CN116760031B

CN116760031B - High-time-resolution photovoltaic power prediction method and device based on meteorological data

Info

Publication number: CN116760031B
Application number: CN202311035011.9A
Authority: CN
Inventors: 施垚; 张清皓; 艾倩雯; 卢樟林; 林超; 裘薇; 钟方军; 张锴; 陆锡航; 李佳萍; 王志毅
Original assignee: Beijing Hongxiang Technology Co ltd
Current assignee: Beijing Hongxiang Technology Co ltd
Priority date: 2023-08-17
Filing date: 2023-08-17
Publication date: 2023-10-27
Anticipated expiration: 2043-08-17
Also published as: CN116760031A

Abstract

The application provides a high-time-resolution photovoltaic power prediction method and device based on meteorological data, and relates to the field of power industry; the method comprises the steps of obtaining a high-resolution multi-source data set of a target area for a preset duration; training the deep learning model by taking the analysis meteorological data as a training sample and the analysis radiation data as a sample label to obtain a pre-training model; taking the fusion meteorological data as a training sample, taking the radiation data as a sample label, and training the pre-training model to obtain a radiation prediction model; the method can use fusion weather forecast data of the preset future time length to forecast the radiation forecast value of the preset future time length through the radiation forecast model. The method solves the problem of insufficient training samples in the process of constructing the radiation prediction model, and also solves the problem of missing radiation monitoring data, thereby further improving the accuracy of the predicted radiation prediction value.

Description

High-time-resolution photovoltaic power prediction method and device based on meteorological data

Technical Field

The application relates to the field of power industry, in particular to a high-time-resolution photovoltaic power prediction method and device based on meteorological data.

Background

Photovoltaic power generation is an important form of renewable energy, and is widely applied and popularized in the global field. Photovoltaic power generation has the characteristics of environmental protection, sustainability and low carbon by converting solar energy into electric energy, and is therefore considered as one of important solutions to cope with climate change and energy crisis.

In recent years, with the rapid development of artificial intelligence and machine learning technologies, new methods and tools have also emerged in the photovoltaic power prediction field. The machine learning-based method utilizes a large amount of historical data and real-time monitoring data to predict the output of photovoltaic power through a training model. The methods can automatically learn complex relationships between data and adjust and optimize the complex relationships in real time according to environmental changes. Photovoltaic power prediction is an important means of improving photovoltaic power generation efficiency. Because the time sequence of the data detected by the photovoltaic station is short, the training sample is insufficient, so that the training model is influenced, and the photovoltaic power of the long-time sequence cannot be predicted further.

Disclosure of Invention

The embodiment of the application aims to provide a high-time-resolution photovoltaic power prediction method and device based on meteorological data, which are used for solving the problems in the prior art and predicting long-time-sequence photovoltaic power.

In a first aspect, a high time resolution photovoltaic power prediction method based on meteorological data is provided, the method may include:

acquiring a high-resolution multi-source data set of a preset duration of a target area; the multi-source dataset comprises analysis data, satellite fusion data and multi-mode weather forecast data; the analysis data includes re-analysis weather data and analysis radiation data;

training the deep learning model by taking the analysis meteorological data as a training sample and the analysis radiation data as a sample label to obtain a pre-training model;

taking the fusion meteorological data as a training sample, taking the radiation data as a sample label, and training the pre-training model to obtain a radiation prediction model; the radiation data are obtained by inverting and correcting the satellite fusion data; the fusion weather data is obtained by fusing and correcting the multi-mode weather forecast data.

In one possible implementation, after obtaining the radiation prediction model, the method further includes:

predicting ground weather data detected by a ground weather station at a position, and determining weather forecast data in a preset future time length;

Adopting the radiation prediction model to process the weather forecast data to obtain a radiation prediction value in the preset future time length;

and determining a photovoltaic power predicted value corresponding to the radiation predicted value based on the conversion relation between the target radiation data and the photovoltaic power.

In one possible implementation, obtaining high resolution, re-analysis meteorological data and re-analysis radiation data for a preset length of time for a target area includes:

acquiring initial re-analysis meteorological data and initial re-analysis radiation data of the preset duration of the target area; the initial re-analysis weather data is low-resolution weather data, and the initial re-analysis radiation data is low-resolution radiation data;

and processing the initial re-analysis meteorological data and the initial re-analysis radiation data by adopting a preset super-resolution reconstruction model, determining the re-analysis meteorological data and the re-analysis radiation data, wherein the preset super-resolution reconstruction model is used for converting low-resolution data into high-resolution data, and the preset super-resolution reconstruction model is a LapSRN model.

In one possible implementation, obtaining high-resolution satellite fusion data for a preset duration of a target area includes:

Acquiring multi-source satellite data of the preset duration in the target area, wherein the multi-source satellite data comprise satellite data with multiple spatial resolutions and satellite data with multiple time resolutions;

processing the satellite data with the multiple spatial resolutions and the satellite data with the multiple time resolutions by adopting a preset space-time fusion model to obtain the satellite fusion data; the preset space-time fusion model is used for fusing data with multiple spatial resolutions and multiple time resolutions so as to obtain fused data with high space-time resolution.

In one possible implementation, the radiation data is obtained by inverting and correcting the satellite fusion data, including:

processing the satellite fusion data by adopting a preset satellite inversion radiation model to obtain initial radiation data; the preset satellite inversion radiation model is used for inverting the satellite fusion data so as to determine the radiation power of the ground;

adopting a preset artificial intelligent correction model, correcting the initial radiation data based on the corresponding ground radiation data and ground photovoltaic power data of the target area, and obtaining the radiation data; the preset artificial intelligence correction model is a BP neural network model.

In one possible implementation, the fused weather data is obtained by fusing and correcting the multi-mode weather forecast data, and includes:

processing the multimode weather forecast data by adopting a preset multimode fusion model to obtain a fusion result; the preset multi-mode fusion model is used for fusing the meteorological data of multiple modes to obtain fused meteorological data;

adopting a preset localization correction model, correcting the fusion result based on the corresponding meteorological data set detected by the power station cluster of the target area, and determining the fusion meteorological data, wherein the preset localization correction model is a Seq2Seq model; the power station cluster comprises a ground radiation station, a ground weather station and a photovoltaic station; the meteorological data set includes ground radiation data detected by the ground radiation station, ground meteorological data detected by the ground meteorological station, and ground photovoltaic power data detected by the photovoltaic station.

