CN117874584A - Neural network ultra-short term photovoltaic power prediction method based on meteorological clustering - Google Patents
Neural network ultra-short term photovoltaic power prediction method based on meteorological clustering Download PDFInfo
- Publication number
- CN117874584A CN117874584A CN202310151448.2A CN202310151448A CN117874584A CN 117874584 A CN117874584 A CN 117874584A CN 202310151448 A CN202310151448 A CN 202310151448A CN 117874584 A CN117874584 A CN 117874584A
- Authority
- CN
- China
- Prior art keywords
- clustering
- data
- photovoltaic power
- neural network
- meteorological
- 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
- 238000000034 method Methods 0.000 title claims abstract description 43
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 33
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 31
- 238000012216 screening Methods 0.000 claims abstract description 31
- 238000007637 random forest analysis Methods 0.000 claims abstract description 22
- 238000012549 training Methods 0.000 claims abstract description 6
- 230000015654 memory Effects 0.000 claims description 25
- 239000003016 pheromone Substances 0.000 claims description 24
- 230000006870 function Effects 0.000 claims description 23
- 230000035772 mutation Effects 0.000 claims description 18
- 238000003066 decision tree Methods 0.000 claims description 13
- 239000013598 vector Substances 0.000 claims description 11
- 241000257303 Hymenoptera Species 0.000 claims description 8
- 238000004458 analytical method Methods 0.000 claims description 6
- 230000000295 complement effect Effects 0.000 claims description 6
- 238000011156 evaluation Methods 0.000 claims description 6
- 238000005259 measurement Methods 0.000 claims description 5
- 238000001704 evaporation Methods 0.000 claims description 4
- 230000008020 evaporation Effects 0.000 claims description 4
- 239000011159 matrix material Substances 0.000 claims description 4
- 230000010354 integration Effects 0.000 claims description 3
- 238000001914 filtration Methods 0.000 claims description 2
- 210000004027 cell Anatomy 0.000 description 8
- 230000008569 process Effects 0.000 description 8
- 238000010248 power generation Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 210000002569 neuron Anatomy 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000000354 decomposition reaction Methods 0.000 description 3
- 230000009467 reduction Effects 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 230000007774 longterm Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 230000006403 short-term memory Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 230000019637 foraging behavior Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000012535 impurity Substances 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000012847 principal component analysis method Methods 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 238000012797 qualification Methods 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 230000003252 repetitive effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 230000001932 seasonal effect Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
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 relates to a neural network ultra-short term photovoltaic power prediction method based on meteorological clustering, which comprises the following steps: s1, constructing a secondary variable; s2, normalizing measured data and predicted data in the secondary variables, and correcting irradiation data; s3, screening the normalized measured data, and clustering based on an improved ant colony clustering algorithm to obtain three kinds of variation clustering results of sunny days, cloudy days and rainy days; s4, screening three kinds of variation clustering results based on random forest feature degrees to obtain a feature variable group; s5, training the LSTM neural network, and inputting the data in the S2 into an LSTM model to obtain a photovoltaic power prediction result; and S6, evaluating the photovoltaic power prediction result, and optimizing the photovoltaic power prediction result. Compared with the prior art, the method has the advantages of improving the accuracy of predicting the fluctuation weather with larger impact on the power grid system and the like.
Description
Technical Field
The invention relates to the field of photovoltaic power prediction, in particular to a neural network ultra-short term photovoltaic power prediction method based on meteorological clustering.
