CN115796004A - Photovoltaic power station ultra-short term power intelligent prediction method based on SLSTM and MLSTNet models - Google Patents
Photovoltaic power station ultra-short term power intelligent prediction method based on SLSTM and MLSTNet models Download PDFInfo
- Publication number
- CN115796004A CN115796004A CN202211375921.7A CN202211375921A CN115796004A CN 115796004 A CN115796004 A CN 115796004A CN 202211375921 A CN202211375921 A CN 202211375921A CN 115796004 A CN115796004 A CN 115796004A
- Authority
- CN
- China
- Prior art keywords
- photovoltaic power
- model
- prediction
- data
- lstm
- 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
Links
Images
Classifications
-
- 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
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses an intelligent photovoltaic power station ultra-short term power prediction method based on SLSTM and MLSTNet models, which specifically comprises the following steps: firstly, processing missing values and abnormal values of original data; then, performing Spearman correlation analysis on the photovoltaic power variable and the meteorological data variable, and determining input variables of a prediction model as temperature, humidity and irradiation intensity; selecting temperature, humidity and irradiation intensity as clustering variables, constructing statistical characteristics of the clustering variables, performing similar daily clustering on photovoltaic historical data by adopting a fuzzy C-means clustering algorithm to obtain Type1-4 similar daily data, and performing normalization processing on the Type1-4 similar daily data; then dividing the data sets of similar days into a training set, a verification set and a test set; constructing an SLSTM photovoltaic power ultra-short term prediction model and an MLSTNet photovoltaic power ultra-short term prediction model, and optimizing parameters of the SLSTM photovoltaic power ultra-short term prediction model and the MLSTNet photovoltaic power ultra-short term prediction model by using an ultra-frequency Bayes optimizer to perform ultra-short term power prediction; and finally, generating a photovoltaic power ultra-short term prediction result on the test set. The method can better track the power curve trend of the photovoltaic power station in four hours in the future, and has the advantages of high running speed and high prediction precision.
Description
Technical Field
The invention relates to the technical field of photovoltaic power generation prediction, in particular to an intelligent prediction method for ultra-short term power of a photovoltaic power station based on SLSTM and MLSTNet models.
Technical Field
In recent years, the problem of environmental pollution is becoming more serious, and the problem of lack of non-renewable energy is becoming more and more prominent. In the new energy development strategy in China, solar energy is developed and utilized on a large scale, and a photovoltaic power generation system is developed rapidly. According to the latest data of the national energy source bureau, the installed capacity of newly added photovoltaic power generation grid-connected machines in 2021 years in China is about 5300 ten thousand kilowatts, the newly added distributed photovoltaic power generation is about 2900 ten thousand kilowatts, the installed capacity of the newly added photovoltaic power generation accounts for about 55% of the whole newly added photovoltaic power generation machines, the first breakthrough of the installed capacity is 50% in history, and the development trend of the centralized and distributed photovoltaic power generation is obvious. By the end of 2021, the installed capacity of the photovoltaic power generation grid-connected system reaches 3.06 hundred million kilowatts, and breaks through the large power limit of 3 hundred million kilowatts, wherein the distributed photovoltaic power generation system reaches 1.075 hundred million kilowatts, breaks through 1 hundred million kilowatts, and accounts for about one third of the installed capacity of the whole photovoltaic power generation grid-connected system. The large-scale photovoltaic power generation grid connection is an effective way for realizing emission reduction, but the grid connection of a photovoltaic system also brings reliability and stability problems, so that the real-time scheduling and safe operation of a power grid are challenged by the characteristics of the grid connection while solar energy is utilized. On a time scale, photovoltaic power prediction can be divided into medium-long term (> 24 h), short term (4-24 h) and ultra-short term (0-4 h). The photovoltaic power generation ultra-short-term power prediction is the basis for adjusting the scheduling plan in advance by a power grid scheduling department, and the higher the photovoltaic prediction precision is, the more negative influence of grid connection of a photovoltaic system on a power grid can be reduced, and the realization of a new energy consumption target is promoted.
The existing photovoltaic power prediction technology can be divided into direct prediction and indirect prediction according to the classification of prediction processes; according to the classification of prediction space scale, single field prediction and region prediction can be divided; according to the classification of the prediction time scale, the prediction method can be divided into ultra-short term prediction, medium term prediction and long term prediction; according to the classification of prediction forms, point prediction, interval prediction and probability prediction can be classified. Classification by prediction methods can be classified into physical methods, statistical methods, and machine learning methods. However, for any classification, prediction studies can be performed by different prediction methods, such as physical methods, statistical methods, and machine learning methods.
The physical method is that a mathematical model is established according to a photovoltaic power generation principle, solar radiation, temperature, humidity, cloud cover, air pressure and wind speed obtained by numerical weather forecast (NWP) are utilized, parameters such as a photovoltaic system installation angle, photovoltaic array conversion efficiency and battery conditions are combined to establish a physical model, and then photovoltaic power is directly calculated. The physical prediction model does not require historical data, but relies on detailed site geographic information, accurate meteorological data, and complete photovoltaic cell information. The prediction accuracy of the physical method greatly depends on the accuracy of the NWP information, but currently, a bottleneck is met in improving the accuracy of the NWP.
Common statistical prediction methods have time-series methods, regression analysis methods, grey theory, fuzzy theory, multi-source data-driven methods, and spatio-temporal correlation methods. The statistical method is that after historical data such as solar radiation, photovoltaic power generation capacity and the like are processed, a relevant mapping relation (namely a data model) between input and output data is established through curve fitting, parameter estimation and relevance analysis, and therefore prediction of future photovoltaic power generation capacity is achieved. However, the statistical method is implemented on the premise that a large amount of correct historical data is needed after processing, and data acquisition and calculation processing are difficult in the implementation process.