In one possible implementation, based on the conversion relationship of the target radiation data to photovoltaic power, comprising:

adopting the radiation prediction model to process test meteorological data to obtain test radiation data, wherein the test meteorological data are historical meteorological data;

And determining a target conversion relation between the test radiation data and the historical photovoltaic power based on the historical photovoltaic power corresponding to the test radiation data and the test meteorological data by adopting a DNN model, and determining the target conversion relation as the conversion relation between the target radiation data and the photovoltaic power.

In a second aspect, a high time resolution photovoltaic power prediction apparatus based on meteorological data is provided, the apparatus may include:

the acquisition unit is used for acquiring a high-resolution multi-source data set of a preset duration of a target area; the multi-source dataset comprises analysis data, satellite fusion data and multi-mode weather forecast data; the analysis data includes re-analysis weather data and analysis radiation data;

the determining unit is used for training the deep learning model by taking the analysis meteorological data as a training sample and the analysis radiation data as a sample label to obtain a pre-training model;

the determining unit is further used for training the pre-training model by taking the fused meteorological data as a training sample and the radiation data as a sample label to obtain a radiation prediction model; the radiation data are obtained by inverting and correcting the satellite fusion data; the fusion weather data is obtained by fusing and correcting the multi-mode weather forecast data.

In a third aspect, an electronic device is provided, the electronic device comprising a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory are in communication with each other via the communication bus;

a memory for storing a computer program;

a processor for implementing the method steps of any one of the above first aspects when executing a program stored on a memory.

In a fourth aspect, a computer-readable storage medium is provided, in which a computer program is stored which, when being executed by a processor, carries out the method steps of any of the first aspects.

The application provides a high-time-resolution photovoltaic power prediction method based on meteorological data, which comprises the steps of obtaining a high-resolution multi-source data set of a target area for a preset duration; the multi-source dataset comprises analysis data, satellite fusion data and multi-mode weather forecast data; the analysis data includes re-analysis weather data and analysis radiation data; training the deep learning model by taking the analysis meteorological data as a training sample and the analysis radiation data as a sample label to obtain a pre-training model; taking the fusion meteorological data as a training sample, taking the radiation data as a sample label, and training the pre-training model to obtain a radiation prediction model; the radiation data are obtained by inverting and correcting the satellite fusion data; the fusion weather data is obtained by fusing and correcting the multi-mode weather forecast data. The method can use fusion weather forecast data of the preset future time length, and forecast the radiation forecast value of the preset future time length through the radiation forecast model, wherein the preset future time length can be a month or other long-time sequence value. The method solves the problem of insufficient training samples in the process of constructing the radiation prediction model, and also solves the problem of missing radiation monitoring data, thereby further improving the accuracy of the predicted radiation prediction value.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed in the embodiments of the present application will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present application and should not be considered as limiting the scope, and other related drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a schematic diagram showing a comparison of high-resolution reconstruction samples of meteorological data under different algorithms and different models according to an embodiment of the present application;

FIG. 2 is a comparative schematic of a standard convolution and a deformable convolution provided by an embodiment of the present disclosure;

FIG. 3 is a schematic diagram of a process for inverting solar radiation by satellites according to an embodiment of the present application;

fig. 4 is a schematic diagram of an empirical relationship between clear sky indexes and cloud indexes according to an embodiment of the present application;

fig. 5 is a schematic structural diagram of a BP neural network according to an embodiment of the present application;

fig. 6 is a schematic structural diagram of a DNN network according to an embodiment of the present application;

FIG. 7 is a system architecture diagram of a high time resolution photovoltaic power prediction method applied to meteorological data according to an embodiment of the present application;

FIG. 8 is a schematic flow chart of a high-time-resolution photovoltaic power prediction method based on meteorological data according to an embodiment of the present application;

FIG. 9 is a detailed flow chart of a high-time-resolution photovoltaic power prediction method based on meteorological data according to an embodiment of the present application;

fig. 10 is a schematic structural diagram of a high-time-resolution photovoltaic power prediction device based on meteorological data according to an embodiment of the present application;

fig. 11 is a schematic structural diagram of an electronic device according to an embodiment of the present application.

Detailed Description

The following description of the embodiments of the present application will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments, but not all embodiments of the present application. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.

A. For convenience of understanding, the terms involved in the embodiments of the present application are explained below:

analyzing data, namely, a meteorological data set which is formed by re-fusing and integrating various types of observation data from different sources with numerical weather forecast by utilizing a perfect data assimilation technology; which may include re-analyzing the meteorological data and re-analyzing the radiation data.

The multi-source satellite data are acquired by a plurality of satellite-borne sensors acquired in a multi-time phase, multi-angle, multi-spectrum, active and passive modes and the like.

The multi-mode weather forecast data is data for establishing a multi-model system aiming at different regions, different time effects and different characteristics, and can comprise global mode weather forecast data, global and regional mode weather forecast data, regional mode weather forecast data and the like.

B. The following is a detailed description of the various models to which the present application relates:

(1) The method comprises the steps of presetting a super-resolution reconstruction model, namely a LapSRN model, wherein the model is formed by respectively carrying out different levels of convolution and up-sampling on an initial image, up-sampling a convolved feature image to obtain sub-band residual information, adding the sub-band residual information with an up-sampled medium-resolution image to obtain the medium-resolution image, and then gradually adopting the operation to obtain a higher-resolution image. It is widely used in image processing, such as face recognition, radar extrapolation, etc., due to its internal unique filter structure, weight sharing, and translation invariant properties. When the spatial resolution of the data is improved, the model selects the original data and the topographic and geomorphic information as input in consideration of regional factors and topographic factors, and the convolutional neural network is utilized to extract the high-dimensional characteristics of the data, so that the high-resolution reconstruction of the data is performed.