Background
Due to the influence of solar radiation intensity and seasonal climate, photovoltaic power generation has the characteristics of strong intermittence, volatility and randomness, and the defect causes that the difficulty of power generation balance is increased due to large-scale access of the photovoltaic to a power grid, and the operation uncertainty of a power system is obviously increased. How to ensure safe and stable and economic operation of a power grid while improving the utilization rate of renewable energy sources is a hot problem in research of the field of new energy sources. Photovoltaic power generation power predictions are generally classified by time scale, and are largely classified into mid-long term, short term and ultra-short term. The long-term prediction is mainly used for predicting power over seven days in the future based on numerical weather forecast (NWP), and is mainly used for power station planning and design, long-term scheduling of a power grid and the like. The short-term and ultra-short-term prediction is mainly power prediction for three days and four hours in the future, and is mainly used for real-time scheduling of a power grid, controlling the running and filling of a generator set and enhancing the running stability of the power grid. CN114626622a discloses a photovoltaic power prediction method, system, device and storage medium, which obtain surface irradiance by obtaining weather prediction data and inputting the weather prediction data into a preset surface irradiance prediction model; calculating to obtain the irradiance of the inclined plane based on the irradiance of the earth surface; the predicted photovoltaic power is obtained based on inclined plane irradiance calculation, the problem of difficult data source is solved, and the precision of the photovoltaic predicted power is improved. The method is beneficial to improving the operation efficiency of the comprehensive energy system in the park under the background of the novel power system, perfecting the energy optimization scheduling management system of the park and improving the energy utilization rate. However, the characteristic screening of the method only adopts the inclined plane irradiance after the surface irradiance conversion, and the influence factors of the photovoltaic power generation power are complex, such as temperature, humidity and the like, can influence the photovoltaic power generation output. The lack of other weather conditions results in a single model and low accuracy of the predicted results. CN110516844a discloses a multivariate input photovoltaic power prediction method based on EMD-PCA-LSTM, decomposing 5 environmental sequences by an empirical mode decomposition method to obtain eigenmode decomposition and residual components at different time scales, and decomposing the environmental sequences into various different fluctuation sequences; and (3) screening out key factors influencing the photovoltaic output power by using a principal component analysis method, reducing the dimension of the input parameters of the model, and eliminating the redundancy and correlation of different fluctuation sequences obtained by EMD decomposition. And finally, modeling the dynamic time between the multivariable time sequence and the photovoltaic power sequence through the LSTM neural network, constructing a prediction model, and realizing the prediction of the photovoltaic output power. But this approach lacks feature screening for multi-variable inputs and the variables analyzed are mainly one-dimensional variables, lacking the data construction of this variable. Meanwhile, the adopted prediction model is a single integral unclassified model, and the power fluctuation and meteorological conditions of different meteorological states have large differences, so that when all data are input into the model for prediction, the weather with strong fluctuation cannot be predicted accurately.
In summary, the accuracy of the current machine learning-based prediction of the generated power needs to be improved, the current machine learning-based prediction is influenced by the change of weather conditions, the ultra-short-term fluctuation characteristics of irradiance and power data are various, each prediction algorithm also has own limitations, and the prediction algorithm aiming at different types of weather is rarely available.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, and provides the ultra-short-term photovoltaic power prediction method of the neural network based on meteorological clustering, which is used for accurately predicting photovoltaic power.
The aim of the invention can be achieved by the following technical scheme:
a neural network ultra-short term photovoltaic power prediction method based on meteorological clustering comprises the following steps:
s1, acquiring field actual measurement data and prediction data of different types, multiplying the data of different types by each other, and constructing a secondary variable;
s2, normalizing measured data and predicted data in the secondary variable, and correcting irradiation data in the predicted data in the secondary variable;
s3, screening the normalized measured data, and clustering the screened measured data based on an improved ant colony clustering algorithm to obtain three kinds of mutation clustering results of sunny days, cloudy days and rainy days;
s4, screening three types of mutation clustering results based on random forest feature degrees, and screening feature variable groups from each type of mutation clustering results;
s5, inputting the characteristic variable group into an LSTM neural network for training to obtain a trained LSTM model, and inputting the normalized measured data and the corrected predicted data obtained in the S2 into the LSTM model to obtain a photovoltaic power predicted result;
and S6, evaluating the photovoltaic power prediction result to obtain an evaluation result, and optimizing the photovoltaic power prediction result based on the evaluation result.
Further, the parameters of the improved ant colony algorithm comprise pheromone tau, and the update formula of the pheromone tau is as follows:
wherein m is the number of ants, ρ is the evaporation rate of pheromone, and Δτ a Is the amount of pheromone released by the nth ant on the edge where it passes, R a Representing a path memory vector, C k Data representing the path length, i and j represent the ith row and jth column of the matrix of pheromones.
Further, in the improved ant colony clustering algorithm, for clustering results of any one of sunny days, cloudy days and rainy days obtained by clustering, clustering results with a preset proportion are selected from all the clustering results, and the clustering results are randomly distributed into clustering centers of other types to obtain variant clustering results, wherein the preset proportion is 10% -20%.
