CN117743829A - Short-term power load quantity prediction method based on deep learning - Google Patents
Short-term power load quantity prediction method based on deep learning Download PDFInfo
- Publication number
- CN117743829A CN117743829A CN202311831821.5A CN202311831821A CN117743829A CN 117743829 A CN117743829 A CN 117743829A CN 202311831821 A CN202311831821 A CN 202311831821A CN 117743829 A CN117743829 A CN 117743829A
- Authority
- CN
- China
- Prior art keywords
- decomposition
- power load
- short
- deep learning
- term power
- 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 38
- 238000013135 deep learning Methods 0.000 title claims abstract description 14
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 31
- 238000005457 optimization Methods 0.000 claims abstract description 12
- 230000000694 effects Effects 0.000 claims abstract description 5
- 241000287127 Passeridae Species 0.000 claims description 19
- 230000006870 function Effects 0.000 claims description 16
- 238000012549 training Methods 0.000 claims description 9
- 238000011156 evaluation Methods 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000003064 k means clustering Methods 0.000 claims description 4
- 238000010845 search algorithm Methods 0.000 claims description 4
- 230000002776 aggregation Effects 0.000 claims description 3
- 238000004220 aggregation Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 238000012163 sequencing technique Methods 0.000 claims description 3
- 230000001419 dependent effect Effects 0.000 claims description 2
- 239000006185 dispersion Substances 0.000 claims description 2
- 238000012795 verification Methods 0.000 claims 1
- 238000010586 diagram Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000010200 validation analysis Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 238000012360 testing method Methods 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 discloses a short-term power load quantity prediction method based on deep learning, which comprises the following implementation steps: (1) performing empirical mode decomposition on load data; (2) sub-sequence classification; (3) high frequency component re-decomposition and parameter optimization; (4) modeling; (5) Compared with the traditional short-term power load prediction method, the method can convert the non-stable and nonlinear original time sequence into a plurality of subsequences and reconstruct the subsequences by introducing modal decomposition, thereby solving the problems of large short-term power load fluctuation and difficult accurate prediction; the high-frequency component is subjected to secondary decomposition, so that the obvious characteristic of time sequence data can be effectively extracted, and the subsequence obtained by secondary decomposition and the subsequence without secondary decomposition are used as model input characteristics, so that the prediction accuracy is effectively improved; the problem of selecting key parameters of LSTM network and VMD decomposition is solved by using SSA to optimize, and the problems of time consumption and poor prediction effect of manual parameter adjustment are solved.
Description
Technical Field
The invention relates to the technical field of power load prediction, in particular to a power load prediction method based on deep learning.
Background
The power system load prediction is to search the influence of the historical data change rule of the power load on the future load according to the historical data of the power load, economy, society, weather and the like, search the internal relation between the power load and various related factors, and scientifically predict the future power load. Inaccurate power load prediction may cause power system faults to cause large-scale power failure, and normal production and life of society are affected. Therefore, accurate load prediction modeling has important significance for ensuring safe and stable operation of the power system.
The traditional load prediction model lacks of adaptability and prediction capability, has poor robustness, inaccurate prediction results and difficulty in meeting the load prediction requirement in precision.
Disclosure of Invention
Aiming at the problem of power load prediction precision, the invention provides a power load prediction method combining a modal decomposition technology and deep learning, which combines model training time and prediction precision, realizes high-precision short-term power load prediction and provides a certain technical reference for accurate regulation and control of power load. In order to achieve the aim of the invention, the technical scheme adopted by the invention is as follows:
the electric load prediction method based on deep learning comprises the following implementation steps:
(1) Carrying out empirical mode decomposition on the load data; (2) sub-sequence classification; (3) high frequency component re-decomposition and parameter optimization; (4) modeling; and (5) predicting results, specifically:
step 1: decomposing the original load data by adopting empirical mode decomposition to obtain n intrinsic mode functions IMF i (i=1, 2,..n.) and the amount of residue.
The specific method comprises the following steps:
1) Interpolation is carried out on the original signal by utilizing a cubic spline, and an upper envelope curve X and a lower envelope curve X of the original signal are searched max (t) and X min (t) and calculating the average envelope as
2) Original sequence signal and envelope mean value m j-1 (t) subtracting, X (t) -m j-1 (t)＝h j (t) obtainingTo h j (t). H at this point j And (t) is a residual signal.