As shown in fig. 1, the temperature image and the rainfall image with low spatial resolution of LR (little resolution) are respectively processed by using a CUBIC (bilinear interpolation algorithm), a HR (Hierarchical Residual Learning) layered residual learning model and a LapSRN model, so as to obtain high-resolution images output by each algorithm and each model.

The first stage of the LapSRN model is input into low-spatial resolution data and high-spatial resolution topographic data to obtain high-spatial resolution data, and the second stage is input with the obtained high-spatial resolution data and higher-spatial resolution topographic data to obtain higher-spatial resolution data.

(2) The deep learning model can be a ConvGRU model, the ConvGRU is an encoding-decoding model (Encoder-Decoder), the encoding part consists of two modules, the convolution module extracts the spatial characteristics of an input image for training the GRU, the GRU is used for extracting the time sequence characteristics of an input time sequence, the decoding part is the inverse process of encoding, the image characteristic size and the GRU learning image sequence characteristics are increased through transposed convolution up-sampling, and a predicted satellite cloud image and a predicted multi-channel observation value are output. ConvGRU inherits GRU characteristics, has advantages such as unit parameter is few, training speed is fast, can be applicable to space-time sequence data prediction well.

To improve the feature extraction capability of ConvGRU, a deformable convolution was decided to be employed.

In the space-time prediction problem, the change in the time dimension is mainly the change of the pixel point position, the position of the target object in the current frame and the position of the target object in the next frame are quite possibly not corresponding, and the regular lattice point sampling in the standard convolution is difficult to fully extract the characteristics of the geometrically deformed target. Thus, adaptive learning of scale or perceived size is required for accurate positioning.

The method uses a parallel convolution layer to learn offset, so that the convolution kernel generates offset at sampling points on an input feature map and concentrates in a target area, namely, an offset is added to the position of each sampling point in the convolution kernel, random sampling can be performed near the current position, and the sampling can not be limited to the previous regular lattice points, thereby improving the feature extraction capability of a space-time prediction model and obtaining a hidden layer with more abundant features. The input image U, whose size is [ H, W, C ], is subjected to a normal convolution operation with a convolution kernel size of (kernel_size×kernel_size), the convolution is filled with the same, the feature map V with the size of [ H, W,2C ] is output, and the output result is the offset of each pixel of the input picture, which is 2C because it is the offset in both X-axis and Y-axis directions. The values of V and U are added to generate the offset coordinate value position, and the pixel offset value obtained by the convolution layer is often a fraction, so that it is necessary to obtain the actual pixel offset position by linear interpolation. As shown in fig. 2, where (a) is a standard convolution and (b) is a deformable convolution, and finally an output result is generated for the shifted feature map operation using a standard convolution operation.

(3) The preset space-time fusion model can be an SPSTFM model (remote sensing space-time fusion model), and the model is a method for fusing multiple satellite data with different space-time resolutions so as to obtain the satellite data with high space-time resolution. The SPSTFM model is an algorithm based on dictionary learning. The traditional dictionary learning fusion model generally comprises two phases, namely a dictionary pair training phase and a reconstruction phase, which are independently carried out in each wave band, wherein the earliest dictionary learning fusion model is shown as the following mathematical expression of the two phases in SPSTFM:

a. dictionary pair training phase

During the training phase, the difference image pair L is utilized ₃₁ And M ₃₁ Training a high resolution dictionary D _l And a low resolution dictionary D _m, The equation is as follows:

wherein Y and X represent a column combination of image patches, respectively from L ₃₁ And M ₃₁ A represents a combination of columns of coefficients, each column corresponding to each column lambda in Y and X, respectively, is a regularized parameter by K-The SVD algorithm can find the D in the equation _l And D _m 。

b. Reconstruction stage

Image M using dictionary Dm ₂₁ Each patch block in the (b) is encoded, and the corresponding sparse representation coefficient alpha is obtained.

The method can be obtained by solving the following optimization problems:

wherein m is ₂₁ Represents M ₂₁ Then the patch of the corresponding high resolution image may be obtained by the following equation:

then, all L ₂₁ The patch blocks of the image (L) are fused to obtain a high-resolution image (L) ₂₁ The method comprises the steps of carrying out a first treatment on the surface of the Image L ₃₂ Can also be obtained in the same way, finally by the equationThe target image L can be predicted ₂ 。

(4) The satellite inversion radiation model is preset, and interactions between solar radiation on the outer layer of the earth and the earth atmosphere, surfaces and objects should be considered in the modeling process of satellite inversion solar radiation.

The cloud acts as the most intense surface solar radiation regulator (much stronger than other atmospheric components) and plays a key role in surface solar radiation estimation. In view of the decisive influence of the cloud on the surface solar radiation, the estimation of the surface solar radiation satellites is to some extent done around how accurately the radiation attenuation of the cloud in the atmosphere (Yun Sanshe and cloud absorption) is considered.

(5) The preset artificial intelligence correction model can be a BP neural network model, and the BP neural network is a multi-layer feedforward neural network and is mainly characterized in that: the signal is forward propagating and the error is backward propagating. A typical BP neural network comprises an input layer, an hidden layer and an output layer, and generally, there is only one input layer and one output layer, and there is at least one hidden layer. Each layer of the network has a certain number of parallel neurons, and the connection of the neurons follows the rules that different layers are connected with each other and the same layer is not connected with each other. The multi-layer network of different numbers of neurons forms a nonlinear system that can solve many problems with respect to complexity self-learning of the signal. Taking three layers of BP neural network as an example, the network structure is shown in figure 5.

When the BP neural network is trained, input vectors are respectively and positively input into respective neurons, after the weighted and summed data are converted through an excitation function, the obtained output result is continuously used as the input of an implicit layer to repeat the above process until the error between the final output layer result and the expected result is calculated, and then the reverse transmission of the error is started. The aim of updating the weight of each layer is achieved through the transmission of errors in opposite directions. The network starts the next round of forward signal calculation and backward weight correction based on the updated weight, and by repeating the calculation in the forward and backward directions for a plurality of times, the error between the network output and the expected output becomes smaller and smaller until the error falls into the allowable minimum range, and the learning training process is ended.