Further, screening is performed based on random forest feature degree, and the feature variable group screening in each type of mutation clustering result is specifically as follows:
the method comprises the steps of calculating the base index of each feature in each type of mutation clustering result, calculating the base index variation before and after node branching of any decision tree in a random forest based on the base index, obtaining importance scores of the feature in the decision tree in the random forest, integrating the importance scores of all the features in all the decision trees, and screening out a feature variable group based on the importance scores obtained by integration.
Further, the expression of the keni index is:
wherein Gini is a keni index, K represents the kth category, K 'represents the kth' category, K represents the total number of categories, p k Representing the proportion of class k, p, in each node k′ Representing the proportion of category k' in each node.
Further, the expression of the update formula for long memories of LSTM neural network is:
wherein C is t F for updated long memory t As a forgetting function, C t-1 I is a long memory of input t For the status updating function,for the state update vector, t represents time t.
Further, in S2, the irradiation data in the prediction data is corrected, where the irradiation data is irradiance of the ground in vertical projection, and the corrected expression is:
wherein alpha is the inclination angle of the battery plate, beta is the complementary angle of the included angle between sunlight and the ground, R NWP To predict irradiance data in the data, R P Is the irradiance data corrected based on the predicted data.
Further, the expression of the complementary angle of the included angle between the sunlight and the ground is:
β＝α-δ
wherein δ is the latitude of the direct point, which is different according to different time periods.
Further, in S3, the filtering of the normalized measured data specifically includes: and screening the normalized measured data by combining the pearson correlation coefficient and the weather classification standard, and screening out an equivalent analysis variable.
Further, the measured data and the predicted data include one or more of horizontal irradiation intensity, temperature, humidity, customs, wind direction, and air pressure.
Compared with the prior art, the invention has the following beneficial effects:
(1) And clustering the large-scale data samples in different weather categories according to a variation ant colony clustering algorithm. The method has the advantages that three different types of weather including sunny days, cloudy days and rainy days are subjected to feature importance analysis and feature screening by adopting a random forest algorithm, the accuracy of photovoltaic prediction is greatly improved through the flow of clustering and feature screening, and meanwhile, the accuracy of predicting the fluctuation weather with larger impact on a power grid system is improved.
(2) For different kinds of field actual measurement data and prediction data, the different kinds of data are multiplied by each other to construct a secondary variable, and compared with a one-dimensional variable in the prior art, the weather with strong fluctuation can be predicted more accurately.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a schematic diagram of the irradiance data modification of the present invention;
FIG. 3 is a diagram of classification effect of sunny days of the present invention;
FIG. 4 is a graph showing classification effect of cloudy days according to the present invention;
FIG. 5 is a diagram showing the classification effect of the rainy days of the present invention;
FIG. 6 is a block diagram of an LSTM neural network of the present invention;
FIG. 7 is a diagram of a dataset construction of an LSTM model of the present invention;
fig. 8 is a graph of the predicted outcome of the present invention.
Detailed Description
The invention will now be described in detail with reference to the drawings and specific examples. The present embodiment is implemented on the premise of the technical scheme of the present invention, and a detailed implementation manner and a specific operation process are given, but the protection scope of the present invention is not limited to the following examples.
Noun interpretation:
NWP (Numerical Weather Prediction) -numerical weather forecast
According to the actual condition of the atmosphere, under a certain initial condition, a numerical calculation is carried out through a large-scale computer, and the weather evolution describing process is solved.
RNN (Recurrent Neural Network) -circulating neural network
The recurrent neural network takes sequence data as input, performs recursion in the evolution direction of the sequence and connects all nodes in a chained mode.
LSTM (Long Short-Term Memory) -Long and Short-Term Memory network
A neural network for deep learning.
ACO (Ant Colony Optimization) -ant colony algorithm
And adopting ant cavity cleaning behaviors of the ant colony and ant colony foraging behaviors to carry out a clustering algorithm simulation process.
RF (Random Forests) random forest algorithm
A classifier is provided for training and predicting samples using a plurality of trees.