Machine learning has the ability to efficiently extract and map high-dimensional complex nonlinear features directly to the output. Machine learning-based prediction methods take advantage of this and have become one of the most common methods for predicting time series. Machine learning methods such as Artificial Neural Networks (ANNs), random Forests (RFs) and Deep Extreme Learning Machines (DELMs) can manage nonlinearity in meteorological data, and are widely applied to photovoltaic power generation output prediction. Machine learning methods have also recently been applied to ultra-short term photovoltaic power prediction, requiring higher spatial and temporal resolution than traditional day-ahead predictions.
The main reason influencing the photovoltaic power prediction accuracy is randomness, intermittence and fluctuation of photovoltaic power generation power, which are derived from factors such as complicated and variable weather states, cloud layer movement and ambient temperature, and further improvement of the photovoltaic power prediction accuracy is hindered. Meanwhile, because meteorological data have certain prediction errors, the errors can be further amplified when the meteorological data are used for photovoltaic prediction, and therefore a prediction algorithm with high precision becomes an urgent problem to be solved.
Disclosure of Invention
The invention aims to provide an intelligent photovoltaic power station ultra-short term power prediction method based on SLSTM and MLSTNet models, and solves the problem that the photovoltaic power generation ultra-short term power prediction result is inaccurate.
The technical scheme adopted by the invention is that the photovoltaic power station ultra-short term power intelligent prediction method based on SLSTM and MLSTNet models is implemented according to the following steps:
step 1: acquiring preprocessed meteorological data and historical photovoltaic power data, performing correlation analysis on the photovoltaic power and the meteorological data, and determining input variables of a prediction model;
step 2: selecting a clustering variable, and constructing statistical characteristics of the clustering variable;
and 3, step 3: according to the clustering variables and the statistical characteristics thereof selected in the step 2, clustering photovoltaic power historical data by adopting a fuzzy C-means clustering algorithm to obtain Type1, type2, type3 and Type4 similar day data sets;
and 4, step 4: dividing the Type1, type2, type3 and Type4 similar day data sets in the step 3 into a training set, a verification set and a test set;
and 5: carrying out normalization processing on the training set, the verification set and the test set in the step 4;
step 6: constructing an SLSTM photovoltaic power station ultra-short-term power prediction model, setting relevant parameters of model training, and inputting data of a training set into the model for training;
and 7: constructing an MLSTNet photovoltaic power station ultra-short-term power prediction model, setting relevant parameters of model training, and inputting data of a training set into the model for training;
and 8: respectively inputting the data of the verification set into the SLSTM and MLSTNet photovoltaic power station ultra-short-term power prediction models trained in the step 6 and the step 7, verifying the prediction result, and respectively returning to the step 6 and the step 7 if the error between the prediction result and the true value is larger; if the error with the true value is smaller, performing step 9;
and step 9: respectively inputting the data of the test set into the SLSTM and MLSTNet photovoltaic power station ultra-short-term power prediction models trained in the steps 6 and 7 to perform photovoltaic power ultra-short-term prediction;
step 10: performing inverse normalization processing on the ultra-short term prediction result in the step 9;
the present invention is also characterized in that,
in the step 1, the method specifically comprises the following steps:
step 1.1: selecting preprocessed meteorological data and preprocessed photovoltaic power data;
the time resolution of the meteorological variable and the photovoltaic power variable is 15min, and the wind speed, the wind direction, the temperature, the humidity, the air pressure and the irradiation intensity are selected as original meteorological data variables;
step 1.2: measuring the degree of correlation among a plurality of meteorological variables by using a Spearman correlation coefficient R;
step 1.3: and selecting meteorological data variables of which the Spearman correlation coefficient R absolute value of the photovoltaic power is not less than 0.5, and inputting the meteorological data variables as a prediction model.
In the step 2, the method specifically comprises the following steps:
step 2.1: selecting a meteorological data variable of which the Spearman correlation coefficient R absolute value of the photovoltaic power is not less than 0.5 as a clustering variable;
step 2.2: and selecting the average value, the standard deviation and the maximum value of the clustering variables as statistical characteristics.
In step 3, the method specifically comprises the following steps:
step 3.1: calculating the numerical values of 3 statistical characteristics of the clustering variables on each day according to the statistical characteristics of the clustering variables constructed in the step 2;
step 3.2: determining the number c of data clustering categories, initializing a clustering center matrix V, giving a fuzzy weighting index m, initializing a membership degree matrix U, and giving a termination standard epsilon of an algorithm, wherein epsilon is positive and infinitesimal;
step 3.3: calculating all clustering centers of the t iteration according to the formula (5) to obtain a clustering center matrix:
in the formula: u. of ij The membership degree of the ith sample belonging to the jth class; x is the number of i Is a sample point; m is a fuzzy weighting index(ii) a t is the number of iterations; c represents the number of clustering categories; (ii) a d ij Is the distance from the ith sample to the class j center;
step 3.4: updating the membership degree matrix, wherein the calculation method is shown as the following formula:
in the formula: u. of ij The membership degree of the ith sample belonging to the jth class; x is the number of i Is a sample point; m is a fuzzy weighting index; t is the number of iterations; c represents the number of clustering categories; d ij Is the distance from the ith sample to the class j center;
step 3.5: calculate | U (t) -U (t-1) And verifying whether an iteration stop condition | U is satisfied (t) -U (t-1) |<Epsilon, if the condition is met, stopping iteration; if the condition is not met, then steps 3.3 and 3.4 are repeated until the condition is reached, resulting in a Type1, type2, type3 and Type4 similar day dataset.
