CN116384583A - Photovoltaic power prediction method based on multiple neural networks - Google Patents

Photovoltaic power prediction method based on multiple neural networks Download PDF

Info

Publication number
CN116384583A
CN116384583A CN202310409491.4A CN202310409491A CN116384583A CN 116384583 A CN116384583 A CN 116384583A CN 202310409491 A CN202310409491 A CN 202310409491A CN 116384583 A CN116384583 A CN 116384583A
Authority
CN
China
Prior art keywords
cloud
data
layer
neural network
current moment
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202310409491.4A
Other languages
Chinese (zh)
Inventor
朱一昕
徐伊洁
管梦瑶
毕恺韬
樊启高
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Jiangnan University
Original Assignee
Jiangnan University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Jiangnan University filed Critical Jiangnan University
Priority to CN202310409491.4A priority Critical patent/CN116384583A/en
Publication of CN116384583A publication Critical patent/CN116384583A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Electricity, gas or water supply
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/13—Satellite images
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • H02J3/004—Generation forecast, e.g. methods or systems for forecasting future energy generation
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • H02J3/38—Arrangements for parallely feeding a single network by two or more generators, converters or transformers
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
    • H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Abstract

The invention discloses a photovoltaic power prediction method based on a multi-neural network, which relates to the technical field of photovoltaic power generation, and comprises the following steps: acquiring cloud image and meteorological data at the current moment; determining actual data of solar radiation reaching the photovoltaic panel under the weather type of the current moment according to the cloud layer image of the current moment; inputting the actual data of the solar radiation quantity and the meteorological data at the current moment into a trained cyclic neural network sub-model to obtain a meteorological factor predicted value; and inputting the meteorological factor predicted value into the trained long-short-period memory neural network submodel to obtain a photovoltaic power predicted result at the current moment. The method collects data information of various different architectures, fully analyzes characteristics of meteorological data, cloud layer images and historical photovoltaic power data and relations among the data, and then fuses richer information to enable photovoltaic power prediction to be more accurate.

