A photovoltaic power generation prediction method based on DeepIM deep learning algorithm
Through the DeepIM deep learning algorithm combined with EMD and PCA, the key factors of photovoltaic power generation environment data are screened, and the problem of accuracy differences in photovoltaic power generation prediction is solved, and higher accuracy prediction is achieved to ensure the stable operation of the power system.
Patent Information
- Application Number
- CN202311175114.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-09-13
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2043-09-13
AI Technical Summary
The accuracy of the existing photovoltaic power generation prediction methods varies greatly in short-term and ultra-short-term predictions, and factors such as the operating status of the photovoltaic station and the operating status of the energy storage system are not considered.
DeepIM deep learning algorithm combined with empirical modal decomposition (EMD) and principal component analysis (PCA), uses Decomposition and dimensionality reduction to process photovoltaic power generation environment data, screens key factors, and uses deep neural networks to perform dynamic time modeling to perform photovoltaic power generation. Quantity prediction.
The accuracy of photovoltaic power generation prediction is improved to ensure the safe scheduling and stable operation of the power system.
Smart Images

Figure CN117194962B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of photovoltaic energy storage, and in particular to a photovoltaic power generation prediction method based on a DeepIM deep learning algorithm. Background Art
[0002] Under the dual carbon goals, photovoltaic power generation has developed rapidly across the country in recent years because it meets the conditions of clean, efficient and renewable. However, photovoltaic power generation is easily affected by the external environment and has problems such as unstable output and large fluctuations. The grid absorption capacity and power system regulation resources that are compatible with its development scale are gradually facing challenges.
[0003] The introduction of energy storage technology can achieve energy transfer, reduce the fluctuation range of power generation, and smooth the output of power generation. The combined operation of photovoltaic and storage has become an effective means to solve this series of problems. The current application and control of energy storage in photovoltaic power generation systems are mainly aimed at reducing the fluctuation of photovoltaic power generation and coordinating the participation of grid auxiliary services. The energy storage system and photovoltaic power generation are independent of each other. Usually, the energy storage system passively receives the power generation forecast results and the photovoltaic power generation and other power generation data to calculate the output of active or reactive compensation. At the same time, the short-term (next 24 hours) and ultra-short-term (next 4 hours) power generation forecasts of photovoltaic power station power generation forecast systems are only based on the station meteorological forecast data and historical data, and are calculated using physical, statistical and deep learning methods. Affected by the prediction time scale, there is a large difference in the accuracy of the two predictions, and factors such as the operating status of the photovoltaic station and the operating status of the energy storage system are not taken into account. Summary of the invention
[0004] The purpose of the present invention is to provide a photovoltaic power generation prediction method based on the DeepIM deep learning algorithm to solve the problem that there is a large difference in the accuracy of the two predictions proposed in the above background technology, and factors such as the operating status of the photovoltaic station and the operating status of the energy storage system are not taken into account.
[0005] To achieve the above object, the present invention provides the following technical solution: a photovoltaic power generation prediction method based on DeepIM deep learning algorithm, specifically comprising the following steps:
[0006] S1: Collect historical environmental data of photovoltaic power generation, perform decomposition and dimensionality reduction processing in sequence after noise reduction to obtain a sample data set;
[0007] The historical environmental data is decomposed by an EMD algorithm to obtain first characteristic sequence data, the first characteristic sequence data is subjected to dimensionality reduction processing by a PCA algorithm to screen out key factors affecting photovoltaic output power generation to obtain second characteristic sequence data, and the second characteristic sequence data is normalized with the pre-collected historical data of photovoltaic power generation to obtain a sample data set;
[0008] S2: Divide the sample data set into a training set and a test set, and import them into the DeepIM deep learning network for training verification, adjust the parameters of the DeepIM deep learning network, obtain the trained DeepIM model, and input the real-time collected photovoltaic power generation environment data into the DeepIM model after decomposition and dimensionality reduction processing to output the prediction results.
[0009] Preferably, the environmental data in S1 and S2 specifically include solar radiation temperature, component temperature, air temperature, relative humidity and atmospheric pressure.
[0010] Preferably, the noise reduction in S1 specifically includes cleaning the environmental data on a daily basis to remove bad data in the environmental data;
[0011] Wherein, the bad data is caused by a communication failure.
[0012] Preferably, the decomposition process in step S1 is specifically: decomposing the environmental data by using an EMD algorithm to obtain the first characteristic sequence data is specifically: decomposing the environmental data into eigenmode components {IMF1, IMF2, ..., IMF m} and the remaining component r n .