In step 6, the method specifically comprises the following steps:
step 6.1: inputting the normalized Type1 similar day data set into an LSTM neural network in a single step rolling input mode to extract a characteristic vector of dynamic change of photovoltaic power;
step 6.2: re-inputting the output special direction vector into the LSTM neural network so as to realize single-step rolling prediction;
for example: and taking the meteorological variable and the photovoltaic power at the t-1 moment and the meteorological variable at the t moment as input to obtain a photovoltaic power predicted value at the t moment. Taking the meteorological variable and the photovoltaic power predicted value at the time t and the meteorological variable at the time t +1 as input, and obtaining the photovoltaic power predicted value at the time t +1 until the prediction is finished;
step 6.3: performing linear fitting on the prediction result obtained in the step 6.2 and the actual photovoltaic power to obtain a final prediction result;
Y LSTM ＝β 1 h t +β 0
in the formula: h is t Output of the result, beta, for the LSTM neural network 0 And beta 1 Is a linear fitting coefficient, Y LSTM The final photovoltaic power ultra-short-term predicted value is obtained;
step 6.4: setting parameters of the model and training the model;
the SLSTM model network structure comprises three layers of LSTMs, a full connection layer and a correction model;
setting the number of LSTM network layers to be 3, optimizing the number of first layer neurons, the number of second layer neurons, the number of third layer neurons, the learning rate and batch processing parameters by a Bayesian optimizer, automatically optimizing in the operation of a model, wherein the maximum iteration number is 100, all layers use Relu activation functions, and the optimizer is Adam;
the training set of Type1 similar day data sets was imported into the constructed SLSTM for model training.
In step 7, the method specifically comprises the following steps:
step 7.1: dividing the normalized Type2, type3 and Type4 similar day data sets into long-term historical data and short-term historical data according to the time interval between the data sets and the prediction day;
step 7.2: inputting the long-term historical data and the short-term historical data into a TCANN neural network, and extracting a time characteristic vector of the photovoltaic power by utilizing a causal convolution layer, an expansion convolution layer and a sparse attention mechanism layer of the TCANN neural network;
step 7.3: respectively sending the output of the TCANN neural network of the step 7.2 into an LSTM neural network model, an LSTM-Skip neural network model and an MTNet model, extracting feature vectors of photovoltaic power dynamic change by using an LSTM layer in the LSTM neural network model and the LSTM-Skip neural network model, and improving the long and short memory extraction capability of data by using the MTNet model;
step 7.4: adding the prediction results of the LSTM neural network, the LSTM-Skip neural network and the MTNet in the step 7.3 to obtain a final prediction result;
Y MLSTNet ＝Y LSTM +Y LSTM-Skip +Y MTNet
in the formula: y is LSTM For the LSTM neural network output results, Y LSTM-Skip For LSTM-Skip neural network transmissionAs a result, Y MTNet Output results for MTNet neural network, Y MLSTNet The final photovoltaic power ultra-short-term predicted value is obtained;
step 7.5: setting parameters of the model and training the model;
the MLSTNet model network structure consists of a TCANN neural network, an LSTM-Skip neural network and an MTNet neural network;
parameters of the MLSTNet model are optimized by a Bayesian optimizer, automatic optimization is carried out in the running of the model, the maximum iteration number is 100, all layers use Relu activation functions, and the optimizer is Adam;
inputting training sets of Type2, type3 and Type4 similar day data sets into the constructed MLSTNet for model training.
The invention has the beneficial effects that:
the photovoltaic power station ultra-short term power intelligent prediction method based on the SLSTM and the MLSTNet models is based on a Spearman correlation analysis method, so that model input variables can be determined, and invalid information of irrelevant data in historical data is reduced to improve the training speed of the models; selecting meteorological data variables with high correlation with photovoltaic power as clustering variables, and selecting 3 statistical characteristics such as average values of the clustering variables as clustering characteristics to comprehensively reflect the characteristics of randomness, intermittency and volatility of daily historical data so as to facilitate efficient clustering by a fuzzy C-means clustering algorithm; the SLSTM model integrates a single step rolling model, an LSTM model and a correction model, and compared with a traditional deep learning prediction model, the SLSTM model has more temporal input and higher prediction accuracy; the MLSTNet model integrates a TCANN model, an LSTM-Skip model and an MTNet model, compared with a traditional deep learning prediction model, the MLSTNet model improves the extraction capability of time characteristics and has higher prediction precision; optimizing the model parameters by using an over-frequency Bayesian optimizer, and solving the problem of model parameter optimization; when the future photovoltaic power is predicted by taking 15min as the interval time, under the condition of non-clear weather, the photovoltaic power taking 15min as the interval presents more change characteristics, the trend of a photovoltaic power station power curve in four hours in the future can be well tracked, and the method has the advantages of high operation speed and high prediction precision.
Drawings
FIG. 1 is a flow chart of the photovoltaic power plant ultra-short term power intelligent prediction method based on SLSTM and MLSTNet models;
FIG. 2 is a flow chart of a fuzzy C-means clustering algorithm in the photovoltaic power station ultra-short term power intelligent prediction method based on SLSTM and MLSTNet models;
FIG. 3 is a graph of the results of clustering a data set using a fuzzy C-means clustering algorithm in the method of the present invention;
detailed description of the invention
The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.