(6) The preset multimode fusion model can be a CNN model, and can be used for considering the distribution of surrounding weather when adjusting certain weather data by utilizing the characteristics of a convolutional neural network model. The CNN model contains a plurality of feature extractors consisting of a convolutional layer and a pooling layer. In the convolutional layer of the CNN model, one neuron is connected to only a part of adjacent layer neurons. In a convolutional layer of CNN, a number of feature maps (featuremaps) are usually included, each feature map is composed of some neurons arranged in a rectangle, the neurons of the same feature map share weights, the weights shared here are convolution kernels, the convolution kernels are initialized in a random decimal matrix, and the convolution kernels learn to obtain reasonable weights in the training process of the network. A direct benefit of sharing weights (convolution kernels) is to reduce the connections between layers of the network while reducing the risk of overfitting. Subsampling is also known as pooling (pooling), and typically takes two forms, mean subsampling (mean pooling) and maximum subsampling (max pooling). Sub-sampling can be seen as a special convolution process. The convolution and sub-sampling greatly simplify the complexity of the model and reduce the parameters of the model.

(7) The localization correction model is preset and can be a Seq2Seq model, and the core of the model is ConvLSTM, which is beneficial to processing data in a space-time direction. The Seq2Seq model consists of two parts, namely an Encoder and a deconer, wherein the Encoder is used for analyzing the history information, and the deconer is used for compiling the history information to generate a correction result.

ConvLSTM is mainly used for replacing the product operation in LSTM with convolution operation, even if the ConvLSTM has convolution structure in the process of input to state and state-to-state conversion, so that the ConvLSTM is more suitable for processing space-time data. According to research, convLSTM networks are found to be better able to capture spatio-temporal correlations and consistently better than FC-LSTM.

In the training process, the model trains the Encoder firstly, then transmits the training state of the Encoder model to the Decode, and continues to train the Decode, wherein the cell structures of the Decode and the Encoder can be different, and the weights of the Decode and the Encoder can be not shared. The model has the advantages that not only is the history information transmitted, but also the space information is compared, so that grid point forecast correction can be accurately carried out.

(8) Kernel PCA model, kernel PCA, is a modified version of PCA that converts nonlinear separable data onto a new low-dimensional subspace suitable for alignment for linear classification, and Kernel PCA can convert the data into one high-dimensional space by nonlinear mapping, map it into another low-dimensional space using PCA in the high-dimensional space, and partition the samples by linear classifier. Kernel function: the function of similarity between vectors is measured by two vector dot products. Common functions are: polynomial kernels, hyperbolic tangent kernels, radial Basis Functions (RBFs) (gaussian kernel functions), and the like.

(9) In the DNN model, the DNN network structure is shown in fig. 6, where the leftmost layer in the network is an input layer (input layer), the middle two layers are hidden layers (hidden layers), the rightmost layer is an output layer (output layer), and the training process of the neural network is divided into a forward process and a reverse process, where the reverse process is used for updating weights and bias constants on neurons in the training process. The weight value and the bias constant multiplied by the corresponding neuron in the forward process are input into the activation function for calculation. The data flowing in from the input layer is firstly subjected to first-layer neuron operation to obtain output which is used as the input of the next-layer neuron, so as to obtain the output of the second layer until reaching the output layer operation to obtain a result.

C. Photovoltaic power generation is an important form of renewable energy, and is widely applied and popularized in the global field. Photovoltaic power generation has the characteristics of environmental protection, sustainability and low carbon by converting solar energy into electric energy, and is therefore considered as one of important solutions to cope with climate change and energy crisis. However, the power output of photovoltaic power generation is affected by a variety of factors, including solar radiation intensity, weather conditions, air temperature, cloud cover, and the like. These factors vary such that there is a large fluctuation and uncertainty in the power output of the photovoltaic power generation, thereby presenting challenges to energy production and supply. To solve this problem, photovoltaic power prediction is an important component of photovoltaic power generation management and operation. Photovoltaic power prediction aims to accurately predict the output of photovoltaic power over a period of time in the future based on historical data and current environmental conditions. This is critical to optimize energy scheduling, achieve reliable grid operation, and increase the economics of photovoltaic power generation. Photovoltaic power prediction is a complex problem that requires consideration of interactions and non-linear relationships of multiple factors.

First, solar radiation is one of the key factors affecting photovoltaic power output. The intensity and distribution of solar radiation is affected by geographic location, seasonal variations, and weather conditions. Second, weather factors such as cloud cover, wind speed, and temperature can also have a significant impact on photovoltaic performance. In addition, characteristics of the photovoltaic power generation itself, such as component type, layout, loss, and the like, also have an influence on power output.

In recent years, with the rapid development of artificial intelligence and machine learning technologies, new methods and tools have also emerged in the photovoltaic power prediction field. The machine learning-based method utilizes a large amount of historical data and real-time monitoring data to predict the power output of the photovoltaic by training a model. The methods can automatically learn complex relationships between data and adjust and optimize the complex relationships in real time according to environmental changes. Photovoltaic power prediction is an important means of improving photovoltaic power generation efficiency.

At present, a plurality of photovoltaic medium-long term prediction algorithms exist; algorithms suitable for photovoltaic power high time resolution month prediction are roughly of two types:

1. a method based on time series analysis: such algorithms are based primarily on historical photovoltaic power analysis and are predicted using time series models, such as autoregressive moving average (ARMA) and autoregressive conditional heteroscedastic (ARIMA). By modeling the trend and periodicity of historical photovoltaic power, future photovoltaic power output can be predicted.

Regression analysis-based method: the regression model is used for prediction by establishing a regression relationship between photovoltaic power output and weather factors (such as solar radiation, temperature, etc.). Common regression models include linear regression, polynomial regression, support vector regression, and the like.

2. Prediction algorithm based on machine learning:

the method based on the artificial neural network comprises the following steps: a multi-layer feed forward neural network (e.g., a multi-layer perceptron) or a recurrent neural network (e.g., a long and short term memory network) is used to establish a mapping between photovoltaic power output and input characteristics. The neural network can automatically learn the nonlinear relation of the data and has stronger prediction capability.