Example 1:
the invention provides a neural network ultra-short term photovoltaic power prediction method based on meteorological clustering, and the flow of the method is shown in figure 1. The method comprises the following steps:
s1, acquiring field actual measurement data and prediction data of different types, multiplying the data of different types by each other, and constructing a secondary variable;
s2, normalizing measured data and predicted data in the secondary variable, and correcting irradiation data in the predicted data in the secondary variable;
s3, screening the normalized measured data, and clustering the screened measured data based on an improved ant colony clustering algorithm to obtain three kinds of mutation clustering results of sunny days, cloudy days and rainy days;
s4, screening three types of mutation clustering results based on random forest feature degrees, and screening feature variable groups from each type of mutation clustering results;
s5, inputting the characteristic variable group into an LSTM neural network for training to obtain a trained LSTM model, and inputting the normalized measured data and the corrected predicted data obtained in the S2 into the LSTM model to obtain a photovoltaic power predicted result;
and S6, evaluating the photovoltaic power prediction result to obtain an evaluation result, and optimizing the photovoltaic power prediction result based on the evaluation result.
In the S1, the predicted data is numerical weather forecast, namely NWP data, which means that according to the actual condition of the atmosphere, numerical calculation is carried out by a large-scale computer under a certain initial condition, and the data describing the weather evolution process is solved. The measured data and the predicted data include one or more of horizontal irradiation intensity, temperature, humidity, customs, wind direction, and air pressure.
In S2, due to different data types, the difference between data dimensions may affect the weight value of the output power, and the invention adopts the minimum-maximum min-max method to normalize the data, so that the process not only can keep the variation trend of the data, but also can reduce the operation complexity of the program. The normalized formula is:
wherein x is i Representing raw data representing input characteristic variables or output power;is the data after normalization.
The irradiation data in the NWP data is irradiance of the vertical projection ground of the city where the power plant is located, and the irradiance accepted by the section of the photovoltaic cell panel affects the generated power, so that the irradiance of the NWP needs to be corrected. In order to improve the illumination utilization rate, the fixed photovoltaic cell panel placement mode adopted by the power plant is that the cell panel is perpendicular to the noon solar illumination direction in spring, winter and winter, so that the corrected expression is:
wherein alpha is the inclination angle of the battery plate, beta is the complementary angle of the included angle between sunlight and the ground, R NWP To predict irradiance data in the data, R P Is the irradiance data corrected based on the predicted data. The modified schematic is shown in fig. 2.
The expression of the complementary angle of the included angle between sunlight and the ground is as follows:
β＝α-δ
wherein δ is the latitude of the direct point, which is different according to different time periods. The sun in spring and autumn of each year is directly projected to the earth equator, the sun in summer and winter is respectively directly projected to the coming back line from north and coming back line from south, and the spring diary is used as the 0 th day, so that the latitude of the direct point can be calculated by using the formula in the table 1.
TABLE 1 direct point latitude Meter
And S3, adopting an improved ant colony clustering algorithm to perform clustering. In nature, the ant population relies on a substance called Pheromone (pheomone) to realize indirect communication of ant colony in the process of finding food, both the cooperation between ants and the interaction between ants and the environment, so that the shortest path from the ant cavity to the food source is found through cooperation.
The ant colony algorithm mainly depends on pheromone parameters as classification references, and the ant colony algorithm with known cluster number is adopted in the invention, and the thought is that the ant colony algorithm is firstly selected and then mutated.
Firstly initializing a pheromone concentration matrix, enabling a plurality of ants to traverse data points for clustering, and enabling the ants to release pheromones on paths and release pheromones related to path lengths. And selecting the result with the optimal objective function value in one iteration as the condition of the next iteration, wherein the pheromone concentration is inversely proportional to the path length. When the following ants hit the intersection again, a path with higher pheromone concentration is selected. The pheromone concentration on the optimal path is larger and larger, and finally the ant colony finds the optimal feeding path.
The update formula of the pheromone tau is:
wherein m is the number of ants, ρ is the evaporation rate of pheromone, and Δτ a Is the amount of pheromone released by the nth ant on the edge where it passes, R a Representing a path memory vector, C k Representing the path length, i and j are the data (i.e., pheromones) that construct the ith row and jth column of the pheromone matrix. The path length is the sum of the lengths of all edges in the path memory vector. The evaporation rate ρ of the pheromone satisfies 0 < ρ < 1.