The invention relates to a photovoltaic power station ultra-short term power intelligent prediction method based on an SLSTM (selective laser desorption/ionization) model and an MLSTNet model, which is specifically implemented according to the following steps as shown in figure 1:
step 1: acquiring preprocessed meteorological data and historical photovoltaic power data, performing variable correlation analysis on the photovoltaic power and the meteorological data, and determining input variables of a prediction model, wherein the method specifically comprises the following steps:
step 1.1: selecting preprocessed meteorological data and photovoltaic power generation power data, wherein the time resolution of meteorological variables and the time resolution of photovoltaic power variables are kept consistent;
the time resolution of the meteorological variable and the photovoltaic power variable is 15min, the wind speed, the wind direction, the temperature, the humidity, the air pressure and the irradiation intensity are selected as original meteorological data variables, and the power at night is 0 because the photovoltaic power station does not generate electricity at night;
filling vacancy values of historical meteorological data and photovoltaic power generation power data through an interpolation method, and removing abnormal values through a box line graph;
step 1.2: measuring the degree of correlation among a plurality of meteorological variables by using a Spearman correlation coefficient R;
the Spearman correlation coefficient R is defined as follows:
in the formula: r represents a Spearman correlation coefficient; d i Is the difference of the two groups of variable rank; n is the total number of samples;
the closer the Spearman correlation coefficient R is to 1, the higher the correlation between the meteorological variable and the photovoltaic power is, and if the correlation coefficient R is a positive number, the positive correlation is shown; if the correlation coefficient R is negative, indicating that the correlation is negative;
step 1.3: selecting meteorological variables of which the Spearman correlation coefficient R absolute value of the photovoltaic power is not less than 0.5, and inputting the meteorological variables as a prediction model;
in this example, the Spearman correlation coefficients R for the photovoltaic power variable and the various meteorological variables are shown in table 1:
TABLE 1 Meteorological variables
Selecting a plurality of meteorological variables with high correlation with photovoltaic power, namely meteorological variables with the Spearman correlation coefficient absolute value of the photovoltaic power not less than 0.5, and inputting the meteorological variables as a prediction model, so that the temperature, the humidity and the irradiation intensity are selected as the meteorological variables;
step 2: selecting a clustering variable, and constructing statistical characteristics of the clustering variable, wherein the statistical characteristics specifically comprise the following steps:
step 2.1: selecting a meteorological variable with a Spearman correlation coefficient absolute value of photovoltaic power not less than 0.5 as a clustering variable, and selecting temperature, humidity and irradiation intensity as the clustering variable in the embodiment;
step 2.2: selecting the average value, the standard deviation and the maximum value of the clustering variables as statistical characteristics;
the mean, standard deviation and maximum are defined as follows:
X max ＝max{x 1 ,x 2 ,…,x n-1 ,x n }
in the formula (I), the compound is shown in the specification,
denotes the mean value of the variables, σ denotes the standard deviation of the variables, X max Denotes the maximum value of the variable, x i A certain sample representing the variable, n represents the total number of samples of the variable;
and step 3: according to the clustering variables and the statistical characteristics thereof selected in the step 2, as shown in fig. 2, similar daily clustering of photovoltaic historical data is performed by adopting a fuzzy C-means clustering algorithm to obtain a photovoltaic similar daily data set, wherein the specific positions are as follows:
step 3.1: calculating the numerical values of 3 statistical characteristics of the clustering variables on each day according to the statistical characteristics of the clustering variables constructed in the step 2;
step 3.2: determining the number c of data clustering categories, initializing a clustering center matrix V, giving a fuzzy weighting index m, initializing a membership matrix U, and giving a termination standard epsilon of an algorithm;
step 3.3: calculating all clustering centers of the t iteration according to the following formula to obtain a clustering center matrix:
in the formula: u. of ij The membership degree of the ith sample belonging to the jth class; x is the number of i Is a sample point; m is a fuzzy weighting index; t is the number of iterations; c represents the number of clustering categories;
step 3.4: updating the membership degree matrix, wherein the calculation method is shown as the following formula:
in the formula: u. of ij For the ith sample belonging to class jDegree of membership of; x is the number of i Is a sample point; m is a fuzzy weighting index; t is the number of iterations; c represents the number of clustering categories; d ij Is the distance from the ith sample to the class j center;
step 3.5: calculate | U (t) -U (t-1) And verifying whether an iteration stop condition | U is satisfied (t) -U (t-1) |<Epsilon, if the condition is met, stopping iteration; if the conditions are not met, repeating the step 3.3 and the step 3.4 until the conditions are met, and finally obtaining similar day data sets under various weather conditions.