Description

Photovoltaic power prediction method based on multiple neural networks
Technical Field
The invention relates to the technical field of photovoltaic power generation, in particular to a photovoltaic power prediction method based on a multi-neural network.
Background
With the implementation of national new energy development strategy, more and more large-scale photovoltaic power generation is applied to power generation systems. However, the photovoltaic power generation is influenced by the illumination intensity, and the output power of the photovoltaic power generation is unstable, so that a plurality of problems can be brought to power grid planning, system scheduling and reliable and stable operation of a power grid, and the research of the photovoltaic power generation power prediction has important significance.
The light power prediction can be mainly divided into short-term prediction (1-3 days for power generation planning) and ultra-short-term prediction (0-4 hours for real-time scheduling), and a prediction method based on numerical weather prediction and a prediction method based on sky images and satellite cloud images are mainly adopted at home and abroad. The former predicts the atmospheric motion state and weather by solving the hydrodynamic and thermodynamic equation sets describing the weather evolution process through numerical calculation, and the method is not suitable for ultra-short-term prediction because of slow change of the data. The cloud cluster distribution situation is predicted through cloud cluster motion tracking, irradiance is measured and calculated to conduct ultra-short-term prediction, and prediction accuracy is required to be improved because irradiance is not obtained through actual sampling. The prediction method based on statistics and intelligent learning theory is mainly adopted in China, is suitable for ultra-short-term prediction, and can also prolong the prediction time by introducing meteorological data. Therefore, how to effectively use meteorological data, power generation data and site data is still a key problem.
Disclosure of Invention
Aiming at the problems and the technical requirements, the inventor provides a photovoltaic power prediction method based on a multi-neural network, which fully fuses the characteristics of meteorological data, cloud image data and historical photovoltaic power data and realizes the short-term and ultra-short-term prediction of photovoltaic power. The technical scheme of the invention is as follows:
a photovoltaic power prediction method based on a multi-neural network comprises the following steps:
acquiring cloud image and meteorological data at the current moment;
determining actual data of solar radiation reaching the photovoltaic panel under the weather type of the current moment according to the cloud layer image of the current moment;
inputting the actual data of the solar radiation quantity and the meteorological data at the current moment into a trained cyclic neural network sub-model to obtain a meteorological factor predicted value;
inputting the meteorological factor predicted value into a trained long-short-term memory neural network sub-model to obtain a photovoltaic power predicted result at the current moment; the trained long-term and short-term memory neural network sub-model comprises a mapping relation between a meteorological factor predicted value and photovoltaic power data.
According to a further technical scheme, the method for determining the actual data of the solar radiation quantity reaching the photovoltaic panel under the weather type of the current moment according to the cloud layer image of the current moment comprises the following steps:
inputting the cloud layer image at the current moment into a trained convolutional neural network submodel to obtain cloud amount index data;
determining the weather type of the current moment according to cloud cover index data;
determining a cloud coverage index according to the weather type to which the current moment belongs;
determining the influence weight of cloud cover shielding at the current moment on solar irradiance according to the cloud cover index;
and multiplying the influence weight by solar radiation data contained in the meteorological data to obtain actual solar radiation data reaching the photovoltaic panel.
The further technical scheme is that the method for determining the weather type of the current moment according to cloud cover index data comprises the following steps:
if the cloud amount index data is not greater than the first threshold value, determining that the weather type to which the current moment belongs is a sunny day; if the cloud quantity index data is larger than the first threshold value and smaller than the second threshold value, determining that the weather type to which the current moment belongs is cloudy; and if the cloud amount index data is not smaller than the second threshold value, determining that the weather type to which the current moment belongs is cloudy.
The cloud coverage index determining method based on the weather type comprises the following steps:
