

the discrimination result determining module 630 is configured to determine discrimination input data of a discriminator in the countermeasure generation network according to the historical output data, the corresponding generated output data, and the weather data, and input the discrimination input data into the discriminator to obtain a discrimination result;

and the photovoltaic power station output prediction network determining module 640 is used for updating the discriminator by adopting a discriminator loss function according to the discrimination result, updating the generator by adopting a generator loss function, and performing iterative training until the countermeasure generation network is stable, so as to obtain the photovoltaic power station output prediction network.

Optionally, the generating output data determining module 620 includes:

an initial weather feature determination unit for mapping weather data into initial weather features by the encoder;

the node quantity determining unit is used for determining the node quantity of the target meteorological features according to the association relation between the initial meteorological features;

The target meteorological feature determining unit is used for mapping meteorological data into target meteorological features corresponding to the number of nodes through the encoder;

and the generated input data determining unit is used for splicing the target meteorological features and the preset noise features to obtain generated input data of the generator in the countermeasure generation network.

Optionally, the photovoltaic power plant output prediction network determining module 640 includes:

and the photovoltaic power station output prediction network determining unit is used for updating the discriminator by adopting a discriminator loss function according to the discrimination result, updating the generator and the encoder by adopting a generator loss function, and performing iterative training until the countermeasure generation network is stable, so as to obtain the photovoltaic power station output prediction network.

Optionally, the discrimination result determining module 630 includes:

the first input data determining unit is used for splicing the target meteorological features corresponding to the historical output data and the meteorological data to serve as first input data;

the second input data discrimination determining unit is used for splicing the generated output data and the target meteorological features to be used as second input data discrimination;

and the judging input data determining unit is used for taking judging first input data and judging second input data as judging input data.

Optionally, the device further includes:

the judging device loss function determining module is used for determining the judging device loss function according to the distribution distance between the historical output data and the corresponding generated output data and Li Puxi Roots condition before the judging device loss function is adopted to update the judging device and the generator loss function is adopted to update the generator;

and the generator loss function determining module is used for determining a generator loss function according to the output result of the discriminator corresponding to the generated output data.

Optionally, the encoder is a fully connected neural network; wherein:

The training device of the photovoltaic power station output prediction network provided by the embodiment of the invention can execute the training method of the photovoltaic power station output prediction network provided by any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

Example five

Fig. 7 is a schematic structural diagram of a photovoltaic power station output predicting device according to a fifth embodiment of the present invention. As shown in fig. 7, the apparatus includes:

the photovoltaic power station output prediction network generating module 710 is configured to generate a photovoltaic power station output prediction network by adopting the training method of the photovoltaic power station output prediction network provided by any embodiment of the present invention;

The next-day output data prediction module 720 is configured to input the current-day output data of the photovoltaic power station and the next-day weather forecast data into the photovoltaic power station output prediction network, and predict the next-day output data of the photovoltaic power station.

The photovoltaic power station output prediction device provided by the embodiment of the invention can execute the photovoltaic power station output prediction method provided by any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

Example six

Fig. 8 shows a schematic diagram of the structure of an electronic device 10 that may be used to implement an embodiment of the invention. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. Electronic equipment may also represent various forms of mobile devices, such as personal digital processing, cellular telephones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be exemplary only, and are not meant to limit implementations of the inventions described and/or claimed herein.

As shown in fig. 8, the electronic device 10 includes at least one processor 11, and a memory, such as a Read Only Memory (ROM) 12, a Random Access Memory (RAM) 13, etc., communicatively connected to the at least one processor 11, in which the memory stores a computer program executable by the at least one processor, and the processor 11 may perform various appropriate actions and processes according to the computer program stored in the Read Only Memory (ROM) 12 or the computer program loaded from the storage unit 18 into the Random Access Memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 may also be stored. The processor 11, the ROM 12 and the RAM 13 are connected to each other via a bus 14. An input/output (I/O) interface 15 is also connected to bus 14.

Various components in the electronic device 10 are connected to the I/O interface 15, including: an input unit 16 such as a keyboard, a mouse, etc.; an output unit 17 such as various types of displays, speakers, and the like; a storage unit 18 such as a magnetic disk, an optical disk, or the like; and a communication unit 19 such as a network card, modem, wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information/data with other devices via a computer network, such as the internet, and/or various telecommunication networks.