And 4, step 4: carrying out normalization processing on the photovoltaic similar day data set so as to eliminate the influence of different dimensions among different variables on a prediction result;
the normalization method is shown as follows:
in the formula: x denotes the data to be normalized, x normal watch Data after normalization, x max Representing the maximum value, x, of some variable data min Represents the minimum value of a certain variable data;
and 5: dividing similar day data sets under various weather types into a training set, a verification set and a test set;
step 6: constructing a single-step rolling long-short term memory neural network correction (SLSTM) ultra-short term prediction model, setting LSTM model parameters, setting relevant parameters of model training, and inputting a training set of similar day data sets into the constructed SLSTM model for model training;
step 6.1: inputting the normalized similar day data set into an LSTM neural network in a single step rolling input mode, and extracting feature vectors of photovoltaic power dynamic change by using an LSTM layer of the LSTM neural network;
step 6.2: re-inputting the output special direction vector into the LSTM neural network so as to realize single-step rolling prediction;
for example: and taking the meteorological variable and the photovoltaic power at the t-1 moment and the meteorological variable at the t moment as input to obtain a photovoltaic power predicted value at the t moment. Taking the meteorological variable and the photovoltaic power predicted value at the time t and the meteorological variable at the time t +1 as input, obtaining the photovoltaic power predicted value at the time t + 1, and stopping until the fourth hour is predicted;
the calculation process of the LSTM neural network is as follows:
the forget gate determines which input information is to be deleted from the memory cell state as shown in the following equation:
f t ＝σ(W f x t +U f h t-1 +b f )
inputting the output value of the previous moment and the input value of the current moment into the input gate, and obtaining the output value of the input gate after calculation, as shown in the following formula:
i t ＝σ(W i x t +U i h t-1 +b i )
inputting the output value of the previous moment and the input value of the current moment into an input gate, and obtaining the state of the candidate cell after calculation, wherein the state is shown as the following formula:
the current cell state is updated as shown in the following formula:
c t ＝f t ⊙c t-1 +i t ⊙c t-1
inputting the output value of the previous moment and the input value of the current moment into an output gate, and obtaining the output value of the output gate after calculation, wherein the output value is shown as the following formula;
o t ＝σ(W o x t +U o h t-1 +b o )
calculating the output of the output gate and the cell state to obtain an output value as shown in the following formula;
h t ＝o t ⊙tanh(c t )
in the formula: i.e. i t Representing the output of the input gate at time t, f t Indicating forgetting at time tOutput of the gate, o t Representing the output of the output gate at time t, h t-1 Represents data output information at time t-1, h t Indicating data output information at time t, x t Representing data input information at time t, c t-1 Represents the state of the cells at time t-1, c t Indicating the cellular state at time t, W i 、W f 、W o 、W c 、U i 、U f 、U o 、U c Weight matrix representing the input matrix of the aforementioned three gates, b i 、b f 、b o Representing the bias vectors of three gates, sigma representing a Sigmoid activation function, and tanh representing a hyperbolic tangent activation function;
step 6.3: performing linear fitting on the prediction result obtained in the step 6.2 and the actual photovoltaic power to obtain a final prediction result;
Y LSTM ＝β 1 h t +β 0
in the formula: h is t Output of the result, beta, for the LSTM neural network 0 And beta 1 Is a linear fitting coefficient, Y LSTM The final photovoltaic power ultra-short-term predicted value is obtained;
step 6.4: setting parameters of the model and training the model;
the SLSTM model network structure comprises three LSTMs, a full connection layer and a correction model;
setting the number of LSTM network layers to be 3, optimizing the number of first layer neurons, the number of second layer neurons, the number of third layer neurons, the learning rate and batch processing parameters by a Bayesian optimizer, automatically optimizing in the operation of a model, wherein the maximum iteration number is 100, all layers use Relu activation functions, and the optimizer is Adam;
inputting a training set of a Type1 data set on a similar day into the constructed SLSTM for model training;
and 7: constructing a memory long-short-term sequence (MLSTNet) ultra-short-term prediction model, setting parameters of TCANN, LSTM-Skip and MTNet models, setting relevant parameters of model training, and inputting a training set of similar day data sets into the constructed SLSTM model for model training, specifically;
step 7.1: dividing the normalized similar day data set into long-term historical data and short-term historical data according to the time interval between the similar day data set and the prediction day;
and 7.2: inputting the long-term historical data and the short-term historical data into a TCANN neural network, and extracting a time characteristic vector of the photovoltaic power by utilizing a causal convolution layer, an expansion convolution layer and a sparse attention mechanism layer of the TCANN neural network;
step 7.3: respectively sending the output of the TCANN neural network of the step 7.2 into an LSTM neural network model, an LSTM-Skip neural network model and an MTNet model, extracting feature vectors of photovoltaic power dynamic change by using an LSTM layer in the LSTM neural network model and the LSTM-Skip neural network model, and improving the long and short memory extraction capability of data by using the MTNet model;
step 7.4: adding the prediction results of the LSTM neural network, the LSTM-Skip neural network and the MTNet in the step 7.3 to obtain a final prediction result;
Y MLSTNet ＝Y LSTM +Y LSTM-Skip +Y MTNet
in the formula: y is LSTM For the LSTM neural network output results, Y LSTM-Skip For the LSTM-Skip neural network output results, Y MTNet Output results for MTNet neural network, Y MLSTNet The final photovoltaic power ultra-short-term predicted value is obtained;
step 7.5: setting parameters of the model and training the model;
the MLSTNet model network structure consists of a TCANN neural network, an LSTM-Skip neural network and an MTNet neural network;
parameters of the MLSTNet model are optimized by a Bayesian optimizer, automatic optimization is carried out in the running of the model, the maximum iteration number is 100, all layers use Relu activation functions, and the optimizer is Adam;
inputting training sets of Type2, type3 and Type4 data sets of similar days into the constructed MLSTNet for model training;
and 8: respectively inputting the data of the verification set into the SLSTM and MLSTNet photovoltaic power station ultra-short-term power prediction models trained in the step 6 and the step 7, verifying the prediction result, and respectively returning to the step 6 and the step 7 if the error between the prediction result and the true value is larger; if the error with the true value is smaller, performing step 9;
and step 9: respectively inputting the data of the test set into the SLSTM and MLSTNet photovoltaic power station ultra-short-term power prediction models trained in the steps 6 and 7 to perform photovoltaic power ultra-short-term prediction;
step 10: performing inverse normalization processing on the ultra-short term prediction result in the step 9;
evaluating the photovoltaic power ultra-short term prediction result by adopting average absolute error (MAE), normalized average absolute error (NMAE), mean Square Error (MSE), root Mean Square Error (RMSE), normalized Root Mean Square Error (NRMSE) and goodness of fit (R) 2 ) The 6 indexes are used for evaluating the predicted performance of the neural network, and are shown as the following formula:
NMAE＝(MAE/(y max -y min ))×100％
NRMSE＝(RMSE/(y max -y min ))×100％
in the formula: y is i In order to be a true photovoltaic power data,
for the predicted value, m is the number of test sets, y max And y min The maximum and minimum values of the power in the test set are respectively.