if the weather type of the current moment is determined to be a sunny day, converting the cloud layer image of the current moment into a numpy matrix, wherein each element in the numpy matrix is the pixel coordinate of a corresponding point in the image, extracting the pixel coordinates of all points into a single one-dimensional matrix, and dividing the total pixel point into two types by using K-means clustering in SPSS, wherein one type is a cloud pixel point and the other type is a sky pixel point;
let the total number of pixels be s and the number of cloud pixels be c, the cloud coverage index CC is: cc=c/s.
The further technical scheme is that the method for determining the cloud coverage index according to the weather type of the current moment further comprises the following steps:
if the weather type to which the current moment belongs is determined to be cloudy, defining a threshold NRBR of the ith pixel point as follows: NRBR i ＝(B i -R i )/(B i +R i )；
Wherein B is i R is the value of a B channel in an RGB channel of an ith pixel point in a cloud layer image i The value of an R channel in an RGB channel of an ith pixel point in the cloud layer image;
when NRBR i When the pixel point is not larger than the third threshold value, determining that the ith pixel point is a cloud pixel point, otherwise, determining that the ith pixel point is a sky pixel point;
let the total number of pixels of the image be s and the number of cloud pixels be c, the cloud coverage index CC is: cc=c/s.
The further technical scheme is that the method for determining the cloud coverage index according to the weather type of the current moment further comprises the following steps:
if the weather type of the current moment is determined to be cloudy, converting the cloud layer image into a gray level image, and setting a critical gray level value;
comparing the gray value of each point in the gray map with a critical gray value, and determining the point larger than the critical gray value as a cloud pixel point, or else as a sky pixel point;
let the total number of pixels of the image be s and the number of cloud pixels be c, the cloud coverage index CC is: cc=c/s.
The further technical scheme is that according to the cloud coverage index, the expression of the influence weight CR of the cloud cover shielding on the solar irradiance at the current moment is determined as follows: cr=1-CC; wherein CC is a cloud coverage index.
The training convolutional neural network submodel comprises an input layer, a convolutional layer, a pooling layer, a full communication layer and an output layer which are sequentially connected, and the training convolutional neural network submodel carries out cloud layer detection based on color characteristics according to the difference of color components contained in a white cloud layer and a blue sky in a cloud layer image to obtain cloud index data.
The method for obtaining the weather factor predicted value comprises the following steps of:
set as meteorological data and actual data sequence x= [ X ] of solar radiation quantity of model input data 1 ，X 2 ，…，X t ]When the current moment is t moment, the state value S of the hidden layer output of the circulating neural network submodel trained at t moment t The method comprises the following steps: s is S t ＝g(U·X t +W·S t-1 )；
Output value Y of output layer of trained cyclic neural network submodel at t moment t The method comprises the following steps: y is Y t ＝σ(V·S t )；
Wherein X is t The acquired t-moment meteorological data and solar radiation quantity actual data are obtained, wherein the t-moment meteorological data comprise temperature, humidity, wind speed and rainfall data at the t moment; y is Y t For outputting the weather factor predicted value output by the output layer, and S t Equal; u is the weight matrix in the input layer, V is the weight matrix in the output layer, W is the hidden layer state value S at the moment t-1 t-1 Inputting a weight matrix at the time t; sigma is the activation function of the output layer and g is the activation function of the hidden layer.
The method further comprises the following steps: the long-term and short-term memory neural network sub-model uses the historical photovoltaic power data and the meteorological factor predicted value output by the cyclic neural network sub-model as input data to carry out model training so as to establish a mapping relation between the meteorological factor predicted value and the photovoltaic power data;
the trained long-short-term memory neural network submodel is an LSTM network comprising three LSTM layers, two dropout layers and a dense layer, and is sequentially connected with the first LSTM layer, the first dropout layer, the second LSTM layer, the third LSTM layer, the second dropout layer and the dense layer in sequence.
The beneficial technical effects of the invention are as follows:
the method integrates the data information of various different architectures, fully analyzes the characteristics of meteorological data, data contained in cloud layer images, historical photovoltaic power data and the relation among the data, and then fuses unified, richer and more stereoscopic information than single data; the proper sub-model is selected according to the type characteristics of different data information, so that the characteristics of the data under different weather conditions are fully mined, and the fitting degree of the model is improved; and the cloud layer image is obtained in real time through local aerial shooting, so that the shielding degree of the cloud layer on solar radiation can be more intuitively and accurately embodied, the influence of cloud dynamic on photovoltaic power generation power prediction can be fully excavated, and the accuracy of photovoltaic power prediction is improved.
Drawings
Fig. 1 is a flowchart of a photovoltaic power prediction method based on a multi-neural network provided by the application.
Fig. 2 is a schematic block diagram of a convolutional neural network submodel provided herein.
Fig. 3 is a schematic block diagram of a recurrent neural network submodel provided herein.
Fig. 4 is a schematic block diagram of a long-short term memory neural network submodel provided herein.
Fig. 5 is a graph of a short-term prediction result of photovoltaic power obtained by using the method provided by the application.
Fig. 6 is a graph of the result of ultra-short-term prediction of photovoltaic power obtained by the method provided by the application.
Detailed Description
The following describes the embodiments of the present invention further with reference to the drawings.
As shown in fig. 1, the present embodiment provides a photovoltaic power prediction method based on a multi-neural network, including the following steps:
step 1: and acquiring cloud image and meteorological data at the current moment.
The weather data at the current moment comprises temperature, humidity, wind speed, rainfall and solar radiation data at the current moment.
Step 2: and determining actual data of the solar radiation quantity reaching the photovoltaic panel under the weather type of the current moment according to the cloud image of the current moment.
The cloud layer image at the current moment is obtained in real time through local aerial shooting, so that the influence degree of cloud layer shielding on solar radiation can be more intuitively reflected, and the accuracy of photovoltaic power prediction is improved.
The method specifically comprises the following steps of:
step 21: and inputting the cloud layer image at the current moment into a trained convolutional neural network submodel to obtain cloud amount index data.
Step 22: and determining the weather type of the current moment according to the cloud cover index data.
According to the acquired current day cloud amount index cl and the set first threshold value a and second threshold value b, dividing weather types into sunny days, cloudy days and cloudy days, wherein the specific dividing standard is as follows: if cl is less than or equal to a, determining that the weather type of the current moment is sunny; if a is less than cl and less than b, determining that the weather type to which the current moment belongs is cloudy; if cl is more than or equal to b, determining that the weather type to which the current moment belongs is cloudy.
In this embodiment, a=4 and b=8 are obtained according to experimental experience, and the threshold value may be adjusted according to actual conditions, which is not limited.
Step 23: and determining a cloud coverage index according to the weather type to which the current moment belongs.
(1) If the weather type of the current moment is determined to be a sunny day, converting the cloud layer image of the current moment into a numpy matrix, wherein each element in the numpy matrix is the pixel coordinate of a corresponding point in the image, extracting the pixel coordinates of all points into a single one-dimensional matrix, and dividing the total pixel point into two types by using K-means clustering in SPSS, wherein one type is a cloud pixel point, and the other type is a sky pixel point.
Let the total number of pixels be s, and the number of cloud pixels obtained by division be c, the cloud coverage index CC is: cc=c/s.
(2) If the weather type to which the current moment belongs is determined to be cloudy, defining a threshold NRBR of the ith pixel point as follows: NRBR i ＝(B i -R i )/(B i +R i )；
Wherein B is i R is the value of a B channel in an RGB channel of an ith pixel point in a cloud layer image i The value of the R channel in the RGB channel of the ith pixel point in the cloud image.
Setting the third threshold value as d, when NRBR i And when d is less than or equal to d, determining the ith pixel point as a cloud pixel point, otherwise, determining the ith pixel point as a sky pixel point. The expression of the subsequent calculation cloud coverage index CC is the same as in (1)And will not be described in detail herein.
In this embodiment, d=0.05 is obtained according to experimental experience, and the threshold value may be adjusted according to actual conditions, which is not limited.
(3) If the weather type of the current moment is determined to be cloudy, converting the cloud image into a gray level image, and setting a critical gray level value G, wherein the value of G is generally 80-90.
And comparing the gray value G of each point in the gray map with G, if G is larger than G, determining the point as a cloud pixel point, otherwise, determining the point as a sky pixel point. The expression of the subsequent calculation cloud coverage index CC is the same as that in (1), and will not be described here again.