The ant colony algorithm is a self-organizing heuristic algorithm, is insensitive to noise and has good result quality by simulating a natural ant path planning principle. In the invention, a mutation step is added in an ant colony algorithm, for any one of the clustering results of sunny days, cloudy days and rainy days, a clustering result with a preset proportion is selected from all the clustering results, and the clustering results are randomly distributed into the clustering centers of other types, wherein the preset proportion is 10% -20%, so as to obtain a mutation clustering result. And retaining the variation with better cluster performance. The randomness is introduced through the mutation process, so that the algorithm can be effectively prevented from falling into local optimum, and the clustering quality of the algorithm is improved. And 3, classifying effect graphs of the clustered sunny days, cloudy days and rainy days are shown in figures 3, 4 and 5.
And S3, before an ant colony algorithm is performed, screening the normalized measured data, and screening the normalized measured data by combining the Pearson correlation coefficient and the weather classification standard to obtain an equivalent analysis variable, wherein the screened equivalent analysis variable is one type of data in a data group with higher correlation, so that the data dimension is reduced, the operation speed is greatly improved, and the operation efficiency is improved.
In S4, the idea of evaluating the importance of the features by using the random forest is based on the contribution degree of each feature in each decision tree in the random forest, and the average value is taken and compared. There are two common calculation methods, one is the reduction of average non-purity (mean decrease impurity), the measurement is carried out by commonly used coefficient of kunning, and the method is adopted in the model adopted by the invention; another is a reduction in average accuracy (mean decrease accuracy), often measured by the out-of-bag error rate. The invention adopts a method for evaluating the base index. Screening is carried out based on random forest feature degree, and the characteristic variable group screening in each type of mutation clustering result is specifically as follows:
the method comprises the steps of calculating the base index of each feature in each type of mutation clustering result, calculating the base index variation before and after node branching of any decision tree in a random forest based on the base index, obtaining importance scores of the feature in the decision tree in the random forest, integrating the importance scores of all the features in all the decision trees, and screening out a feature variable group based on the importance scores obtained by integration. The method comprises the following specific steps:
variable importance scores (variable importance measures) are expressed as VIM, assuming m features X1, X2, X3, xm, gini index score VIM for each feature is calculated.
The expression of the Gini (Gini) index is:
wherein Gini is a keni index, K represents the kth category, K 'represents the kth' category, K represents the total number of categories, p k Representing the proportion of class k, p, in each node k′ Representing the proportion of category k' in each node.
The importance of feature Xj at node m can be expressed as a reduction in weighted non-purity, i.e., the Gini index change before and after branching at node m is:
VIM＝Gini-Gini′-Gini″
wherein Gini' and Gini "represent Gini indices of two new nodes after branching, respectively.
If the node of the feature that appears in decision tree n is in set M, then the importance of the feature in the nth tree is:
the importance algorithm of the single feature on the single decision tree is generalized to N decision trees as follows:
and finally, normalizing the importance scores of all the features.
And evaluating the importance of each feature through a random forest algorithm, and eliminating the influence of the repetitive features. The contribution of each feature to the accuracy of the predictive model is derived. In some embodiments, the characteristic variable groups of the first five names in each classification are screened through classification conditions after an ant colony clustering algorithm and substituted into a subsequent LSTM model to predict.
The LSTM network in S5, i.e., the long-short term memory neural network, is one of the cyclic neural networks (RNNs), is a special improvement based on the gradient disappearance or the mention of explosion problem of the RNN network, and can make full use of the historical data information. The photovoltaic power generation power and time have obvious dependency relationship, and the focus of the photovoltaic ultra-short term prediction is to explore time sequences, so that the characteristics of a pre-use period and a period are used, and the prediction precision is improved. The structure of the LSTM network is shown in fig. 6.
Between neurons, from time t to time t+1 LSTM is transmitted a long memory Ct and a short memory ht, and the input Xt and the short memory ht in this period are combined to be used as input parameters. Inside neurons, parameters are calculated mainly by means of several gating functions.
The forget gate decides which information should be discarded or retained, and screens through the output value of the sigmoid function (between 0 and 1).