The photovoltaic power station ultra-short term power intelligent prediction method based on the SLSTM and the MLSTNet model, provided by the invention, is based on a Spearman correlation analysis method, can determine model input variables, and reduces invalid information of irrelevant data in historical data to improve the model training speed; selecting meteorological data variables with high correlation with photovoltaic power as clustering variables, and selecting 3 statistical characteristics such as average values of the clustering variables as clustering characteristics to comprehensively reflect the characteristics of randomness, intermittence and volatility of each ephemeris data so as to facilitate efficient clustering by a fuzzy C-means clustering algorithm; the SLSTM model integrates a single step rolling model, an LSTM model and a correction model, and compared with a traditional deep learning prediction model, the SLSTM model has more temporal input and higher prediction precision; the MLSTNet model integrates a TCANN model, an LSTM-Skip model and an MTNet model, compared with a traditional deep learning prediction model, the MLSTNet model improves the extraction capability of time characteristics and has higher prediction precision; model parameters are optimized by using the super-frequency Bayes optimizer, the problem of model parameter optimization is solved, and a prediction result with higher quality can be obtained by the method.
Claims (6)
1. A photovoltaic power station ultra-short term power intelligent prediction method based on SLSTM and MLSTNet models is characterized by comprising the following steps:
step 1: acquiring preprocessed meteorological data and historical photovoltaic power data, performing Spearman correlation analysis on the photovoltaic power and the meteorological data, and determining input variables of a prediction model;
and 2, step: selecting a clustering variable, and constructing statistical characteristics of the clustering variable;
and step 3: according to the clustering variables and the statistical characteristics thereof selected in the step 2, clustering photovoltaic power historical data by adopting a fuzzy C-means clustering algorithm to obtain Type1, type2, type3 and Type4 similar day data sets;
and 4, step 4: dividing the data sets of Type1, type2, type3 and Type4 similar days in the step 3 into a training set, a verification set and a test set;
and 5: carrying out normalization processing on the training set, the verification set and the test set in the step 4;
step 6: constructing an SLSTM photovoltaic power station ultra-short-term power prediction model, setting relevant parameters of model training, and inputting data of a training set into the model for training;
and 7: constructing an MLSTNet photovoltaic power station ultra-short-term power prediction model, setting relevant parameters of model training, and inputting data of a training set into the model for training;
and 8: respectively inputting the data of the verification set into the SLSTM and MLSTNet photovoltaic power station ultra-short-term power prediction models trained in the steps 6 and 7, verifying the prediction result, and respectively returning to the steps 6 and 7 if the error between the prediction result and the true value is larger; if the error with the true value is smaller, performing step 9;
and step 9: respectively inputting the data of the test set into the SLSTM and MLSTNet photovoltaic power station ultra-short-term power prediction models trained in the steps 6 and 7 to perform photovoltaic power ultra-short-term prediction;
step 10: and (4) performing inverse normalization processing on the ultra-short term prediction result in the step (9).
2. The photovoltaic power station ultra-short term power intelligent prediction method based on SLSTM and MLSTNet models according to claim 1, wherein in the step 1, specifically:
step 1.1: selecting preprocessed meteorological data and preprocessed photovoltaic power data;
the time resolution of the meteorological variable and the photovoltaic power variable is 15min, and the wind speed, the wind direction, the temperature, the humidity, the air pressure and the irradiation intensity are selected as original meteorological data variables;
step 1.2: measuring the degree of correlation among a plurality of meteorological variables by using a Spearman correlation coefficient R;
step 1.3: and selecting meteorological data variables of which the Spearman correlation coefficient R absolute value of the photovoltaic power is not less than 0.5, and inputting the meteorological data variables as a prediction model.
3. The photovoltaic power station ultra-short term power intelligent prediction method based on SLSTM and MLSTNet models according to claim 1, wherein in the step 2, specifically:
step 2.1: selecting a meteorological data variable of which the Spearman correlation coefficient R absolute value of the photovoltaic power is not less than 0.5 as a clustering variable;
step 2.2: and selecting the average value, the standard deviation and the maximum value of the clustering variables as statistical characteristics.
4. The photovoltaic power station ultra-short term power intelligent prediction method based on SLSTM and MLSTNet models according to claim 1, wherein in the step 3, specifically:
step 3.1: calculating the numerical values of 3 statistical characteristics of the clustering variables on each day according to the statistical characteristics of the clustering variables constructed in the step 2;
step 3.2: determining the number c of data clustering categories, initializing a clustering center matrix V, giving a fuzzy weighting index m, initializing a membership matrix U, and giving an iteration stop standard epsilon of an algorithm, wherein the epsilon is positive and infinitesimal;
step 3.3: calculating all clustering centers of the t iteration according to the formula (5) to obtain a clustering center matrix:
in the formula: u. u ij The membership degree of the ith sample belonging to the jth class; x is the number of i Is a sample point; m is a fuzzy weighting index; t is the number of iterations; c represents the number of clustering categories;
step 3.4: updating the membership matrix, wherein the calculation method is shown as the formula (6):
in the formula: u. u ij The membership degree of the ith sample belonging to the jth class; x is the number of i Is a sample point; m is a fuzzy weighting index; t is the number of iterations; c represents the number of clustering categories; d ij Is the distance from the ith sample to the class j center;
step 3.5: calculate | U (t) -U (t-1) And verifying whether an iteration stop condition | U is satisfied (t) -U (t-1) |<If yes, stopping iteration; if not, repeating the step 3.3 and the step 3.4 until the iteration stop condition is reached, and finally obtaining the Type1, type2, type3 and Type4 similar day data sets.