Step 24: according to the cloud coverage index, determining the influence weight of cloud cover shielding at the current moment on solar irradiance, wherein the expression is as follows: cr=1-CC.
Step 25: and multiplying the influence weight CR by solar radiation data contained in the meteorological data to obtain actual solar radiation data reaching the photovoltaic panel.
Step 3: and inputting the actual data of the solar radiation quantity and other meteorological data at the current moment into a trained cyclic neural network sub-model to obtain a meteorological factor predicted value.
Step 4: and inputting the meteorological factor predicted value into the trained long-short-period memory neural network submodel to obtain a photovoltaic power predicted result at the current moment.
The trained long-term and short-term memory neural network sub-model comprises a mapping relation between a meteorological factor predicted value and photovoltaic power data.
Optionally, the method further comprises the step of training three sub-models using the cloud image, the meteorological data and the historical photovoltaic power data. Specific:
(1) Convolutional Neural Network (CNN) submodel
The convolutional neural network is good at extracting image features, can directly input an original image, and avoids complex preprocessing of the image. The built CNN submodel comprises an input layer, a convolution layer, a pooling layer, a full communication layer and an output layer which are sequentially connected, wherein the convolution layer and the pooling layer are key layers. As shown in fig. 2, the convolution layer is the most core part in the CNN submodel, and the main work of the layer is to convolve the input cloud layer image by convolution check to extract the characteristic data; the pooling layer is mainly used for sampling and dimension-reducing the characteristic data and improving the operation efficiency; after the convolution layer and the pooling layer, each node of the full-communication layer is communicated with all nodes of the upper layer, and the extracted features of the upper layer are integrated, and the parameters of the full-communication layer are the most generally due to the full-communication property of the full-communication layer; and finally, outputting cloud amount index data by an output layer, and determining a final weather type according to the cloud amount index data.
Based on the working principle of the CNN submodel, the trained CNN submodel carries out cloud layer detection based on color characteristics according to the difference of color components contained in white cloud layer and blue sky in the cloud layer image, and cloud amount index data is obtained.
(2) Cyclic neural network (RNN) submodel
The RNN circulates in the original network, and the neural network can recursively process the existing solar radiation actual data and other meteorological data, so that the special internal modeling can be applied to solving the problems of multiple associated information of space or time sequence. The RNN input layer can handle a wide time series structure, and find out the relation between data in different time periods, so that the RNN input layer is often applied to prediction in time series.
As shown in fig. 3, which shows the internal structure of a standard RNN neural network, there are interconnected time nodes in a hidden layer, and neurons in this layer can record the corresponding information of the previous moment and the current moment. In the figure: tanh is the activation function; x is X t The input value is the acquired t-moment weather data and solar radiation actual data in the embodiment, wherein the t-moment weather data comprises temperature, humidity, wind speed and rainfall data at the t moment; y is Y t Is an output value, in this embodiment, a weather factor predicted value output by the RNN output layer; s is S t Is the state value output by the RNN hidden layer at the moment t and Y t Equal.
Meteorological data and sun set as model input dataRadiation quantity actual data sequence x= [ X ] 1 ，X 2 ，…，X t ]When the current time is t time, the state value S output by the hidden layer of the RNN submodel at t time t The method comprises the following steps: s is S t ＝g(U·X t +W·S t-1 )；
Output layer output value Y of RNN submodel at time t t The method comprises the following steps: y is Y t ＝σ(V·S t )；
Wherein U is the weight matrix in the input layer, V is the weight matrix in the output layer, W is the hidden layer state value S at the moment t-1 t-1 Inputting a weight matrix at the time t; sigma is the activation function of the output layer and g is the activation function of the hidden layer.
The back propagation algorithm (BPTT) is a method applied to learning and training of RNN neural networks, wherein the included optimized micro loss function L adds the loss of power load data per time period during RNN learning. And iterating in the gradient descent process, so that proper cyclic neural network weight matrix parameters can be obtained.
The BPTT loss function expression is:
Figure BDA0004182813420000071