The expression of the forgetting function is:
f t ＝σ(W f ·[h t-1 ，x t ]+b f )
wherein:
sigma-Sigmoid function:
W f -a weight term;
b f -a bias term.
Entry of neuron [ h ] t-1 ，x t ]After the combination operation of the weight item and the bias item, the data is multiplied by the sigmoid function, when the data input with higher importance degree is larger, the output value of the sigmoid function is close to 1, so that the data is reserved, otherwise, when the input is smaller, the output value is close to 0, and the part of information is forgotten.
After the information is initially processed, the input gate updates the cell state by selecting information of newly added neurons:
wherein:
W i -inputting a weight term of the sigmoid function;
b i -inputting a bias term of the sigmoid function;
W C -inputting a weight term of the tanh function;
b C -inputting a bias term of the tanh function;
the input information enters a sigmoid function and a tanh function, the principle formula is shown in the formula, the sigmoid gating function selects which information in the input needs to be updated, the tanh function generates a new candidate value vector for updated content, and the two candidate value vectors are multiplied, so that the output value of the sigmoid determines which information in the output value of the tanh is important and needs to be kept.
The cell state of the previous layer is multiplied by the forgetting vector point by point, and then the value is added with the output value of the input gate point by point, and the process is to update the long memory and also update the cell state:
wherein C is t F for updated long memory t As a forgetting function, C t-1 I is a long memory of input t For the status updating function,for the state update vector, t represents time t.
After updating the cell state, short memory updating is performed in the output gate. Short memories are required to be transferred between neurons and to be propagated to the next layer. In the output layer, a sigmoid function is used to multiply the input parameters, select the parameters that need to be output, and a tanh function is multiplied with the parameters generated by the long memories to generate vectors that can be generated to be output as short memories. And finally, multiplying the two to determine the information carried by the short memory.
o t ＝σ(W o ·[h t-1 ，x t ]+b o )
h t ＝o t ⊙tanh(C t )
The short memories are then taken as the output of the current cell, and the new long memories and the new short memories are passed on to the next time step.
The time series must be converted into samples with input and output components. The method is divided into predicted input data X and predicted output labels y. For the univariate time series problem, observations at a previous point in time may be used as input, and observations at a current point in time as output.
The invention relates to a multivariable time sequence problem, which adopts the characteristics of power output and NWP meteorological data structure of the previous point time and the characteristics of NWP meteorological data structure of the current point as input tags, and the current power output is an output tag. Therefore, based on the characteristics of low photovoltaic ultra-short-term prediction resolution and short prediction time, an LSTM data set model is constructed. The dataset composition of the LSTM model is shown in fig. 7.
NWP data is obtained in advance, is more suitable for prediction, but has lower accuracy. The local measured data is obtained in real time, and the measured data at the time t-1 can be used for predicting the power output at the time t. Through the LSTM neural network, the characteristic quantity screened by the random forest algorithm is selected as the input of the neural network, and the power value is selected as the output to carry out network training.
In S6, in order to measure the accuracy of the prediction method, the national standard adopts Root Mean Square Error (RMSE), mean Absolute Error (MAE) and month Qualification Rate (QR) to evaluate the prediction result.
P i -i moment actual power value;
P i i, predicting the power value at the moment;
C i -i moment photovoltaic panel installation capacity;
n is the number of samples;
the value range of the RMSE and the MAE is [0, + ] and is equal to 0 when the predicted value is completely consistent with the true value, namely a perfect model; the larger the error, the larger the value. The value range [0,1] of the QR is equal to 1 when the predicted value is completely matched with the true value, namely a perfect model; the larger the error, the smaller the value.
The invention can optimize the predicted result through evaluating the predicted result.
The results of the prediction in S5 are shown in table 2 and fig. 8.
Table 2 prediction results table
As can be seen from Table 2 and FIG. 8, the RF algorithm has improved prediction accuracy over the correlation coefficient algorithm, and the predicted data of ACO-RF-LSTM is more accurate than the predicted data without ACO clustering. After ACO clustering, the learning degree of weather volatility of the RF-LSTM prediction model is greatly improved, and the prediction of the volatility weather is further improved.