[0013] Preferably, the specific steps of decomposing environmental data using the EMD algorithm are as follows:
[0014] S1-11: Select an original data sequence x(t) in the environmental data for screening, find all its maximum points and determine them as the upper envelope, and all its minimum points and determine them as the lower envelope, m(t) represents the mean of the upper envelope and the lower envelope, and the first component h1(t) = x1(t) - m1(t);
[0015] S1-12: Perform a second screening on the first component, h1(t) is regarded as the original data, m1(t) is the mean of the upper and lower envelopes of h1(t), and repeat step S21 to determine the second component h2(t);
[0016] S1-13: The screening process is repeated n times until h n (t) is an intrinsic mode function or residual component r n (t) is a monotonic function, terminating the decomposition process;
[0017] Where q1(t)=h1(t), q2(t)=h2(t), ..., q i (t) = h n (t), x(t) is finally decomposed into n eigenmode components q i(t) and a residual component r n (t), as shown in the following formula:
[0018]
[0019] Preferably, in step S1, the PCA algorithm is used to perform dimensionality reduction processing on the first characteristic sequence data to screen out key factors affecting photovoltaic output power generation, and the specific steps of obtaining the second characteristic sequence data are as follows:
[0020] S1-21: Get the standardized matrix of the first characteristic sequence data, and use the Pearson correlation coefficient analysis to get the correlation between each environmental factor data and the photovoltaic power generation data:
[0021]
[0022] In the formula: X represents environmental factor data, y represents power generation;
[0023] S1-22: Calculate contribution rate π i and the cumulative contribution rate η i , according to η i Determine the number of principal components to be selected:
[0024]
[0025]
[0026] Where: λ represents the eigenvector, i,k represents the eigenvectors of the i,kth factors;
[0027] S1-23: By selecting the eigenvalues of the principal components and the corresponding eigenvectors, the reduced-dimensional data can be finally obtained.
[0028] Preferably, in step S1-22, according to n i The specific method for determining the number of principal components to be selected is to select η i Greater than 95% of the principal components.
[0029] Preferably, in step S2, the sample data set is divided into a training set and a test set, and the sets are imported into a DeepIM deep learning network for training verification, and the parameters of the DeepIM deep learning network are adjusted to obtain a trained DeepIM model. Specifically, the sample data set is divided into a training set and a test set, and the training set is imported into the DeepIM deep learning network for training, and it is determined whether the accuracy of the output result of the DeepIM deep learning network reaches the standard value. If not, the parameters of the DeepIM deep learning network are adjusted, and the training is repeated until the accuracy reaches the standard. If so, the test set is imported into the DeepIM deep learning network, and the evaluation result is output and compared with the historical data of photovoltaic power generation for verification to obtain the optimal DeepIM model.
[0030] Compared with the prior art, the beneficial effects of the present invention are as follows: the present application proposes a photovoltaic power generation prediction model combining empirical mode decomposition (EMD) and a deep learning network model (DeepIM), which fully considers the five environmental factors that restrict photovoltaic output power generation. First, the environmental factor sequence is decomposed by using EMD to obtain the changes in the data signal on different time scales, thereby reducing the non-stationarity of the environmental factor sequence; secondly, PCA is used to extract the key influencing factors of the feature sequence, eliminate the correlation and redundancy of the original sequence, and reduce the dimension of the model input; finally, a deep neural network is used to perform dynamic time modeling on the multivariate feature sequence, thereby realizing the prediction of photovoltaic power generation and improving the prediction accuracy of photovoltaic power generation, which is of great significance for ensuring the safe dispatch and stable operation of the power system. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] Figure 1 It is a schematic diagram of a flow chart of a method for constructing a photovoltaic power generation prediction model of the present invention;
[0032] Figure 2 Schematic diagram of the DeepIM neural network of the present invention. DETAILED DESCRIPTION
[0033] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0034] Example 1
[0035] See also Figure 1 and Figure 2 , a photovoltaic power generation prediction method based on DeepIM deep learning algorithm, specifically comprising the following steps:
[0036] Step 1: Real-time data collection. Taking into full consideration the main environmental factors that restrict photovoltaic power generation, environmental data in five dimensions, including solar irradiance, component temperature, air temperature, relative humidity and atmospheric pressure, are collected. Parallel data cleaning is performed to eliminate "bad data" caused by communication failures and other reasons in actual production on a daily basis.
[0037] Step 2: Decompose the environmental data into eigenmode components of different frequencies using the EMD algorithm
[0038] {IMF1, IMF2, …, IMF m} and the remaining component r n , decompose the original environmental sequence into various characteristic fluctuation sequences, thereby decomposing the fluctuations or trends of different scales in the original environmental signal step by step.