wherein L is t A loss function for each position of each sequence. And then calculating a new weight matrix by using the gradient expression of the weight matrix U, V, W through the loss function L, and repeatedly iterating until model training is completed so as to establish a mapping relation between the new weight matrix and each meteorological data at the moment t. The processing process of the trained RNN submodel on the weather data and the solar radiation actual data at the current moment is the same as that described above, and will not be described again here.
(3) Long-short term memory neural network (LSTM) submodel
The LSTM is a deep learning algorithm based on RNN and improved, and the core idea is to change the gradient with decimal value from a continuous multiplication form to an accumulation form, so that the LSTM can solve the problem of gradient disappearance when processing long-term time sequence information. Its general structure is shown in FIG. 4, where x t-1 、x t 、x t+1 Network input variables at times t-1, t and t+1 are respectively weather factor predicted values in the embodiment; h is a t-1 、h t 、h t+1 The short-term states output at times t-1, t and t+1 are respectively, and in the embodiment, photovoltaic power prediction results are obtained; c t-1 、c t The network memory states at the time t-1 and t are respectively the weights of the input meteorological factors in the embodiment; f (f) t 、i t 、o t A forget gate, an input gate and an output gate respectively; sigma is a sigmoid activation function; g t To update the control gate of the input state at time t.
As can be seen from FIG. 4, module A in LSTM reads input x t And output h t Information is transferred from the last step to the current step for calculation. The three control gates and the functions realized by the three control gates are respectively: forget the door, is responsible for screening out the information that needs to be discarded; an input gate for determining the amount of new information input; and the output gate is responsible for outputting the position of the state information. The three control gates are connected by sigmoid activation function sigma and a point-by-point multiplication, so as to realize the control of information. In the information transmission process, the weight c of each input characteristic at the time t-1 t-1 Updating self state to c by information processing of forget gate and input gate t The method comprises the steps of carrying out a first treatment on the surface of the Then, c is a pair of tanh functions t Processing, processing result and o t Multiplying output items to determine a photovoltaic power prediction result h at t moment required to be output by a module A in LSTM t This value is equal to c t Together to the next module.
The long-term memory network is a special recurrent neural network, solves the problems of gradient disappearance and gradient explosion, and is particularly good at processing time series. Because the weather factor predicted value data sequence and the historical photovoltaic power data sequence are both time sequences and have obvious time characteristics (such as the photovoltaic fluctuation rule of each day is approximately the same), the LSTM sub-model is used for combining the historical photovoltaic power data with the weather factor predicted value so as to realize analysis of the historical photovoltaic power sequence and prediction of photovoltaic power at future time. And carrying out model training on the LSTM sub-model by using the weather factor predicted value and the historical photovoltaic power data of the moment to be predicted, which are obtained by the RNN sub-model, as input data so as to establish a mapping relation between the weather data and the photovoltaic power. The historical photovoltaic power data comprises all photovoltaic power data of the previous moment, such as photovoltaic power data of the previous moment, and the like.
In this embodiment, the trained long-short-term memory neural network sub-model is an LSTM network including three LSTM layers, two dropout layers and one dense layer, and is sequentially connected according to the sequence of the first LSTM layer, the first dropout layer, the second LSTM layer, the third LSTM layer, the second dropout layer and the dense layer, and the working principle of the model is the same as that described above, and will not be repeated here. Referring to fig. 5 and 6, the change curves of the photovoltaic power short-term and ultra-short-term prediction results obtained by the photovoltaic power prediction method provided by the embodiment are close to the actual photovoltaic power change curve, so that the method is proved to be practical and has considerable accuracy of the photovoltaic power prediction results.
What has been described above is only a preferred embodiment of the present application, and the present invention is not limited to the above examples. It is to be understood that other modifications and variations which may be directly derived or contemplated by those skilled in the art without departing from the spirit and concepts of the present invention are deemed to be included within the scope of the present invention.