The foregoing describes in detail preferred embodiments of the present invention. It should be understood that numerous modifications and variations can be made in accordance with the concepts of the invention by one of ordinary skill in the art without undue burden. Therefore, all technical solutions which can be obtained by logic analysis, reasoning or limited experiments based on the prior art by the person skilled in the art according to the inventive concept shall be within the scope of protection defined by the claims.
Claims (10)
1. A neural network ultra-short term photovoltaic power prediction method based on meteorological clustering is characterized by comprising the following steps:
s1, acquiring field actual measurement data and prediction data of different types, multiplying the data of different types by each other, and constructing a secondary variable;
s2, normalizing measured data and predicted data in the secondary variable, and correcting irradiation data in the predicted data in the secondary variable;
s3, screening the normalized measured data, and clustering the screened measured data based on an improved ant colony clustering algorithm to obtain three kinds of mutation clustering results of sunny days, cloudy days and rainy days;
s4, screening three types of mutation clustering results based on random forest feature degrees, and screening feature variable groups from each type of mutation clustering results;
s5, inputting the characteristic variable group into an LSTM neural network for training to obtain a trained LSTM model, and inputting the normalized measured data and the corrected predicted data obtained in the S2 into the LSTM model to obtain a photovoltaic power predicted result;
and S6, evaluating the photovoltaic power prediction result to obtain an evaluation result, and optimizing the photovoltaic power prediction result based on the evaluation result.
2. The method for predicting ultra-short-term photovoltaic power of a neural network based on meteorological clustering according to claim 1, wherein the parameters of the improved ant colony algorithm comprise pheromone tau, and an update formula of the pheromone tau is as follows:
wherein m is the number of ants, ρ is the evaporation rate of pheromone, and Δτ a Is the amount of pheromone released by the nth ant on the edge where it passes, R a Representing a path memory vector, C k Data representing the path length, i and j represent the ith row and jth column of the matrix of pheromones.
3. The neural network ultra-short term photovoltaic power prediction method based on meteorological clustering according to claim 1, wherein in an improved ant colony clustering algorithm, for clustering results of any one of sunny days, cloudy days and rainy days obtained by clustering, clustering results with a pre-configured proportion are selected from all the clustering results, and the clustering results are randomly distributed in the clustering centers of other types to obtain variant clustering results, wherein the pre-configured proportion is 10% -20%.
4. The method for predicting ultra-short-term photovoltaic power of a neural network based on meteorological clustering according to claim 1, wherein the screening is performed based on random forest feature degree, and the screening of feature variable groups in each type of mutation clustering result is specifically as follows:
the method comprises the steps of calculating the base index of each feature in each type of mutation clustering result, calculating the base index variation before and after node branching of any decision tree in a random forest based on the base index, obtaining importance scores of the feature in the decision tree in the random forest, integrating the importance scores of all the features in all the decision trees, and screening out a feature variable group based on the importance scores obtained by integration.
5. The method for predicting ultra-short term photovoltaic power of a neural network based on meteorological clustering according to claim 4, wherein the expression of the keni index is:
wherein Gini is a keni index, K represents the kth category, K 'represents the kth' category, K represents the total number of categories, p k Representing the proportion of class k, p, in each node k′ Representing the proportion of category k' in each node.
6. The method for predicting ultra-short term photovoltaic power of a neural network based on meteorological clustering according to claim 1, wherein the expression of the update formula of the LSTM neural network for long memory is:
wherein C is t F for updated long memory t As a forgetting function, C t-1 I is a long memory of input t For the status updating function,for the state update vector(s),t represents time t.
7. The method for predicting ultra-short-term photovoltaic power of a neural network based on meteorological clustering according to claim 1, wherein in S2, irradiation data in the predicted data is corrected, the irradiation data is irradiance of a vertical projection ground, and the corrected expression is:
wherein alpha is the inclination angle of the battery plate, beta is the complementary angle of the included angle between sunlight and the ground, R NWP To predict irradiance data in the data, R P Is the irradiance data corrected based on the predicted data.
8. The method for predicting ultra-short-term photovoltaic power of a neural network based on meteorological clustering according to claim 7, wherein the expression of the complementary angle of the included angle between sunlight and the ground is:
β＝α-δ
wherein δ is the latitude of the direct point, which is different according to different time periods.