The method based on the support vector machine comprises the following steps: support vector machines are a common method of machine learning, classifying or regressing by finding an optimal hyperplane. In photovoltaic power prediction, the support vector machine can output future photovoltaic power by learning the relation between historical ground photovoltaic power data and weather factors.

Method based on ensemble learning: the ensemble learning combines a plurality of individual prediction models together to obtain a final prediction result by voting or weighted average. Common ensemble learning methods include random forests and gradient-lifted trees. The method can effectively utilize the advantages of a plurality of models and improve the accuracy and stability of prediction.

The above method can be further classified according to the difference of input characteristics. For example, some methods consider only historical photovoltaic power, called univariate predictive models; some methods combine historical photovoltaic power with weather factors, known as multivariate predictive models. In addition, some methods consider other factors, such as holidays, temporal characteristics, etc., to improve the accuracy of the predictions.

When the output model of the photovoltaic power is trained by using any method, the training sample is insufficient due to the fact that the time sequence of data detected by the photovoltaic station is short, so that the training model is influenced, and the prediction result of the photovoltaic power of a long time sequence is further influenced.

The high-time resolution photovoltaic power prediction method based on meteorological data provided by the embodiment of the application can be applied to a system architecture shown in fig. 7, and as shown in fig. 7, the system can comprise: server and terminal. The server may be a physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN), basic cloud computing services such as big data and artificial intelligent platforms. The Terminal may be a Mobile phone, a smart phone, a notebook computer, a digital broadcast receiver, a Personal Digital Assistant (PDA), a tablet personal computer (PAD), or other User Equipment (UE), a handheld device, a car-mounted device, a wearable device, a computing device, or other processing device connected to a wireless modem, a Mobile Station (MS), a Mobile Terminal (Mobile Terminal), or the like. The terminal and the server may be directly or indirectly connected through wired or wireless communication, which is not limited herein.

And the terminal is used for acquiring the multi-source data set and sending the multi-source data set to the server.

The server is used for receiving the multi-source data set and executing the high-time resolution photovoltaic power prediction method based on the meteorological data.

The method makes up the defect of short time sequence of the power station data, can provide sufficient training samples, and utilizes the multi-mode weather forecast data to combine with a preset super-resolution reconstruction model, a preset multi-mode fusion model and a preset localization correction model so as to generate the weather data localization, refinement and gridding forecast of the photovoltaic station in a small scale range. This will increase the spatial-temporal resolution and accuracy of the data and provide the necessary data support for subsequent photovoltaic power predictions.

The preferred embodiments of the present application will be described below with reference to the accompanying drawings of the specification, it being understood that the preferred embodiments described herein are for illustration and explanation only, and not for limitation of the present application, and embodiments of the present application and features of the embodiments may be combined with each other without conflict.

Fig. 8 is a flow chart of a high-time-resolution photovoltaic power prediction method based on meteorological data according to an embodiment of the present application. As shown in fig. 8, the method may include:

Step 810, acquiring a high-resolution multi-source data set of a preset duration of a target area.

Wherein the multi-source dataset includes analysis data, satellite fusion data, and multi-mode weather forecast data. The analysis data includes re-analysis weather data and analysis radiation data.

Before executing step S810, the initial analysis weather data and the initial analysis radiation data need to be processed to obtain high-resolution analysis weather data and analysis radiation data, which may specifically include:

the low-resolution initial analysis meteorological data and the low-resolution initial analysis radiation data are respectively input into a preset super-resolution reconstruction model (LapSRN model), and the high-spatial-resolution analysis meteorological data and the high-spatial-resolution analysis radiation data are output through the preset super-resolution reconstruction model.

In this way, as shown in fig. 1, the result output by the LapSRN model has high quality, and the reconstruction of data with high spatial resolution can be accurately performed.

And step S820, pre-training the deep learning model by taking the analysis meteorological data as a training sample and the analysis radiation data as a sample label to obtain a pre-training model.

Specifically, a training sample set is constructed according to the analysis weather data acquired in the step S810, and the deep learning model is pre-trained by taking analysis radiation data corresponding to the analysis weather data acquired in the step S810 as a sample tag set, so that a pre-training model can be obtained. The deep learning model may be a convglu model.

In this way, the training sample set composed of meteorological data is analyzed, and the problem that samples of a long time sequence (month sequence) are insufficient in the formal training process can be solved.

And step 830, performing formal training on the pre-training model by taking the fused meteorological data as a training sample and the radiation data as a sample label to obtain a radiation prediction model.

Before the pre-training model is formally trained, a training sample set fused with meteorological data is needed to be constructed, and a sample tag set of radiation data corresponding to the meteorological data is fused.

Specifically, a, acquiring multi-mode weather forecast data of a target area history for a preset time length.

And taking the multi-mode weather forecast data as input of a preset super-resolution reconstruction model to obtain the multi-mode weather forecast data with high spatial resolution.

Due to the fact that the generation principle and the characteristics of the weather forecast data of each mode in the multi-mode weather forecast data are different, the weather forecast data have advantages, and each weather forecast data has applicable scenes. If the model is formally trained by using weather forecast data of a single mode, it is difficult to ensure that the model obtained after the formal training can be suitable for various scenes. A multi-model system is established aiming at different regions, different timelines and different characteristics. Proper weather forecast data can be fused, the weather forecast data are mutually supplemented, the advantages and the disadvantages are increased, and the accuracy of the forecast data can be greatly improved.

Therefore, the weather forecast data of each mode is fused, specifically, the weather forecast data of each mode is used as input of a preset multimode fusion model, multimode fusion is performed by using a convolutional neural network, the weight is automatically adjusted according to recent data, and the weight is effectively adjusted according to the quality of the weather forecast data of each mode, so that a fusion result with high accuracy can be obtained.

Because the obtained fusion result may have a certain error with the actual weather forecast data, a preset localization correction model is adopted, and the fusion result is corrected based on the corresponding historical weather data set detected by the power station cluster of the target area, so as to determine the fusion weather data.