Claims (10)

1. A photovoltaic power prediction method based on a multi-neural network, the method comprising:
acquiring cloud image and meteorological data at the current moment;
determining actual data of solar radiation quantity reaching a photovoltaic panel under the weather type of the current moment according to the cloud layer image of the current moment;
inputting the solar radiation quantity actual data and the weather data at the current moment into a trained cyclic neural network sub-model to obtain a weather factor predicted value;
inputting the meteorological factor predicted value into a trained long-short-term memory neural network sub-model to obtain a photovoltaic power predicted result at the current moment; the trained long-term and short-term memory neural network sub-model comprises a mapping relation between a meteorological factor predicted value and photovoltaic power data.
2. The multi-neural network-based photovoltaic power prediction method according to claim 1, wherein the method for determining the actual data of the solar radiation amount reaching the photovoltaic panel under the weather type of the current moment according to the cloud image of the current moment comprises the following steps:
inputting the cloud layer image at the current moment into a trained convolutional neural network submodel to obtain cloud amount index data;
determining the weather type of the current moment according to the cloud cover index data;
determining a cloud coverage index according to the weather type of the current moment;
determining the influence weight of cloud cover shielding at the current moment on solar irradiance according to the cloud cover index;
and multiplying the influence weight by solar radiation amount data contained in the meteorological data to obtain actual solar radiation amount data reaching the photovoltaic panel.
3. The multi-neural network-based photovoltaic power prediction method according to claim 2, wherein the method for determining the weather type to which the current time belongs according to the cloud cover index data comprises the following steps:
if the cloud amount index data is not greater than the first threshold value, determining that the weather type to which the current moment belongs is a sunny day; if the cloud quantity index data is larger than the first threshold value and smaller than the second threshold value, determining that the weather type to which the current moment belongs is cloudy; and if the cloud amount index data is not smaller than the second threshold value, determining that the weather type to which the current moment belongs is cloudy.
4. The multi-neural network-based photovoltaic power prediction method according to claim 2, wherein the method of determining a cloud coverage index according to the weather type to which the current time belongs comprises:
if the weather type of the current moment is determined to be a sunny day, converting the cloud layer image of the current moment into a numpy matrix, wherein each element in the numpy matrix is the pixel coordinate of a corresponding point in the image, extracting the pixel coordinates of all points into a single one-dimensional matrix, and dividing the total pixel point into two types by using K-means clustering in SPSS, wherein one type is a cloud pixel point and the other type is a sky pixel point;
let the total number of pixels be s and the number of cloud pixels be c, the cloud coverage index CC is: cc=c/s.
5. The multi-neural network-based photovoltaic power prediction method according to claim 2, wherein the method of determining a cloud coverage index according to the weather type to which the current time belongs further comprises:
if the weather type to which the current moment belongs is determined to be cloudy, defining a threshold NRBR of the ith pixel point as follows: NRBR i ＝(B i -R i )/(B i +R i )；
Wherein B is i R is the value of a B channel in an RGB channel of an ith pixel point in a cloud layer image i The value of an R channel in an RGB channel of an ith pixel point in the cloud layer image;
when NRBR i When the pixel point is not larger than a third threshold value, determining that the ith pixel point is a cloud pixel point, otherwise, determining that the ith pixel point is a sky pixel point;
let the total number of pixels of the image be s and the number of cloud pixels be c, the cloud coverage index CC is: cc=c/s.
6. The multi-neural network-based photovoltaic power prediction method according to claim 2, wherein the method of determining a cloud coverage index according to the weather type to which the current time belongs further comprises:
if the weather type of the current moment is determined to be cloudy, converting the cloud layer image into a gray level image, and setting a critical gray level value;
comparing the gray value of each point in the gray map with the critical gray value, and determining the point larger than the critical gray value as a cloud pixel point, or else as a sky pixel point;
let the total number of pixels of the image be s and the number of cloud pixels be c, the cloud coverage index CC is: cc=c/s.
7. The multi-neural network-based photovoltaic power prediction method according to claim 2, wherein the expression for determining the influence weight CR of cloud cover shielding on solar irradiance at the current moment according to the cloud cover index is: cr=1-CC; wherein CC is the cloud coverage index.
8. The multi-neural-network-based photovoltaic power prediction method according to claim 2, wherein the trained convolutional neural network sub-model comprises an input layer, a convolutional layer, a pooling layer, a full communication layer and an output layer which are sequentially connected, and cloud layer detection based on color characteristics is performed by the trained convolutional neural network sub-model according to the difference of color components contained in white cloud layers and blue sky in a cloud layer image, so as to obtain cloud quantity index data.
9. The multi-neural network-based photovoltaic power prediction method according to claim 1, wherein the method for inputting the solar radiation amount actual data and the weather data at the current moment into the trained cyclic neural network sub-model to obtain the weather factor predicted value comprises the following steps:
set as meteorological data and actual data sequence x= [ X ] of solar radiation quantity of model input data 1 ，X 2 ，…，X t ]The current time is t time, and the state value S output by the hidden layer of the trained cyclic neural network submodel is at t time t The method comprises the following steps: s is S t ＝g(U·X t +W·S t-1 )；
The output value Y of the output layer of the trained cyclic neural network submodel at the time t t The method comprises the following steps: y is Y t ＝σ(V·S t )；
Wherein X is t For the acquired weather data and actual solar radiation data at the time t,the weather data at the moment t comprise temperature, humidity, wind speed and rainfall data at the moment t; y is Y t A weather factor predicted value output by the output layer is equal to S t Equal; u is the weight matrix in the input layer, V is the weight matrix in the output layer, W is the hidden layer state value S at the moment t-1 t-1 Inputting a weight matrix at the time t; sigma is the activation function of the output layer and g is the activation function of the hidden layer.
10. The multi-neural network-based photovoltaic power prediction method of claim 1, further comprising: the long-term and short-term memory neural network sub-model uses the historical photovoltaic power data and the meteorological factor predicted value output by the cyclic neural network sub-model as input data to carry out model training so as to establish a mapping relation between the meteorological factor predicted value and the photovoltaic power data;
the trained long-short-term memory neural network submodel is an LSTM network comprising three LSTM layers, two dropout layers and a dense layer, and is sequentially connected with the first LSTM layer, the first dropout layer, the second LSTM layer, the third LSTM layer, the second dropout layer and the dense layer according to the sequence of the first LSTM layer, the first dropout layer, the second LSTM layer, the third LSTM layer, the second dropout layer and the dense layer.
CN202310409491.4A 2023-04-17 2023-04-17 Photovoltaic power prediction method based on multiple neural networks Pending CN116384583A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310409491.4A CN116384583A (en) 2023-04-17 2023-04-17 Photovoltaic power prediction method based on multiple neural networks