5. The intelligent photovoltaic power station ultra-short term power prediction method based on SLSTM and MLSTNet models according to claim 1, wherein in the step 6, specifically:
step 6.1: inputting the normalized Type1 similar day data set into an LSTM neural network in a single step rolling input mode to extract a characteristic vector of dynamic change of photovoltaic power;
step 6.2: re-inputting the output feature vector into the LSTM neural network so as to realize single-step rolling prediction;
for example: and taking the meteorological variable and the photovoltaic power at the t-1 moment and the meteorological variable at the t moment as input to obtain a photovoltaic power predicted value at the t moment. Taking the meteorological variable and the photovoltaic power predicted value at the time t and the meteorological variable at the time t +1 as input, and obtaining the photovoltaic power predicted value at the time t +1 until the prediction is finished;
step 6.3: performing linear fitting on the prediction result obtained in the step 6.2 and the actual photovoltaic power to obtain a final prediction result;
Y LSTM ＝β 1 h t +β 0
in the formula: h is t Output of the result, beta, for the LSTM neural network 0 And beta 1 Is a linear fitting coefficient, Y LSTM The final photovoltaic power ultra-short-term predicted value is obtained;
step 6.4: setting parameters of the model and training the model;
the SLSTM model network structure comprises three layers of LSTMs, a full connection layer and a correction model;
setting the number of LSTM network layers to be 3, optimizing the number of first layer neurons, the number of second layer neurons, the number of third layer neurons, the learning rate and batch processing parameters by a Bayesian optimizer, automatically optimizing in the running of a model, wherein the maximum iteration number is 100, all layers use Relu activation functions, and the optimizer is Adam;
the training set of Type1 similar day data sets was imported into the constructed SLSTM for model training.
6. The intelligent ultra-short term power prediction method for the photovoltaic power plant based on the SLSTM and MLSTNet models as claimed in claim 1, wherein in the step 7, specifically:
step 7.1: dividing the normalized Type2, type3 and Type4 similar day data sets into long-term historical data and short-term historical data according to the time interval between the data sets and the prediction day;
and 7.2: inputting the long-term historical data and the short-term historical data into a TCANN neural network, and extracting a time characteristic vector of the photovoltaic power by utilizing a causal convolution layer, an expansion convolution layer and a sparse attention mechanism layer of the TCANN neural network;
step 7.3: respectively sending the output of the TCANN neural network of the step 7.2 into an LSTM neural network model, an LSTM-Skip neural network model and an MTNet model, extracting feature vectors of photovoltaic power dynamic change by using an LSTM layer in the LSTM neural network model and the LSTM-Skip neural network model, and improving the long and short memory extraction capability of data by using the MTNet model;
step 7.4: adding the prediction results of the LSTM neural network, the LSTM-Skip neural network and the MTNet in the step 7.3 to obtain a final prediction result;
Y MLSTNet ＝Y LSTM +Y LSTM-Skip +Y MTNet
in the formula: y is LSTM For the LSTM neural network output results, Y LSTM-Skip For the LSTM-Skip neural network output results, Y MTNet Output results for MTNet neural network, Y MLSTNet The final photovoltaic power ultra-short-term predicted value is obtained;
step 7.5: setting parameters of the model and training the model;
the MLSTNet model network structure consists of a TCANN neural network, an LSTM-Skip neural network and an MTNet neural network;
parameters of the MLSTNet model are optimized by a Bayesian optimizer, automatic optimization is carried out in the running of the model, the maximum iteration number is 100, all layers use Relu activation functions, and the optimizer is Adam;
inputting training sets of Type2, type3 and Type4 similar day data sets into the constructed MLSTNet for model training.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211375921.7A CN115796004A (en) | 2022-11-04 | 2022-11-04 | Photovoltaic power station ultra-short term power intelligent prediction method based on SLSTM and MLSTNet models |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211375921.7A CN115796004A (en) | 2022-11-04 | 2022-11-04 | Photovoltaic power station ultra-short term power intelligent prediction method based on SLSTM and MLSTNet models |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115796004A true CN115796004A (en) | 2023-03-14 |
Family
ID=85435475
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211375921.7A Pending CN115796004A (en) | 2022-11-04 | 2022-11-04 | Photovoltaic power station ultra-short term power intelligent prediction method based on SLSTM and MLSTNet models |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115796004A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116227743A (en) * | 2023-05-06 | 2023-06-06 | 中国华能集团清洁能源技术研究院有限公司 | Photovoltaic power generation power abnormal rate determining method and system based on tuna swarm algorithm |
| CN117272851A (en) * | 2023-11-23 | 2023-12-22 | 太原理工大学 | Modeling prediction method for plant light receiving quantity under saline-alkali soil photovoltaic panel |
| CN117709556A (en) * | 2024-02-06 | 2024-03-15 | 山东建筑大学 | Photovoltaic power generation short-term prediction method, system, medium and equipment |