3) To obtain a smoother sequence, the above steps are repeated, and the remaining components are decomposed until the termination condition is satisfied.
Step 2: sample entropy calculation and classification are performed on the sub-sequences.
The IMF components obtained after the empirical mode decomposition are arranged from high to low in frequency. Computing an eigenmode function IMF using a sample entropy function i Sample entropy of (i=1, 2, n) and residual, assessing modal component complexity; the K-means clustering algorithm is utilized to divide the K-means clustering algorithm into three types to form high, medium and low frequency components.
Step 3: and re-decomposing the high-frequency component by adopting a variational mode VMD algorithm to obtain a series of mode components.
VMD decomposition is essentially a solution to the problem of variance, which can be solved by establishing the following optimization problems:
wherein alpha is a penalty factor; lambda is Lagrange multiplier, u k Is the kth modal function component; w (w) k Center frequency of the kth mode function; delta (t) is a dirac function, |·| represents a norm, f (t) represents an original signal, and x represents a convolution operator.
Step 4: establishing a corresponding LSTM model for each subsequence to perform training prediction, wherein sparrow search optimization algorithm is used for assisting model parameter selection;
the method comprises the following steps: first, a set of superparameters is randomly generated for LSTM model training. The fitness of each set of hyper-parameters is then determined by evaluating the performance of the LSTM model on the validation set. And sorting all sparrows in the sparrow search algorithm according to the fitness evaluation result, and arranging the sparrows with higher fitness in front. And selecting sparrows to mate and mutate according to the sequencing result so as to generate a new super-parameter combination. And repeating the operation until the algorithm converges, namely finding the optimal super-parameter combination. And sending parameters obtained by the sparrow search optimization algorithm into the LSTM to complete model training.
Step 5: model prediction and evaluation.
And carrying out inverse normalization processing on the predicted value obtained by each LSTM predicted model, and then carrying out aggregation operation to obtain a final actual predicted value. Introducing Root Mean Square Error (RMSE), mean Absolute Error (MAE), determining coefficient (R 2 ) The index is used to check the predictive effect of the model.
The beneficial effects of the invention are as follows:
according to the invention, by introducing modal decomposition, a non-stationary and nonlinear original time sequence can be converted into a plurality of subsequences and reconstructed, so that the problems of large short-term power load fluctuation and difficulty in accurate prediction are solved; the high-frequency component is subjected to secondary decomposition, so that the obvious characteristic of time sequence data can be effectively extracted, and the subsequence obtained by secondary decomposition and the subsequence without secondary decomposition are used as model input characteristics, so that the prediction accuracy is effectively improved; the problem of selecting key parameters of LSTM network and VMD decomposition is solved by using SSA to optimize, and the problems of time consumption and poor prediction effect of manual parameter adjustment are solved.
Drawings
Fig. 1 is: a flow chart of the method of the invention;
fig. 2 is: in the method, in the step 2, a sub-sequence classification reconstruction schematic diagram obtained by decomposition is obtained;
fig. 3 is: in the method, a sub-sequence diagram obtained by VMD (virtual machine direction detector) re-decomposition is adopted for the high-frequency component;
fig. 4 is: the predicted result and the true value of the method are shown in a schematic diagram;
Detailed Description
The invention is further described below with reference to the accompanying drawings: the flow chart of the method of the invention is shown in figure 1, and the specific implementation process of the method is as follows:
step 1: the original power load data is decomposed by adopting empirical mode decomposition (EMD, empirical Model Decomposition) to obtain a plurality of intrinsic mode functions IMF and residual quantity residual. The specific method comprises the following steps:
1) Interpolation is carried out on the original signal by utilizing a cubic spline, and an upper envelope curve X and a lower envelope curve X of the original signal are searched max (t) andX min (t) and calculating the average envelope as
2) Original sequence signal and envelope mean value m j-1 (t) subtracting, X (t) -m j-1 (t)＝h j (t) obtaining h j (t)。
3) To obtain a smoother sequence, the above steps are repeated, and the remaining components are decomposed until the termination condition is satisfied.