The power station cluster comprises a ground radiation station, a ground weather station and a photovoltaic station. The meteorological dataset comprises: ground radiation data detected by the ground radiation station, ground weather data detected by the ground weather station, and ground photovoltaic power data detected by the photovoltaic station.

And constructing a training sample set according to the determined fusion meteorological data.

b. Acquiring multi-source satellite data of a target area history in a preset time length, wherein the multi-source satellite data is low-resolution multi-source satellite data; the low resolution includes a low temporal resolution and/or a low spatial resolution. The multi-source satellite data may include satellite data of different temporal resolutions and satellite data of different spatial resolutions.

It should be noted that, the historical multi-mode weather forecast data with preset duration corresponds to the historical multi-source satellite data with preset duration.

Satellite data with different time resolutions and satellite data with different spatial resolutions are used as input of a preset space-time fusion model (SPSTFM model, remote sensing space-time fusion model), and satellite fusion data with high space-time resolution can be obtained.

The satellite fusion data is used as input of a preset satellite inversion radiation model, the output of the preset satellite inversion radiation model is radiation data corresponding to the satellite fusion data, and the radiation data is determined to be initial radiation data.

The process of processing the satellite fusion data by the preset satellite inversion radiation model is a process of satellite inversion solar radiation, as shown in fig. 3, and mainly comprises the following steps:

firstly, satellite fusion data are processed through a clear sky model, so that clear sky horizontal total radiation is obtained (irradiance reaching the ground when clear sky exists is set). In the ESRA5i clear sky model, the clear sky level total radiationIs divided into two parts: direct component->And scattering component->. Irradiance is given in +.>。

And then, calculating a cloud index through satellite fusion data, and further calculating a clear sky index to quantify the weakening effect of the cloud cover. It should be noted that the cloud index may also be understood as a cloud albedo.

Finally, the total level radiation (initial radiation data) throughout the day is derived by combining the total level radiation of clear sky and the clear sky index.

Specifically, a, direct radiation in clear sky:

direct irradiance (direct component) of horizontal plane under clear sky conditionGiven by the formula:

wherein, the liquid crystal display device comprises a liquid crystal display device,is solar constant, i.e. irradiance outside (extraterrestrial radiation) at average solar-to-earth distance, equal to；/>Is a correction value for taking into account the change in the distance between the earth and the sun and the average value thereof; />Is the solar altitude. Sunrise and sunset time->Is 0 degrees; />The effect of the present invention is that, for a linkman turbidity factor for an atmospheric mass equal to 2, the linkman turbidity factor is a function of aerosol scattering (aerosol optical thickness) and gas (mainly water vapor, i.e. reduced water vapor) absorption, when combined with atmospheric molecular scattering, it summarizes the turbidity of the atmosphere, thus summarizing the importance of the attenuation and diffusion fraction of the direct light beam, the greater the linkman turbidity factor, the greater the attenuation of the clear atmosphere to radiation; />For relative atmospheric mass>To integrate the rayleigh optical thickness.

Indicating the transmittance of direct radiation under cloudless sky. The relative atmospheric mass represents the ratio of the path length of the solar beam through the atmosphere to the path length of the sun through the sea level standard atmosphere at zenith. As the solar altitude decreases, the relative optical path length increases. The relative path length also follows the altitude of the station The increase in z decreases. A correction method is used to obtain the ratio of the average atmospheric pressure p of the site altitude to the average atmospheric pressure p0 of the sea level. The relative optical quality has no unit, and the calculation mode is as follows: />

The station height correction is as follows:

wherein z is the elevation of the station,equal to 8434.5 m.

Solar altitude used in the processThe folding is corrected:

rayleigh optical thicknessThe parameterization of (2) is as follows:

the variation of the direct transmittance with the atmospheric mass is contained inIs a product of (c) and (d).

b. Clear sky scattered radiation:

clear sky horizontal planeScattered irradiance onThe calculation of (2) also depends on the linkman turbidity factor +.>. In practice, the proportion of scattered energy in the atmosphere increases with increasing turbidity, and as the direct irradiance decreases, the scattered irradiance generally increases. However, under very low solar altitude and high turbidity conditions, the scattered irradiance may decrease with increasing turbidity due to higher total radiant energy losses. Therefore, scattering irradiance (scattering component)/(scattering component)>Determined by the following formula:

in this equation, scattered radiation is expressed as a scattered transmission function at zenithAnd the scattering angle function->Is a product of (a) and (b).

For a very clear sky, the scattering transmittance is low, i.e. there is little scattering. As haze increases, diffuse transmittance increases and direct transmittance decreases. In general, the number of the devices used in the system, Ranging from clear sky (+)>0.05 to turbid atmosphere (++2) under =2)>=7) at 0.22. Scattering angle function->Depending on the elevation angle of the sun, and fitting by means of a second order sinusoidal polynomial function: />

Wherein the coefficient is And->Only on the linkman turbidity factor.

c. Adding the direct irradiance (direct component) and the scattered irradiance (scattered component) under cloudless weather conditions on the basis of the above a and b to obtain clear sky level total radiation G _c ：。

Further, to solve the problem of efficiently and accurately estimating solar radiation under all-cloudy conditions, many studies have been conducted in the past to take a large number of atmospheric and surface parameters as inputs to a radiation transmission model to account for the effects of the cloud on the radiation. Although these models have a definite physical process, the spatial resolution of the final result is limited due to the large number of variables involved and the large computational effort. In contrast, the Heliosat-2 method selects the cloud albedo as the comprehensive index of the cloud attenuation effect, attempts to determine the influence of the cloud by utilizing the comprehensive characteristics of the whole atmosphere, does not need the assumption of the vertical structure of the atmosphere, and has higher calculation efficiency, so that the method is widely applied.

The Heliosat-2 adopts a semi-parameterized solar radiation model, satellite data are adopted to identify cloud characteristics, the atmospheric attenuation process of most solar radiation is also considered in the calculation scheme, and some input physical parameters are also adopted. Thus, this calculation scheme can sufficiently simulate the actual situation. The estimation effect is generally superior to that of radiation products such as global energy and water circulation experiments, ERA analysis data and the like. The Heliosat-2 method is very widely used because of its sufficient versatility and flexibility.