[0039] The essence of the EMD algorithm is to obtain different numbers of intrinsic mode functions (IMFs) from the characteristic time scale of the data. Different intrinsic mode components IMF represent different characteristic fluctuation sequences, which make the fluctuation characteristics of the original data stand out at different time scales. Since the five environmental time series have certain randomness and discontinuity, through EMD decomposition, while enriching the diversity of input variables, according to the obtained IMF components, the local characteristics of the environmental series at different time scales can be highlighted, reflecting the volatility, periodicity and trend changes of the original environmental series. The specific decomposition process is as follows:
[0040] 1) For an original data sequence x(t), find all its maximum points and determine them as the upper envelope, and all its minimum points and determine them as the lower envelope. m(t) represents the mean of the upper envelope and the lower envelope, and the first component h1(t) = x1(t) - m1(t);
[0041] 2) In the second screening process, h1(t) is regarded as the original data, m1(t) is the mean of the upper and lower envelopes of h1(t), and the second component h2(t) is determined as shown in 1);
[0042] 3) The screening process is repeated n times until h n (t) is an intrinsic mode function or residual component r n (t) is a monotonic function, terminating the decomposition process;
[0043] 4) In summary, specify q1(t) = h1(t), q2(t) = h2(t), ..., (t) = h n (t), x(t) is finally decomposed into n eigenmode components q i (t) and a residual component r n (t), as shown in the following formula:
[0044]
[0045] Step 3: Perform PCA dimensionality reduction on the data decomposed in step 2, use the PCA algorithm to screen out the key factors affecting the photovoltaic output power generation, and eliminate the redundancy and correlation of different time series data obtained by EMD decomposition.
[0046] PCA is a classic data dimensionality reduction method. It converts the original data into a new feature space through linear transformation to extract the most important linear components of the data. The algorithm steps are as follows:
[0047] 1) Obtain the standardized matrix of the original data, and use the Pearson correlation coefficient analysis to obtain the correlation between the environmental factor data and the power generation data;
[0048]
[0049] 2) Calculate the contribution rate π i and the cumulative contribution rate η i , the calculation method is shown in formula 3) and formula 4), according to η i Determine the number of principal components to be selected;
[0050]
[0051]
[0052] 5) By selecting the eigenvalues of the principal components and the corresponding eigenvectors, the reduced-dimensional data can be finally obtained.
[0053] Step 4: Normalize the data after dimension reduction in step 3 and the historical data of photovoltaic power generation, convert them into a data set suitable for DEEPIM network training, and divide them into training set and test set.
[0054] Step 5: Initialize the DeepIM model parameters and input the sample training set into the DeepIM model for training to obtain the target accuracy of the model.
[0055] The hidden layer cell structure of DeepIM is shown in the figure below. By introducing the control gate, the problem of RNN gradient disappearance and explosion is solved, making the deep neural network unfolded in time easier to train. For a given sequence x = (x1, x2, ..., x n ), the forward calculation process is shown as follows:
[0056] i t =σ(W xi x t +W hi h t-1+W ci c t-1 +b i )
[0057] f t =σ(W xf x t +W hf h t-1 +W cf c t-1 +b f )
[0058] c t =f t c t-1 +i t tanh(W xc x t +W hc h t-1 +b c )
[0059] o t =σ(W xo x t +W ho h t-1 +W co c t +b o )
[0060] h t =o t tanh(c t )
[0061] Where: W: corresponding weight coefficient matrix; b: corresponding bias term; σ: sigmoid activation function; tanh: hyperbolic tangent activation function.
[0062] Step 6: After the model training is completed, save the training file and input the test set for testing. Output the model evaluation indicators MAPE and MSE;
[0063]
[0064]
[0065] It should be noted that after the test set is input into the model, the result (prediction result) is output. By comparing the result with the historical data, the accuracy of the model prediction can be verified.