Step 2: the IMF components obtained after the empirical mode decomposition are arranged from high to low in frequency. Sample entropy of all the modal components is calculated, classification is carried out by using a K-means clustering algorithm, and the modal components in the same class are reconstructed to form three kinds of high-medium-low frequency reconstruction components, as shown in figure 2.
The sample entropy calculation method comprises the following steps:
1) if the original signal is a sequence with length N, X (1), X (2), X (3) … X (N), samples with length m are sequentially taken, where the i-th sample can be represented as X (i) = [ X (i), X (i+1), L X (i+m-1) ], i=1 to N-m-1.
2) Calculating the distance d [ X (i), X (j) between samples X (i) and X (j)]Defined as the maximum value of the difference between corresponding elements, i.eThe symbol || represents an absolute value operation;
3) Given a threshold value r, counting the number of which the distance is smaller than the threshold value r and comparing the number with N-m, and recording as
4) Solving the sample entropy of the sequence:
then, for the solved sequence sample entropy, the given data set is divided into K clusters (K is set to 3 in the present invention) by kmens, and a center point corresponding to each sample data is given.
Step 3: and re-decomposing the reconstructed high-frequency component by adopting a VMD algorithm, and optimizing parameters, such as the number of decompositions and penalty factors, by using a sparrow search algorithm. The specific method comprises the following steps:
VMD decomposition is essentially a solution to the problem of variance, which can be solved by establishing the following optimization problems:
wherein alpha is a secondary penalty factor; lambda is Lagrange multiplier, u k Is the kth modal function component; w (w) k Center frequency of the kth mode function; alpha t Is a dirac function.
Solving the above to obtain an updated formula as
Where n represents the number of iterations. Wherein the method comprises the steps ofRespectively indicate->fourier transforms of f (t), u (t) and λ (t). The update process termination conditions are: />Where τ is the noise margin, ε>And 0 is a convergence threshold.
Step 4: and building a corresponding LSTM model for each subsequence to perform training prediction, wherein a sparrow search optimization algorithm is used for assisting model parameter selection:
the method comprises the following steps: first, a set of superparameters is randomly generated for LSTM model training. The fitness of each set of hyper-parameters is then determined by evaluating the performance of the LSTM model on the validation set. And sorting all sparrows in the sparrow search algorithm according to the fitness evaluation result, and arranging the sparrows with higher fitness in front. And selecting sparrows to mate and mutate according to the sequencing result so as to generate a new super-parameter combination. And repeating the operation until the algorithm converges, namely finding the optimal super-parameter combination. And sending parameters obtained by the sparrow search optimization algorithm into the LSTM to complete model training.
Step 5: model prediction and evaluation.
And carrying out inverse normalization processing on the predicted value obtained by each LSTM predicted model, and then carrying out aggregation operation to obtain a final actual predicted value. Introducing Root Mean Square Error (RMSE), mean Absolute Error (MAE), determining coefficient (R 2 ) The index is used to check the predictive effect of the model.
The root mean square error reflects the dispersion of the predicted values and the magnitude of the model error.
The average absolute error is the average value of the absolute errors, and can better reflect the actual situation of the predicted value errors.
The goodness of fit may reflect the degree of interpretation of the independent variable by the dependent variable, with a larger value indicating a better goodness of fit.
FIG. 4 is a graph showing the predicted results and the actual values of the method of the present invention. Wherein the prediction results (parts) are shown in table 1:
TABLE 1 prediction results at true values
| Time | Actual load amount kWh | Predicting load kWh |
| 22:30 | 2269.729 | 2209.407 |
| 22:45 | 2203.306 | 2116.752 |
| 23:00 | 2385.568 | 2393.735 |
| 23:15 | 1945.762 | 1975.920 |
| 23:30 | 1753.181 | 1724.883 |
Table 2 shows the evaluation index comparison table of the predicted results of the method and the comparison model. The following are provided:
table 2 evaluation index table
| Model | The method of the invention | EMD-LSTM | VMD-LSTM |
| R 2 | 0.921 | 0.697 | 0.833 |
| RMSE | 128.1 | 236.98 | 175.79 |
| MAE | 83.65 | 122.93 | 100.43 |
From tables 1 and 2, the power load is predicted by using the test data, and the method has the advantages of higher fitting degree of the predicted result and the real result and high prediction precision.
The deep learning algorithm can mine the internal rule of the model through data analysis, so that the aim of constructing the model to predict the load is fulfilled. Aiming at the time sequence change characteristics of the load data, the load data is decomposed into a plurality of modal components with different frequencies through a decomposition algorithm, so that the difficulty of data fitting is further reduced, the data curve is smooth and periodic, and the fitting degree of the neural network is improved.
The foregoing is merely a preferred embodiment of the invention, and it should be noted that modifications could be made by those skilled in the art without departing from the principles of the invention, which modifications would be considered to be within the scope of the invention without inventive faculty.
Claims (7)
1. The short-term power load prediction method based on deep learning is characterized by comprising the following implementation steps: (1) performing empirical mode decomposition on load data; (2) sub-sequence classification; (3) high frequency component re-decomposition and parameter optimization; (4) modeling; and (5) predicting the result.
2. The short-term power load prediction method based on deep learning as claimed in claim 1, wherein the specific method steps are as follows: step (1): decomposing the original load data by adopting empirical mode decomposition to obtain n intrinsic mode functions IMF i (i=1, 2, n) and the residual amount;
step (2): sample entropy calculation and classification are carried out on the subsequences;
arranging the IMF components obtained after empirical mode decomposition according to the frequency from high to low; computing an eigenmode function IMF using a sample entropy function i Sample entropy of (i=1, 2, n) and residual, assessing modal component complexity; dividing the high-frequency component, the medium-frequency component and the low-frequency component into three types by using a K-means clustering algorithm;
step (3): re-decomposing the high-frequency component by adopting a variational mode VMD algorithm to obtain a series of mode components;
step (4): establishing a corresponding LSTM model for each subsequence to perform training prediction, wherein sparrow search optimization algorithm is used for assisting model parameter selection;
step (5): model prediction and evaluation: performing inverse normalization processing on the predicted value obtained by each LSTM predicted model, performing aggregation operation to obtain final actual predicted value, introducing Root Mean Square Error (RMSE), mean Absolute Error (MAE), and determining coefficient (R) 2 ) The index is used to check the predictive effect of the model.
3. The short-term power load prediction method based on deep learning according to claim 2, wherein the specific method of step (1) is:
1) Interpolation is carried out on the original signal by utilizing a cubic spline, and an upper envelope curve X and a lower envelope curve X of the original signal are searched max (t) and X min (t) and calculating the average envelope as
2) Original sequence signal and envelope mean value m j-1 (t) subtracting, X (t) -m j-1 (t)＝h j (t) obtaining h j (t). H at this point j (t) is a residual signal;
3) To obtain a smoother sequence, the above steps are repeated, and the remaining components are decomposed until the termination condition is satisfied.
4. The short-term power load prediction method based on deep learning according to claim 2, wherein the specific method of step (3) is:
the sample entropy calculation method comprises the following steps:
1) if the original signal is a sequence with length of N, respectively X (1), X (2), X (3) … X (N), taking samples with length of m according to the sequence, wherein the ith sample can be expressed as X (i) = [ X (i), X (i+1), L X (i+m-1) ], and i=1 to N-m-1;
2) Calculating the distance d [ X (i), X (j) between samples X (i) and X (j)]Defined as the maximum value of the difference between corresponding elements, i.eThe symbol || represents an absolute value operation;
3) Given a threshold value r, counting the number of which the distance is smaller than the threshold value r and comparing the number with N-m, and recording as
4) Solving the sample entropy of the sequence:
then, for the solved sequence sample entropy, the given data set is divided into K clusters (K is set to 3 in the present invention) by kmens, and a center point corresponding to each sample data is given.
5. The short-term power load prediction method based on deep learning according to claim 2, wherein the specific method of step (3) is:
VMD decomposition is essentially a solution to the problem of variance, which can be solved by establishing the following optimization problems:
wherein alpha is a penalty factor; lambda is Lagrange multiplier, u k Is the kth modal function component; w (w) k Center frequency of the kth mode function; delta (t) is a dirac function, |·| represents a norm, f (t) represents an original signal, and x represents a convolution operator.
6. The short-term power load prediction method based on deep learning according to claim 2, wherein the specific method of step (4) is:
firstly, randomly generating a group of super parameters to train an LSTM model; then, determining the fitness of each group of super parameters by evaluating the performance of the LSTM model on the verification set; sorting all sparrows in the sparrow search algorithm according to the fitness evaluation result, and arranging the sparrows with higher fitness in front; selecting sparrows to mate and mutate according to the sequencing result so as to generate a new super-parameter combination; repeating the operation until the algorithm converges, namely finding out the optimal super-parameter combination; and sending parameters obtained by the sparrow search optimization algorithm into the LSTM to complete model training.
7. The short-term electric power load prediction method based on deep learning according to claim 2, wherein in step (5): the root mean square error reflects the dispersion of the predicted values and the magnitude of the model error:
the average absolute error is the average value of absolute errors, and can better reflect the actual situation of the predicted value error:
the goodness of fit may reflect the degree of interpretation of the independent variable by the dependent variable, with a larger value indicating a better goodness of fit:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311831821.5A CN117743829A (en) | 2023-12-28 | 2023-12-28 | Short-term power load quantity prediction method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311831821.5A CN117743829A (en) | 2023-12-28 | 2023-12-28 | Short-term power load quantity prediction method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117743829A true CN117743829A (en) | 2024-03-22 |
Family
ID=90283131
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311831821.5A Pending CN117743829A (en) | 2023-12-28 | 2023-12-28 | Short-term power load quantity prediction method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117743829A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118014166A (en) * | 2024-04-09 | 2024-05-10 | 陕西德联新能源有限公司 | Load prediction method and system of heating system based on big data |
-
2023
- 2023-12-28 CN CN202311831821.5A patent/CN117743829A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118014166A (en) * | 2024-04-09 | 2024-05-10 | 陕西德联新能源有限公司 | Load prediction method and system of heating system based on big data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111860982A (en) | Wind power plant short-term wind power prediction method based on VMD-FCM-GRU | |
| CN112990556A (en) | User power consumption prediction method based on Prophet-LSTM model | |
| CN111382906B (en) | Power load prediction method, system, equipment and computer readable storage medium | |
| CN109583635B (en) | Short-term load prediction modeling method for operational reliability | |
| CN112434848B (en) | Nonlinear weighted combination wind power prediction method based on deep belief network | |
| CN117743829A (en) | Short-term power load quantity prediction method based on deep learning | |
| CN115822887A (en) | Performance evaluation and energy efficiency diagnosis method and system of wind turbine generator | |
| CN114154716B (en) | Enterprise energy consumption prediction method and device based on graph neural network | |
| CN111060652A (en) | Method for predicting concentration of dissolved gas in transformer oil based on long-term and short-term memory network | |
| CN112364560A (en) | Intelligent prediction method for working hours of mine rock drilling equipment | |
| Wang et al. | A remaining useful life prediction model based on hybrid long-short sequences for engines | |
| CN109921426A (en) | Wind-electricity integration system probability load flow calculation method based on CV-KDE | |
| CN108228978B (en) | Xgboost time sequence prediction method combined with complementary set empirical mode decomposition | |
| CN116933175A (en) | Electric automobile charging load prediction method and device | |
| CN115375026A (en) | Method for predicting service life of aircraft engine in multiple fault modes | |
| CN113780420A (en) | Method for predicting concentration of dissolved gas in transformer oil based on GRU-GCN | |
| CN116960978A (en) | Offshore wind power prediction method based on wind speed-power combination decomposition reconstruction | |
| CN116819423A (en) | Method and system for detecting abnormal running state of gateway electric energy metering device | |
| CN111598313A (en) | Wind power prediction method | |
| CN115221793A (en) | Tunnel surrounding rock deformation prediction method and device | |
| CN117709524A (en) | Bayesian optimization-based carbon emission prediction method and system for steel industry | |
| CN113033898A (en) | Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network | |
| CN116865255A (en) | Short-term wind power prediction method based on improved entropy weight method and SECEEMD | |
| CN112001530A (en) | Predictive maintenance method and system for transformer oil chromatography online monitoring device | |
| CN117009758A (en) | Oil well yield prediction method based on PSO-VMD-LSTM |
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 |