The Heliosat-2 model combines a clear sky model with a "cloud index". The cloud index method is based on the following assumptions: the presence of clouds on the pixels results in an increase in the reflectivity of the visible image; the attenuation of the downstream short-wave irradiance of one pixel by the atmosphere is related to the magnitude of the change between the reflectivity that should be observed in cloudless sky and the reflectivity that is currently observed. The magnitude of this change can be characterized by introducing a cloud index and a clear sky index. Cloud index n is defined as follows:

the clear sky index is defined as the ratio of the actual horizontal irradiance to the clear sky horizontal total radiationThe method comprises the following steps:

wherein, the liquid crystal display device comprises a liquid crystal display device,is the apparent reflectivity observed by the satellite. />Apparent reflectance of brightest clouds, +.>Is the reflectivity of the earth surface>And->The actual irradiance and clear sky level total radiation respectively.

Finally, as shown in fig. 4, an empirical relationship between Clear-sky index (Clear-sky index) and Cloud index (Cloud index) is established.

Mapping relation between clear sky index and cloud index:

according to the mapping relation between the clear sky index and the cloud index, the cloud index (Yun Fanzhao rate) is brought into a corresponding formula, and the target clear sky index can be obtained.

The surface solar radiation under the clear sky condition is multiplied by the target clear sky index to obtain the horizontal total radiation (initial radiation data).

Furthermore, since the initial radiation data may have a certain error in the actual radiation data, the target area is required to be used to correspond to the historical ground photovoltaic power data detected by the photovoltaic station and the historical ground radiation data detected by the radiation station, and the historical ground photovoltaic power data and the historical ground radiation data detected by the radiation station are used as input of a preset artificial intelligence correction model (BP neural network model), and the initial radiation data is corrected through the model, so that more accurate radiation data is obtained. Specifically, initial radiation data to be corrected is used as an input feature of the BP neural network model, high-precision geographic information data, land utilization type data, meteorological data and other features are input, then the radiation data detected by the ground radiation station and the radiation data corresponding to the ground photovoltaic power detected by the photovoltaic station are used as output, and after the BP neural network processing, corrected radiation data is automatically generated.

And constructing a sample label set according to the determined radiation data.

The method can solve the problem of radiation monitoring data missing, provides rich historical radiation data (radiation data) for formally training the model, and can improve the model training precision based on the historical radiation data.

The step S830 may specifically include training the pre-training model (convglu model) obtained in the step S820 according to the sample set and the sample label set constructed above to obtain a radiation prediction model, where the radiation prediction model is a model that can be actually applied and predicts radiation data of a preset future time length.

Further, radiation data for predicting a preset future time length is obtained, a kernalPCA model and a DNN model are needed for predicting the photovoltaic power, the test radiation data, the corresponding historical meteorological data and the corresponding historical photovoltaic power are processed, a target conversion relation between the test radiation data and the corresponding historical photovoltaic power is determined, and the target conversion relation is determined as a conversion relation between the target radiation data and the photovoltaic power. Specifically, a radiation prediction model is adopted to process test meteorological data to obtain test radiation data, the test meteorological data is historical meteorological data, and the obtained test radiation data is also historical radiation data; and extracting abstract features from the test radiation data, the corresponding historical meteorological data and the corresponding historical photovoltaic power by using a kernalPCA model to obtain a data set formed by combining the test radiation data, the corresponding historical meteorological data and the corresponding historical photovoltaic power after extracting abstract deep features, taking the data set as the input of a DNN model, and finally training the data set through the DNN model to obtain the target conversion relation of the test radiation data and the corresponding historical photovoltaic power.

And step S840, adopting a radiation prediction model to process weather forecast data to obtain a radiation prediction value.

Further, the method includes the steps that a ground weather data set detected by a ground weather station at a position is predicted, weather forecast data in a preset future time period are determined, and in the process of determining the weather forecast data in the preset future time period, not only the ground weather data detected by the ground weather station but also other types of data are needed for predicting weather preset data in the preset future time period, and the method is not limited herein;

and processing the weather forecast data by adopting a radiation prediction model to obtain a radiation prediction value in a preset future time length.

It should be noted that, because the radiation prediction model uses the fusion meteorological data obtained by fusion processing of the multi-mode meteorological prediction data in the training process, that is to say, the training process integrates the data of multiple modes, which can be single-mode meteorological data or multi-mode meteorological data; therefore, in the actual application process of the radiation prediction model, the model can consider single-mode meteorological data or multi-mode meteorological data, so that the radiation prediction model is suitable for various scenes in the actual application process.

And determining a photovoltaic power predicted value with high time resolution corresponding to the radiation predicted value based on the conversion relation between the target radiation data and the photovoltaic power.

As shown in fig. 9, the method provided by the basic application is divided into various aspects; the data aspect can be divided into the required basic data sets including analysis data, multi-source satellite data and multi-mode weather forecast data. And carrying out data preprocessing on the basic data to obtain refined meteorological data. The analysis data is processed through a (preset) super-resolution reconstruction model, so that high-resolution analysis meteorological data and analysis radiation data are obtained. The multi-source satellite data determines radiation data through a (preset) space-time fusion model, a (preset) satellite inversion radiation model and a (preset) artificial intelligence correction model. And the multi-mode weather forecast data is used for determining the fusion weather data through a (preset) super-resolution reconstruction model, a (preset) multi-mode fusion model and a (preset) localization correction model. And performing migration learning on the refined meteorological data, obtaining a pre-training model through processing the analyzed data, and obtaining a radiation prediction model through processing the pre-training model, the fused meteorological data and the radiation data through a deep learning model. According to the photovoltaic month prediction method, steps of a photovoltaic month prediction technology are optimized by introducing multisource refined meteorological data and a deep learning algorithm. The method can improve the prediction precision and time resolution of the photovoltaic station and enrich the training data of the prediction model, thereby providing more accurate results for the photovoltaic month power prediction.