[0066] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A photovoltaic power generation prediction method based on DeepIM deep learning algorithm, Features: The specific steps include: S1: Collect historical environmental data of photovoltaic power generation, perform decomposition and dimensionality reduction processing in sequence after noise reduction to obtain a sample data set; The historical environmental data is decomposed by an EMD algorithm to obtain first characteristic sequence data, the first characteristic sequence data is subjected to dimensionality reduction processing by a PCA algorithm to screen out key factors affecting photovoltaic output power generation to obtain second characteristic sequence data, and the second characteristic sequence data is normalized with the pre-collected historical data of photovoltaic power generation to obtain a sample data set; S2: Divide the sample data set into a training set and a test set, and import them into the DeepIM deep learning network for training verification, adjust the parameters of the DeepIM deep learning network, obtain a trained DeepIM model, and input the real-time collected photovoltaic power generation environment data into the DeepIM model after decomposition and dimensionality reduction processing to output prediction results; Among them, DeepIM deep learning network for a given sequence x = (x 1 , x 2 , …, x n ), the forward calculation process is shown as follows: i t =σ(W xi x t +W hi h t-1 +W ci c t-1 +b i ) f t =σ(W xf x t +W hf h t-1 +W cf c t-1 +b f ) c t =f t c t-1 +i t tanh(W xc x t +W hc h t-1 +b c ) o t =σ(W xo x t +W ho h t-1 +W co c t +b o ) h t =o t fishy t ) Where: W: corresponding weight coefficient matrix; b: corresponding bias term; σ: sigmoid activation function; tanh: hyperbolic tangent activation function; The environmental data in S1 and S2 specifically include solar radiation temperature, component temperature, air temperature, relative humidity and atmospheric pressure; The noise reduction in S1 specifically includes cleaning the environmental data on a daily basis to remove bad data in the environmental data; Wherein, the bad data is caused by a communication failure; The decomposition process in step S1 is specifically as follows: decomposing the environmental data by using the EMD algorithm to obtain the first characteristic sequence data is specifically as follows: decomposing the environmental data into eigenmode components {IMF 1 , IMF 2 , ..., IMF m } and the remaining component r n ; The specific steps of the EMD algorithm to decompose environmental data are as follows: S1-11: Select an original data sequence x(t) from the environmental data for screening, find all its maximum points as the upper envelope, and all its minimum points as the lower envelope. m(t) represents the mean of the upper and lower envelopes. The first component h 1 (t) = x 1 (t)-m 1 (t); S1-12: Perform a second screening on the first component, h 1 (t) is regarded as the original data, m 1 (t) is h 1 (t) and repeat steps S1-11 to determine the second component h 2 (t); S1-13: The screening process is repeated n times until h n (t) is an intrinsic mode function or residual component r n (t) is a monotonic function, terminating the decomposition process; Among them, specify q 1 (t) = h 1 (t), q 2 (t) = h 2 (t), ..., q i (t) = h n (t), x(t) is finally decomposed into n eigenmode components q i (t) and a residual component r n (t), as shown in the following formula: In step S1, the PCA algorithm is used to perform dimensionality reduction processing on the first characteristic sequence data to screen out the key factors affecting the photovoltaic output power generation, and the specific steps of obtaining the second characteristic sequence data are as follows: S1-21: Get the standardized matrix of the first characteristic sequence data, and use the Pearson correlation coefficient analysis to get the correlation between each environmental factor data and the photovoltaic power generation data: In the formula: X represents the data of various environmental factors, y represents the power generation; S1-22: Calculate contribution rate π i and the cumulative contribution rate η i , according to η i Determine the number of principal components to be selected; Where: λ represents the eigenvector, i,k represents the eigenvectors of different factors i,k S1-23: Through the selected eigenvalues of the principal components and the corresponding eigenvectors, the reduced-dimensional data can be finally obtained; In step S1-22, according to n i The specific method for determining the number of principal components to be selected is to select η i Greater than 95% of the principal components; In step S2, the sample data set is divided into a training set and a test set, and the sets are imported into a DeepIM deep learning network for training verification, and the parameters of the DeepIM deep learning network are adjusted to obtain a trained DeepIM model. Specifically, the sample data set is divided into a training set and a test set, and the training set is imported into the DeepIM deep learning network for training, and it is determined whether the accuracy of the output result of the DeepIM deep learning network reaches the standard value. If not, the parameters of the DeepIM deep learning network are adjusted, and the training is repeated until the accuracy reaches the standard. If so, the test set is imported into the DeepIM deep learning network, and the evaluation result is output and compared with the historical data of photovoltaic power generation for verification to obtain the optimal DeepIM model.
Citation Information
Patent Citations
Multivariable input photovoltaic power prediction method based on EMD-PCA-LSTM
CN110516844A
SM-OFDM signal grouping detection method based on deep neural network
CN114584448A
Photovoltaic power generation power prediction method based on combined machine learning model and application thereof
CN115860177A
A system for forecasting solar irradiance for a solar-based photovoltaic system and method thereof
IN202011044299A
Device and method for diagnosing / predicting status of solar cell or solar cell array on real time based on machine learing
KR1020180123951A