9. The method for predicting ultra-short-term photovoltaic power of a neural network based on meteorological clustering according to claim 1, wherein in S3, the filtering of the normalized measured data is specifically: and screening the normalized measured data by combining the pearson correlation coefficient and the weather classification standard, and screening out an equivalent analysis variable.
10. The method for predicting ultra-short term photovoltaic power of a neural network based on meteorological clustering according to claim 1, wherein the measured data and the predicted data comprise one or more of horizontal irradiation intensity, temperature, humidity, customs, wind direction and air pressure.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310151448.2A CN117874584A (en) | 2023-02-22 | 2023-02-22 | Neural network ultra-short term photovoltaic power prediction method based on meteorological clustering |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310151448.2A CN117874584A (en) | 2023-02-22 | 2023-02-22 | Neural network ultra-short term photovoltaic power prediction method based on meteorological clustering |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117874584A true CN117874584A (en) | 2024-04-12 |
Family
ID=90587130
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310151448.2A Pending CN117874584A (en) | 2023-02-22 | 2023-02-22 | Neural network ultra-short term photovoltaic power prediction method based on meteorological clustering |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117874584A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118199061A (en) * | 2024-05-17 | 2024-06-14 | 宁波送变电建设有限公司甬城配电网建设分公司 | Short-term power prediction method and system for renewable energy sources |
-
2023
- 2023-02-22 CN CN202310151448.2A patent/CN117874584A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118199061A (en) * | 2024-05-17 | 2024-06-14 | 宁波送变电建设有限公司甬城配电网建设分公司 | Short-term power prediction method and system for renewable energy sources |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109711620B (en) | Short-term power load prediction method based on GRU neural network and transfer learning | |
| CN110942205B (en) | Short-term photovoltaic power generation power prediction method based on HIMVO-SVM | |
| CN113282122B (en) | Commercial building energy consumption prediction optimization method and system | |
| CN111260126B (en) | Short-term photovoltaic power generation prediction method considering correlation degree of weather and meteorological factors | |
| CN114792156B (en) | Photovoltaic output power prediction method and system based on curve characteristic index clustering | |
| CN110929953A (en) | Photovoltaic power station ultra-short term output prediction method based on cluster analysis | |
| CN112465251A (en) | Short-term photovoltaic output probability prediction method based on simplest gated neural network | |
| CN109858665A (en) | Photovoltaic short term power prediction technique based on Feature Selection and ANFIS-PSO | |
| CN116629428A (en) | Building energy consumption prediction method based on feature selection and SSA-BiLSTM | |
| CN115912502A (en) | Intelligent power grid operation optimization method and device | |
| CN117874584A (en) | Neural network ultra-short term photovoltaic power prediction method based on meteorological clustering | |
| CN117277372A (en) | Multi-time-scale joint scheduling method and system for optical storage station and electronic equipment | |
| CN114862023A (en) | Distributed photovoltaic power prediction method and system based on four-dimensional point-by-point meteorological forecast | |
| Chen et al. | Probabilistic Prediction of Photovoltaic Power Using Bayesian Neural Network-LSTM Model | |
| CN115481788B (en) | Phase change energy storage system load prediction method and system | |
| CN116663404A (en) | Flood forecasting method and system coupling artificial intelligence and Bayesian theory | |
| CN116681154A (en) | Photovoltaic power calculation method based on EMD-AO-DELM | |
| CN116432812A (en) | New energy power prediction method for optimizing LSTM (least squares) by using Zun sea squirt algorithm | |
| CN110059871A (en) | Photovoltaic power generation power prediction method | |
| CN115293406A (en) | Photovoltaic power generation power prediction method based on Catboost and Radam-LSTM | |
| CN115392387A (en) | Low-voltage distributed photovoltaic power generation output prediction method | |
| CN114139783A (en) | Wind power short-term power prediction method and device based on nonlinear weighted combination | |
| Wang et al. | Optimization of Convolutional Long Short-Term Memory Hybrid Neural Network Model Based on Genetic Algorithm for Weather Prediction | |
| Ma et al. | Short-Term PV Power Prediction Based on FCM-ISSA-LSTM | |
| Fang et al. | A FCM-XGBoost-GRU Model for Short-Term Photovoltaic Power Forecasting Based on Weather Classification |
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 |