The processor 11 may be a variety of general and/or special purpose processing components having processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various specialized Artificial Intelligence (AI) computing chips, various processors running machine learning model algorithms, digital Signal Processors (DSPs), and any suitable processor, controller, microcontroller, etc. The processor 11 performs the various methods and processes described above, such as the training method of the photovoltaic power plant output prediction network; or a photovoltaic power station output prediction method.

In some embodiments, a method of training a photovoltaic power plant output prediction network; alternatively, the photovoltaic power plant output prediction method may be implemented as a computer program, which is tangibly embodied on a computer-readable storage medium, such as the storage unit 18. In some embodiments, part or all of the computer program may be loaded and/or installed onto the electronic device 10 via the ROM 12 and/or the communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, the above-described training method of the photovoltaic power plant output prediction network may be performed; alternatively, one or more steps of a photovoltaic power plant output prediction method. Alternatively, in other embodiments, the processor 11 may be configured to perform the training method of the photovoltaic power plant output prediction network in any other suitable manner (e.g., by means of firmware); or a photovoltaic power station output prediction method.

Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuit systems, field Programmable Gate Arrays (FPGAs), application Specific Integrated Circuits (ASICs), application Specific Standard Products (ASSPs), systems On Chip (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs, the one or more computer programs may be executed and/or interpreted on a programmable system including at least one programmable processor, which may be a special purpose or general-purpose programmable processor, that may receive data and instructions from, and transmit data and instructions to, a storage system, at least one input device, and at least one output device.

A computer program for carrying out methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the computer programs, when executed by the processor, cause the functions/acts specified in the flowchart and/or block diagram block or blocks to be implemented. The computer program may execute entirely on the machine, partly on the machine, as a stand-alone software package, partly on the machine and partly on a remote machine or entirely on the remote machine or server.

In the context of the present invention, a computer-readable storage medium may be a tangible medium that can contain, or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. The computer readable storage medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. Alternatively, the computer readable storage medium may be a machine readable signal medium. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

To provide for interaction with a user, the systems and techniques described here can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to a user; and a keyboard and a pointing device (e.g., a mouse or a trackball) through which a user can provide input to the electronic device. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form, including acoustic input, speech input, or tactile input.

The systems and techniques described here can be implemented in a computing system that includes a background component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such background, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), wide Area Networks (WANs), blockchain networks, and the internet.

The computing system may include clients and servers. The client and server are typically remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, also called a cloud computing server or a cloud host, and is a host product in a cloud computing service system, so that the defects of high management difficulty and weak service expansibility in the traditional physical hosts and VPS service are overcome.

It should be appreciated that various forms of the flows shown above may be used to reorder, add, or delete steps. For example, the steps described in the present invention may be performed in parallel, sequentially, or in a different order, so long as the desired results of the technical solution of the present invention are achieved, and the present invention is not limited herein.

The above embodiments do not limit the scope of the present invention. It will be apparent to those skilled in the art that various modifications, combinations, sub-combinations and alternatives are possible, depending on design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of the present invention.

Claims

1. The training method of the photovoltaic power station output prediction network is characterized by comprising the following steps of:

2. The method of claim 1, wherein determining the generated input data for the generator in the countermeasure generation network from the meteorological data comprises:

3. The method of claim 2, wherein updating the discriminator with a discriminator loss function and updating the generator with a generator loss function based on the discrimination result, performing iterative training until the countermeasure generation network is stable, and obtaining a photovoltaic power plant output prediction network, comprises:

4. The method of claim 2, wherein determining discrimination input data for a discriminator in the countermeasure generation network based on the historical output data, the corresponding generated output data, and the meteorological data, comprises:

5. The method of claim 1, further comprising, prior to updating the arbiter with the arbiter loss function and updating the generator with the generator loss function based on the discrimination result:

6. The method of claim 2, wherein the encoder is a fully connected neural network; wherein:

7. The photovoltaic power station output prediction method is characterized by comprising the following steps of:

generating a photovoltaic power plant output prediction network by adopting the training method of the photovoltaic power plant output prediction network according to any one of claims 1 to 6;

8. A photovoltaic power plant output prediction network training device, comprising:

9. An electronic device, the electronic device comprising:

at least one processor; and

the memory stores a computer program executable by the at least one processor to enable the at least one processor to perform the training method of the photovoltaic power plant output prediction network of any of claims 1-6; alternatively, the photovoltaic power plant output predicting method of claim 7.

10. A computer readable storage medium storing computer instructions for causing a processor to perform the training method of the photovoltaic power plant output prediction network of any of claims 1-6; alternatively, the photovoltaic power plant output predicting method of claim 7.