The application provides a high-time-resolution photovoltaic power prediction method based on meteorological data, which can use fusion meteorological prediction data of preset future time length to predict a radiation prediction value of the preset future time length through a radiation prediction model, wherein the preset future time length can be month or other long-time sequence values. The method solves the problem of insufficient training samples in the process of constructing the radiation prediction model, and also solves the problem of missing radiation monitoring data, thereby further improving the accuracy of the predicted radiation prediction value.

Corresponding to the method, the embodiment of the application also provides a high-time resolution photovoltaic power prediction device based on meteorological data, as shown in fig. 10, the device comprises:

an acquiring unit 1010, configured to acquire a high-resolution multi-source data set of a preset duration of a target area; the multi-source dataset comprises analysis data, satellite fusion data and multi-mode weather forecast data; the analysis data includes re-analysis weather data and analysis radiation data;

the determining unit 1020 is configured to train the deep learning model by using the analysis weather data as a training sample and the analysis radiation data as a sample tag, so as to obtain a pre-training model;

The determining unit 1020 is further configured to train the pre-training model with the fused meteorological data as a training sample and the radiation data as a sample tag, so as to obtain a radiation prediction model; the radiation data are obtained by inverting and correcting the satellite fusion data; the fusion weather data is obtained by fusing and correcting the multi-mode weather forecast data.

The functions of each functional unit of the high-time-resolution photovoltaic power prediction device based on meteorological data provided by the embodiment of the application can be realized through the steps of the method, so that the specific working process and beneficial effects of each unit in the high-time-resolution photovoltaic power prediction device based on meteorological data provided by the embodiment of the application are not repeated here.

The embodiment of the present application further provides an electronic device, as shown in fig. 11, including a processor 1110, a communication interface 1120, a memory 1130, and a communication bus 1140, where the processor 1110, the communication interface 1120, and the memory 1130 perform communication with each other through the communication bus 1140.

A memory 1130 for storing a computer program;

processor 1110, when executing the program stored in memory 1130, performs the following steps:

The communication bus mentioned above may be a peripheral component interconnect standard (Peripheral Component Interconnect, PCI) bus or an extended industry standard architecture (Extended Industry Standard Architecture, EISA) bus, or the like. The communication bus may be classified as an address bus, a data bus, a control bus, or the like. For ease of illustration, the figures are shown with only one bold line, but not with only one bus or one type of bus.

The communication interface is used for communication between the electronic device and other devices.

The Memory may include random access Memory (Random Access Memory, RAM) or may include Non-Volatile Memory (NVM), such as at least one disk Memory. Optionally, the memory may also be at least one memory device located remotely from the aforementioned processor.

The processor may be a general-purpose processor, including a central processing unit (Central Processing Unit, CPU), a network processor (Network Processor, NP), etc.; but also digital signal processors (Digital Signal Processing, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), field programmable gate arrays (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

Since the implementation manner and the beneficial effects of the solution to the problem of each device of the electronic apparatus in the foregoing embodiment may be implemented by referring to each step in the embodiment shown in fig. 8, the specific working process and the beneficial effects of the electronic apparatus provided by the embodiment of the present application are not repeated herein.

In yet another embodiment of the present application, a computer readable storage medium is provided, in which instructions are stored, which when run on a computer, cause the computer to perform a high time resolution photovoltaic power prediction method based on meteorological data as described in any of the above embodiments.

In yet another embodiment of the present application, a computer program product comprising instructions that, when run on a computer, cause the computer to perform a high time resolution photovoltaic power prediction method based on meteorological data as in any of the above embodiments is also provided.

It will be appreciated by those skilled in the art that embodiments of the present application may be provided as methods, systems, or computer program products. Accordingly, embodiments of the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, embodiments of the present application may take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

Embodiments of the present application are described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

While preferred embodiments of the present application have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts.

It will be apparent to those skilled in the art that various modifications and variations can be made in the embodiments of the present application without departing from the spirit or scope of the embodiments of the application.

Claims

1. A high time resolution photovoltaic power prediction method based on meteorological data, the method comprising:

taking the fusion meteorological data as a training sample, taking the radiation data as a sample label, and training the pre-training model to obtain a radiation prediction model; the radiation data are obtained by inverting and correcting the satellite fusion data; the fusion weather data is obtained by fusing and correcting the multi-mode weather forecast data;

wherein after obtaining the radiation prediction model, the method further comprises:

determining a photovoltaic power predicted value corresponding to the radiation predicted value based on the conversion relation between the target radiation data and the photovoltaic power;

the radiation data is obtained by inverting and correcting the satellite fusion data, and comprises the following steps:

Adopting a preset artificial intelligent correction model, correcting the initial radiation data based on the corresponding ground radiation data and ground photovoltaic power data of the target area, and obtaining the radiation data; the preset artificial intelligence correction model is a BP neural network model;

the fusion weather data is obtained by fusing and correcting the multi-mode weather forecast data, and comprises the following steps:

2. The method of claim 1, wherein acquiring high resolution, re-analyzed weather data and re-analyzed radiation data for a predetermined length of time for the target region comprises:

3. The method of claim 1, wherein obtaining high resolution satellite fusion data for a predetermined length of time for the target area comprises:

4. The method of claim 1, wherein based on the conversion relationship of the target radiation data to photovoltaic power, comprising:

5. A high time resolution photovoltaic power prediction device based on meteorological data, the device comprising:

the determining unit is further used for training the pre-training model by taking the fused meteorological data as a training sample and the radiation data as a sample label to obtain a radiation prediction model; the radiation data are obtained by inverting and correcting the satellite fusion data; the fusion weather data is obtained by fusing and correcting the multi-mode weather forecast data;

after the radiation prediction model is obtained, predicting ground weather data detected by a ground weather station at the position, and determining weather forecast data in a preset future time length;

6. An electronic device, characterized in that the electronic device comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory are in communication with each other through the communication bus;

a memory for storing a computer program;

a processor for implementing the steps of the method of any one of claims 1-4 when executing a program stored on a memory.

7. A computer readable storage medium, characterized in that the computer readable storage medium has stored therein a computer program which, when executed by a processor, implements the steps of the method of any of claims 1-4.