| CN117973639A (en) * | 2024-03-29 | 2024-05-03 | 南京师范大学 | Photovoltaic power prediction method based on multiscale similarity days and improved integration algorithm |
| CN118194918A (en) * | 2024-05-17 | 2024-06-14 | 华能江苏综合能源服务有限公司 | Construction method and system of LSTM model for photovoltaic power generation power prediction |
| CN118364865A (en) * | 2024-06-14 | 2024-07-19 | 云南省交通规划设计研究院股份有限公司 | Method for predicting visibility of highway traffic meteorological environment in minute level and early warning system |
| CN118536407A (en) * | 2024-07-24 | 2024-08-23 | 哈尔滨工业大学（威海） | Sea surface wind speed prediction algorithm and system based on symbol dynamic analysis |
-
2022
- 2022-11-04 CN CN202211375921.7A patent/CN115796004A/en active Pending
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116227743A (en) * | 2023-05-06 | 2023-06-06 | 中国华能集团清洁能源技术研究院有限公司 | Photovoltaic power generation power abnormal rate determining method and system based on tuna swarm algorithm |
| CN116227743B (en) * | 2023-05-06 | 2023-09-01 | 中国华能集团清洁能源技术研究院有限公司 | Photovoltaic power generation power abnormal rate determining method and system based on tuna swarm algorithm |
| CN117272851A (en) * | 2023-11-23 | 2023-12-22 | 太原理工大学 | Modeling prediction method for plant light receiving quantity under saline-alkali soil photovoltaic panel |
| CN117272851B (en) * | 2023-11-23 | 2024-02-02 | 太原理工大学 | Modeling prediction method for plant light receiving quantity under saline-alkali soil photovoltaic panel |
| CN117709556A (en) * | 2024-02-06 | 2024-03-15 | 山东建筑大学 | Photovoltaic power generation short-term prediction method, system, medium and equipment |
| CN117709556B (en) * | 2024-02-06 | 2024-05-31 | 山东建筑大学 | Photovoltaic power generation short-term prediction method, system, medium and equipment |
| CN117973639A (en) * | 2024-03-29 | 2024-05-03 | 南京师范大学 | Photovoltaic power prediction method based on multiscale similarity days and improved integration algorithm |
| CN117973639B (en) * | 2024-03-29 | 2024-06-14 | 南京师范大学 | Photovoltaic power prediction method based on multiscale similarity days and improved integration algorithm |
| CN118194918A (en) * | 2024-05-17 | 2024-06-14 | 华能江苏综合能源服务有限公司 | Construction method and system of LSTM model for photovoltaic power generation power prediction |
| CN118364865A (en) * | 2024-06-14 | 2024-07-19 | 云南省交通规划设计研究院股份有限公司 | Method for predicting visibility of highway traffic meteorological environment in minute level and early warning system |
| CN118536407A (en) * | 2024-07-24 | 2024-08-23 | 哈尔滨工业大学（威海） | Sea surface wind speed prediction algorithm and system based on symbol dynamic analysis |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Li et al. | Photovoltaic power forecasting with a hybrid deep learning approach | |
| CN115796004A (en) | Photovoltaic power station ultra-short term power intelligent prediction method based on SLSTM and MLSTNet models | |
| CN113962364B (en) | Multi-factor power load prediction method based on deep learning | |
| CN109002915B (en) | Photovoltaic power station short-term power prediction method based on Kmeans-GRA-Elman model | |
| CN113128793A (en) | Photovoltaic power combination prediction method and system based on multi-source data fusion | |
| CN109284870A (en) | Short-term method for forecasting photovoltaic power generation quantity based on shot and long term Memory Neural Networks | |
| CN108921339B (en) | Quantile regression-based photovoltaic power interval prediction method for genetic support vector machine | |
| CN111222674A (en) | Short-term photovoltaic power generation capacity prediction method based on long-term and short-term memory neural network | |
| CN111695736B (en) | Photovoltaic power generation short-term power prediction method based on multi-model fusion | |
| CN114792156B (en) | Photovoltaic output power prediction method and system based on curve characteristic index clustering | |
| CN115115125B (en) | Photovoltaic power interval probability prediction method based on deep learning fusion model | |
| CN113537582B (en) | Photovoltaic power ultra-short-term prediction method based on short-wave radiation correction | |
| CN115860177A (en) | Photovoltaic power generation power prediction method based on combined machine learning model and application thereof | |
| CN113516271A (en) | Wind power cluster power day-ahead prediction method based on space-time neural network | |
| CN106446440A (en) | Short-term photovoltaic generation power prediction method based on online sequential extreme learning machine | |
| CN116341613A (en) | Ultra-short-term photovoltaic power prediction method based on Informar encoder and LSTM | |
| CN114511132A (en) | Photovoltaic output short-term prediction method and prediction system | |
| CN105426989A (en) | EEMD and combined kernel RVM-based photovoltaic power short-term prediction method | |
| CN113469427A (en) | Convolution LSTM-based photovoltaic power station day-ahead power generation amount prediction method and system | |
| CN111242355A (en) | Photovoltaic probability prediction method and system based on Bayesian neural network | |
| CN115907131B (en) | Method and system for constructing electric heating load prediction model in northern area | |
| CN116629416A (en) | Photovoltaic power station power prediction method and device | |
| CN116014722A (en) | Sub-solar photovoltaic power generation prediction method and system based on seasonal decomposition and convolution network | |
| CN116826737A (en) | Photovoltaic power prediction method, device, storage medium and equipment | |
| CN116502074A (en) | Model fusion-based photovoltaic power generation power prediction method and system |
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 |