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310409491.4A CN116384583A (en) 2023-04-17 2023-04-17 Photovoltaic power prediction method based on multiple neural networks

Publications (1)

Publication Number Publication Date
CN116384583A true CN116384583A (en) 2023-07-04

Family

ID=86961517

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310409491.4A Pending CN116384583A (en) 2023-04-17 2023-04-17 Photovoltaic power prediction method based on multiple neural networks

Country Status (1)

Country Link
CN (1) CN116384583A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116721356A (en) * 2023-08-10 2023-09-08 深圳航天科创泛在电气有限公司 Output power prediction method of photovoltaic system and related equipment
CN117411088A (en) * 2023-12-13 2024-01-16 山东鼎鑫能源工程有限公司 Operation control method of photovoltaic power generation system and photovoltaic power generation system

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116721356A (en) * 2023-08-10 2023-09-08 深圳航天科创泛在电气有限公司 Output power prediction method of photovoltaic system and related equipment
CN116721356B (en) * 2023-08-10 2023-11-24 深圳航天科创泛在电气有限公司 Output power prediction method of photovoltaic system and related equipment
CN117411088A (en) * 2023-12-13 2024-01-16 山东鼎鑫能源工程有限公司 Operation control method of photovoltaic power generation system and photovoltaic power generation system
CN117411088B (en) * 2023-12-13 2024-02-27 山东鼎鑫能源工程有限公司 Operation control method of photovoltaic power generation system and photovoltaic power generation system

Similar Documents

Publication Publication Date Title
US20220373984A1 (en) Hybrid photovoltaic power prediction method and system based on multi-source data fusion
CN113962364B (en) Multi-factor power load prediction method based on deep learning
Ghaderi et al. Deep forecast: Deep learning-based spatio-temporal forecasting
CN116384583A (en) Photovoltaic power prediction method based on multiple neural networks
CN112561058B (en) Short-term photovoltaic power prediction method based on Stacking-integrated learning
CN110046787A (en) A kind of urban area charging demand for electric vehicles spatio-temporal prediction method
CN113128113B (en) Lean information building load prediction method based on deep learning and transfer learning
CN110232476A (en) A kind of micro-grid load prediction technique based on deep learning
Liu et al. Heating load forecasting for combined heat and power plants via strand-based LSTM
CN112257614B (en) Station building passenger flow space-time distribution prediction method based on graph convolution network
CN114792158A (en) Multi-wind-farm short-term power prediction method based on space-time fusion graph neural network
CN112329990A (en) User power load prediction method based on LSTM-BP neural network
CN111242355A (en) Photovoltaic probability prediction method and system based on Bayesian neural network
CN114462718A (en) CNN-GRU wind power prediction method based on time sliding window
Aliberti et al. Forecasting Short-term Solar Radiation for Photovoltaic Energy Predictions.
CN114239991A (en) Building heat supply load prediction method, device and equipment based on data driving
CN116894504A (en) Wind power cluster power ultra-short-term prediction model establishment method
CN115860215A (en) Photovoltaic and wind power generation power prediction method and system
CN114648176A (en) Wind-solar power consumption optimization method based on data driving
CN109447843B (en) Photovoltaic output prediction method and device
CN116402194A (en) Multi-time scale load prediction method based on hybrid neural network
CN116090635A (en) Meteorological-driven new energy generation power prediction method
Wang et al. Ultra-short-term regional PV power forecasting based on fluctuation pattern recognition with satellite images
CN111461462B (en) Daily load prediction method based on trellis Net-LSTM
CN114372418A (en) Wind power space-time situation description model establishing method

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination