CN112488397B - Load prediction method under extreme scene based on modal decomposition and transfer learning - Google Patents
Load prediction method under extreme scene based on modal decomposition and transfer learning Download PDFInfo
- Publication number
- CN112488397B CN112488397B CN202011389811.7A CN202011389811A CN112488397B CN 112488397 B CN112488397 B CN 112488397B CN 202011389811 A CN202011389811 A CN 202011389811A CN 112488397 B CN112488397 B CN 112488397B
- Authority
- CN
- China
- Prior art keywords
- predicted
- load
- trend
- data
- historical data
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 31
- 238000000354 decomposition reaction Methods 0.000 title claims abstract description 18
- 238000013526 transfer learning Methods 0.000 title claims abstract description 14
- 230000002776 aggregation Effects 0.000 claims abstract description 7
- 238000004220 aggregation Methods 0.000 claims abstract description 7
- 230000007246 mechanism Effects 0.000 claims abstract description 7
- 238000012546 transfer Methods 0.000 claims abstract description 6
- 238000012549 training Methods 0.000 claims description 30
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 7
- 230000005012 migration Effects 0.000 claims description 6
- 238000013508 migration Methods 0.000 claims description 6
- 238000012706 support-vector machine Methods 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 5
- 230000006870 function Effects 0.000 abstract description 11
- 230000008859 change Effects 0.000 description 5
- 230000004913 activation Effects 0.000 description 4
- 230000014509 gene expression Effects 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000007873 sieving Methods 0.000 description 2
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 description 1
- 238000006424 Flood reaction Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/003—Load forecast, e.g. methods or systems for forecasting future load demand
-
- 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
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Operations Research (AREA)
- Entrepreneurship & Innovation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Game Theory and Decision Science (AREA)
- Quality & Reliability (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Power Engineering (AREA)
- Development Economics (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method for predicting power load in an extreme scene based on modal decomposition and transfer learning, which comprises the following steps: 1) counting historical data of the power load in various extreme scenes, and classifying the historical data according to the approximate trend of a load curve; 2) dividing the frequency of the data by using an improved aggregation empirical mode decomposition method to obtain a load trend term and a plurality of high-frequency components; 3) performing transfer learning based on attention mechanism model weight transfer by using the trend item of the historical data to obtain a prediction model of the trend item; 4) and respectively carrying out load prediction on the trend item and the intrinsic mode function by using the prediction model and the LSTM network, and superposing prediction results to obtain a load prediction result. The method can fully utilize the historical data of the power load when various extreme events occur, predict the trend of the power load in the similar scene, and solve the problem that the traditional method is not applicable when the power load is predicted in the extreme scene.
Description
Technical Field
The invention relates to the technical field of load prediction of a power system, in particular to a load prediction method under an extreme scene based on modal decomposition and transfer learning.
Background
The load prediction of the power system is an important means and a key link for planning the power system and guiding power production, however, the power load is subjected to sudden change due to uncontrollable factors such as natural disasters such as earthquake, torrential flood, debris flow, typhoon, frost and the like or equipment failure and the like in some extreme scenes. In this case, the historical data has no reference value relative to future load curve changes, and a more accurate predicted value cannot be calculated by a traditional time series method.
Disclosure of Invention
The invention aims to solve the defects of the prior art, and provides a load prediction method under an extreme scene based on modal decomposition and transfer learning, so that the load history data of other similar extreme scenes can be fully utilized to capture the load change trend of the scene, the power load in the process of generating and recovering extreme events can be accurately predicted, and better guidance effect on power scheduling and planning can be realized in response to the emergency events.
In order to achieve the purpose, the invention adopts the following technical scheme:
the invention discloses a method for predicting power load in an extreme scene based on modal decomposition and transfer learning, which is characterized by comprising the following steps of:
step 3, performing transfer learning based on attention mechanism model weight transfer by using the basic trend of the power load to obtain a prediction model of a trend item;
and 4, respectively carrying out load prediction on the basic trend and the high-frequency components of the power load by using the prediction model of the trend item and the LSTM network, and superposing prediction results to obtain a load prediction result.
The method for predicting the power load in the extreme scene is characterized in that the improved aggregation empirical mode decomposition method in the step 2 comprises the following steps:
step 2.1, using a support vector machine to carry out regression prediction on historical data x of the load to be predicted 0 M sets of history data { x) of the same type as the power load history data with the classification label i Carrying out endpoint continuation on | i ═ 1, 2.., m } to obtain a time sequence y to be predicted after continuation 0 And historical time series y i 1, · m }; wherein x is i Representing the ith set of historical data, y i Representing the ith group of historical time series;
step 2.2, treat the prediction time series y 0 And historical time series y i Adding a white noise sequence a with limited amplitude to obtain a processed time sequence y to be predicted a0 And processed historical time series y ai 1, · m }; wherein, y ai Representing the historical time series after the ith group of processing;
step 2.3, the processed time series y to be predicted is decomposed by using empirical mode a0 Decomposed into a basic trend item of the load to be predicted and high-frequency components of j loads to be predicted,processed historical time series y ai Decomposing | i ═ 1, 2.. multidot.m } into m basic trend terms of the training history data and m × j high-frequency components of the training history data;
step 2.4, after different white noise sequences are added each time, repeating the step 2.2 and the step 2.3 for n times, thereby obtaining n basic trend items of the load to be predicted and n multiplied by j high-frequency components of the load to be predicted, and n multiplied by m basic trend items of the historical data for training and n multiplied by m multiplied by j high-frequency components of the historical data for training;
step 2.5, calculating the mean value of the basic trend items of the n loads to be predicted and taking the mean value as a trend item sequence r to be predicted 0 Calculating the average value of the high-frequency components of the n multiplied by j loads to be predicted and taking the average value as a trend item sequence { r ] to be predicted i 1,2, a.m., calculating the mean value of the basic trend terms of the n × m training history data and using the mean value as the high-frequency component sequence C to be predicted 0t Calculating the average of the high frequency components of n × m × j pieces of training history data as a training high frequency component sequence { C it |i＝1,2,...,m；t＝1,2,...,j}。
The transfer learning method based on attention mechanism model weight transfer in the step 3 comprises the following steps:
step 3.1, connecting an attention model containing a learning parameter q with a coder-decoder structure to form a sequence-to-sequence learning network;
step 3.2, using the trend item sequence to be predicted { r i Taking | i ═ 1,2,. and m } as training data, and repeatedly correcting the learning parameter q by an Adam algorithm until errors meet requirements, so as to obtain a trained migration model;
step 3.3, connecting the migration model with an LSTM network as a trend item sequence r to be predicted 0 The predictive model of (1).
The step 4 comprises the following steps:
step 4.1, a trend item sequence r to be predicted 0 Inputting the data into the prediction model as training data, training parameters of an LSTM network in the prediction model through an Adam algorithm, and predicting future data of a basic trend of the power load;
step 4.2, using another LSTM network to respectively predict j high-frequency component sequences C to be predicted 0t Predicting to obtain future data of j high-frequency component sequences;
and 4.3, adding the future data of the basic trend of the power load and the future data of the j high-frequency component sequences to obtain a load prediction result to be predicted.
Compared with the prior art, the invention has the beneficial effects that:
1. the invention can fully utilize the load historical data of other similar extreme scenes to capture the load change trend of the scenes, thereby more accurately predicting the power load in the process of generating and recovering the extreme events;
2. compared with the traditional aggregation empirical mode decomposition algorithm, the improved aggregation empirical mode decomposition algorithm provided by the invention adopts a support vector machine regression prediction method to carry out endpoint continuation, effectively solves the problem of endpoint effect during decomposition of the existing method, and can extract a better load change trend;
3. the transfer learning algorithm provided by the invention uses historical data of other similar scenes to train a query vector in an attention mechanism model, and then transfers the query vector to a prediction model, thereby realizing full utilization of similar prior knowledge and predicting load change in the process of extreme event occurrence and recovery.
Drawings
FIG. 1 is a schematic flow chart of a load prediction method in an extreme scenario according to the present invention;
FIG. 2 is a flow diagram of a prior art empirical mode decomposition algorithm;
fig. 3 is a block diagram of a basic unit of an LSTM network in the prior art.
Detailed Description
In this embodiment, as shown in fig. 1, a method for predicting a load in an extreme scene based on modal decomposition and transfer learning is performed according to the following processes:
specifically, the extreme scenes comprise scenes in which loads are abnormally mutated due to uncontrollable factors such as natural disasters such as earthquakes, torrential floods, debris flows, typhoons and frost or equipment faults; the historical data of the power load refers to the value of the power of the load sampled at a uniform moment from a period of time before an emergency happens to the time when the power load is completely recovered to be normal;
step 2.1, using a support vector machine to carry out regression prediction on historical data x of the load to be predicted 0 M sets of history data { x) of the same type as the power load history data with the classification label i Carrying out endpoint continuation on | i ═ 1, 2.., m } to obtain a time sequence y to be predicted after continuation 0 And historical time series y i 1, · m }; wherein x is i Representing the ith set of historical data, y i Representing the ith group of historical time series;
specifically, in the support vector machine regression prediction method, the constructed optimal decision function is as follows:
in the formula (1), w is a weight parameter,
for non-linear mapping, b is the threshold parameter.
By using a structural risk minimization criterion, introducing a non-negative relaxation variable and a dot product kernel function K (x, x'), and determining a support vector machine based on a regression algorithm by minimizing an objective function as follows:
in equation (2), f is an optimal decision function.
And introducing a Lagrange multiplier to construct a Lagrange functional to be used as a quadratic programming problem solution.
Step 2.2, treat the prediction time series y 0 And historical time series y i Adding a white noise sequence a with limited amplitude to obtain a processed time sequence y to be predicted a0 And processed historical time series y ai 1, · m }; wherein, y ai Representing the historical time sequence after the ith group of processing;
step 2.3, the processed time series y to be predicted is decomposed by using empirical mode a0 Decomposing the data into a basic trend item of the load to be predicted and high-frequency components of j loads to be predicted, and processing the processed historical time sequence { y ai 1,2, m } is decomposed into basic trend items of the m pieces of training historical data and high-frequency components of the m multiplied by j pieces of training historical data;
specifically, the empirical mode decomposition method is shown in fig. 2, and includes the following steps:
and 2.3.1, for any given signal X (t), firstly determining all extreme values on the X (t), connecting all the extreme values by a cubic spline curve to form an upper envelope line, and forming a lower envelope line by the same method. Mean m of signal X (t) and upper and lower envelope lines 1 Is recorded as h 1 Then, then
h 1 ＝X(t)-m 1 (3)
Repeating the above steps until the ith result h i When the two IMF conditions are satisfied, the IMF becomes the first-order IMF screened from the original signal, and is marked as C 1 . Typically the IMF component C of the first order 1 Containing the highest frequency part of the signal.
Step 2.3.2, adding C 1 Separating from the original signal X (t) to obtain a difference signal r with high frequency components removed 1 The method comprises the following steps:
r 1 ＝X(t)-C 1 (4)
handle r 1 As new signalsThe sieving step of 2.3.1 is repeated until the residual signal of the j-th order becomes a monotonic function and the IMF component can no longer be sieved.
r j ＝r j-1 -C j (5)
In the formula (5), r j As residual signal after the jth sieving, C j Is the j-th order residual signal.
Step 2.3.3, the original given signal x (t) can be expressed as the sum of j IMF components and one residual term, i.e.:
in formula (6): r is j The residual signal after the jth screening represents the average trend in the signal; each IMF component C k Respectively representing the components of different frequency bands of the signal from high to low;
step 2.4, after different white noise sequences are added each time, repeating the step 2.2 and the step 2.3 for n times, thereby obtaining n basic trend items of the load to be predicted and n multiplied by j high-frequency components of the load to be predicted, and n multiplied by m basic trend items of the historical data for training and n multiplied by m multiplied by j high-frequency components of the historical data for training;
step 2.5, solving the mean value of the basic trend items of the n loads to be predicted as a trend item sequence r to be predicted 0 Calculating the average value of the high-frequency components of the n multiplied by j loads to be predicted as a trend item sequence { r ] to be predicted i 1,2,.., m }, and calculating the mean value of the basic trend terms of the n × m training historical data as the high-frequency component sequence C to be predicted 0t The average of the high frequency components of n × m × j pieces of training history data is determined as a training high frequency component sequence { C it |i＝1,2,...,m；t＝1,2,...,j}。
Step 3, performing transfer learning based on attention mechanism model weight transfer by using the basic trend of the power load to obtain a prediction model of a trend item;
step 3.1, connecting an attention model containing a learning parameter q with a coder-decoder structure to form a sequence-to-sequence learning network;
specifically, the attention mechanism model is a dot product model, represented by the following equation:
s(d,q)＝d T q (8)
in the expressions (7) to (9), d is an input sequence, q is a query vector, s (d, q) is an attention scoring function of a dot product model, and is calculated by the expression (8), and α is an attention distribution, and is calculated by the expression (9).
Step 3.2, using the trend item sequence to be predicted { r i Taking | i ═ 1,2,. and m } as training data, and repeatedly correcting the learning parameter q by an Adam algorithm until errors meet requirements, so as to obtain a trained migration model;
specifically, the Adam algorithm is an improved gradient descent method, is a combination of the RMSProp algorithm and a momentum method, uses momentum as a parameter to update the direction, and can adaptively adjust the learning rate. The parameter updating calculation formula is as follows:
M＝β 1 M′+(1-β 1 )g (13)
G＝β 2 G+(1-β 2 )g⊙g (14)
in the formula (10) to the formula (14), beta 1 And beta 2 Respectively, the attenuation rates of two moving averages, usually taken as beta 1 ＝0.9，β 2 0.99; g is the gradient vector, and M and G can be considered as the mean (first moment) and the variance (second moment) of the non-subtracted mean of the gradient, respectively.
Step 3.3, connecting the migration model with the LSTM network as a trend item sequence r to be predicted 0 The predictive model of (1).
As shown in fig. 3, the mathematical model of the LSTM network is shown as follows:
f t ＝σ(W f ·[h t-1 ,x t ]+b f ) (15)
g t ＝σ(W g ·[h t-1 ,x t ]+b i ) (16)
s t ＝tanh(W c ·[h t-1 ,x t ]+b c ) (17)
c t ＝f t ⊙c t-1 +g t ⊙s t (18)
o t ＝σ(W o ·[h t-1 ,x t ]+b o ) (19)
h t ＝o t ⊙tanh(c t ) (20)
y t ＝softmax(h t ) (21)
formula (15) to formula (21): x is the number of t For input, f t To forget the gate value, g t For inputting a gate value, s t Is a temporary state quantity, c t Is a state quantity, o t For intermediate output, h t To output a gate value, y t Is an output; w is a group of f ，W g ，W c ，W o Are respectively corresponding weight matrix, b f ，b g ，b c ，b o Respectively, corresponding offset entries, indicating array multiplication, i.e., multiplication of elements of the vector. Sigma is a sigmoid activation function, and specific expressions of the sigmoid activation function and the tanh activation function and the softmax activation function are shown as the following formulas:
and 4, respectively carrying out load prediction on the basic trend of the power load and a plurality of high-frequency components by using the prediction model of the trend item and the LSTM network, and superposing the prediction results to obtain a load prediction result.
Step 4.1, a trend item sequence r to be predicted 0 Inputting the data serving as training data into a prediction model, training parameters of an LSTM network in the prediction model through an Adam algorithm, and predicting future data of a basic trend of the power load;
step 4.2, using another LSTM network to respectively predict j high-frequency component sequences C to be predicted 0t Predicting to obtain future data of j high-frequency component sequences;
and 4.3, adding the future data of the basic trend of the power load and the future data of the j high-frequency component sequences to obtain a load prediction result to be predicted.
Claims (2)
1. A method for predicting power load in an extreme scene based on modal decomposition and transfer learning is characterized by comprising the following steps:
step 1, counting historical data of power loads in various extreme scenes, and classifying the historical data according to the trend of a load curve to obtain historical data of the power loads with classification labels; then collecting historical data under a scene to be predicted;
step 2, decomposing historical data under a scene to be predicted and historical data of the same type in the historical data of the power load with the classification label by using an improved aggregation empirical mode decomposition method to obtain a basic trend and a plurality of high-frequency components of the power load;
step 2.1,Regression prediction of historical data x of load to be predicted by using support vector machine 0 M groups of historical data { x) of the same type as the power load historical data with classification labels i Carrying out endpoint continuation on | i ═ 1, 2.., m } to obtain a time sequence y to be predicted after continuation 0 And historical time series y i 1, · m }; wherein x is i Representing the ith set of historical data, y i Representing the ith group of historical time series;
step 2.2, time series y to be predicted 0 And historical time series y i Adding a white noise sequence a with limited amplitude to obtain a processed time sequence y to be predicted a0 And processed historical time series y ai 1, · m }; wherein, y ai Representing the historical time series after the ith group of processing;
step 2.3, the processed time series y to be predicted is decomposed by using empirical mode a0 Decomposing the data into a basic trend item of the load to be predicted and high-frequency components of j loads to be predicted, and processing the processed historical time sequence { y ai Decomposing | i ═ 1, 2.. multidot.m } into m basic trend terms of the training history data and m × j high-frequency components of the training history data;
step 2.4, after adding different white noise sequences each time, repeating the step 2.2 and the step 2.3 for n times, thereby obtaining n basic trend items of the load to be predicted and n multiplied by j high frequency components of the load to be predicted, and n multiplied by m basic trend items of the historical data for training and n multiplied by m multiplied by j high frequency components of the historical data for training;
step 2.5, calculating the average value of the n basic trend items of the load to be predicted and taking the average value as a trend item sequence r to be predicted 0 Calculating the average value of the high-frequency components of the n multiplied by j loads to be predicted and taking the average value as a trend item sequence { r ] to be predicted i 1,2, a.m., calculating the mean value of the basic trend terms of the n × m training history data and using the mean value as the high-frequency component sequence C to be predicted 0t Calculating the average of the high frequency components of n × m × j pieces of training history data as a training high frequency component sequence { C it |i＝1,2,...,m；t＝1,2,...,j}；
Step 3, performing transfer learning based on attention mechanism model weight transfer by using the basic trend of the power load to obtain a prediction model of a trend item;
step 3.1, connecting an attention model containing a learning parameter q with a coder-decoder structure to form a sequence-to-sequence learning network;
step 3.2, using the trend item sequence { r ] to be predicted i Taking | i ═ 1,2,. and m } as training data, and repeatedly correcting the learning parameter q by an Adam algorithm until errors meet requirements, so as to obtain a trained migration model;
step 3.3, connecting the migration model with an LSTM network as a trend item sequence r to be predicted 0 The predictive model of (2);
and 4, respectively carrying out load prediction on the basic trend and the high-frequency components of the power load by using the prediction model of the trend item and the LSTM network, and superposing prediction results to obtain a load prediction result.
2. The method for predicting the power load in the extreme scene according to claim 1, wherein the step 4 comprises:
step 4.1, a trend item sequence r to be predicted 0 Inputting the data into the prediction model as training data, training parameters of an LSTM network in the prediction model through an Adam algorithm, and predicting future data of a basic trend of the power load;
step 4.2, using another LSTM network to respectively predict j high-frequency component sequences C to be predicted 0t Predicting to obtain future data of j high-frequency component sequences;
and 4.3, adding the future data of the basic trend of the power load and the future data of the j high-frequency component sequences to obtain a load prediction result to be predicted.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011389811.7A CN112488397B (en) | 2020-12-01 | 2020-12-01 | Load prediction method under extreme scene based on modal decomposition and transfer learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011389811.7A CN112488397B (en) | 2020-12-01 | 2020-12-01 | Load prediction method under extreme scene based on modal decomposition and transfer learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112488397A CN112488397A (en) | 2021-03-12 |
| CN112488397B true CN112488397B (en) | 2022-09-27 |
Family
ID=74938955
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011389811.7A Active CN112488397B (en) | 2020-12-01 | 2020-12-01 | Load prediction method under extreme scene based on modal decomposition and transfer learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112488397B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113139743A (en) * | 2021-05-12 | 2021-07-20 | 平安国际智慧城市科技股份有限公司 | Sewage discharge index analysis method and device, electronic equipment and storage medium |
| CN114675598B (en) * | 2022-03-29 | 2024-06-25 | 中南大学 | Method and system for predicting tool nose modal parameters of different numerical control machine tools based on transfer learning |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104794538A (en) * | 2015-04-20 | 2015-07-22 | 浙江大学 | Power system short-term load prediction method based on IEEMD-BPNN (improved ensemble empirical mode decomposition-back propagation neutral network) |
| CN110059844A (en) * | 2019-02-01 | 2019-07-26 | 东华大学 | Energy storage device control method based on set empirical mode decomposition and LSTM |
| CN110070226A (en) * | 2019-04-24 | 2019-07-30 | 河海大学 | Photovoltaic power prediction technique and system based on convolutional neural networks and meta learning |
| CN111191841A (en) * | 2019-12-30 | 2020-05-22 | 润联软件系统(深圳)有限公司 | Power load prediction method and device, computer equipment and storage medium |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120020216A1 (en) * | 2010-01-15 | 2012-01-26 | Telcordia Technologies, Inc. | Cognitive network load prediction method and apparatus |
-
2020
- 2020-12-01 CN CN202011389811.7A patent/CN112488397B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104794538A (en) * | 2015-04-20 | 2015-07-22 | 浙江大学 | Power system short-term load prediction method based on IEEMD-BPNN (improved ensemble empirical mode decomposition-back propagation neutral network) |
| CN110059844A (en) * | 2019-02-01 | 2019-07-26 | 东华大学 | Energy storage device control method based on set empirical mode decomposition and LSTM |
| CN110070226A (en) * | 2019-04-24 | 2019-07-30 | 河海大学 | Photovoltaic power prediction technique and system based on convolutional neural networks and meta learning |
| CN111191841A (en) * | 2019-12-30 | 2020-05-22 | 润联软件系统(深圳)有限公司 | Power load prediction method and device, computer equipment and storage medium |
Non-Patent Citations (3)
| Title |
|---|
| Power Load Prediction Method Based on VMD and Dynamic Adjustment BP;Fengtian Kuang;《2019 CAA Symposium on Fault Detection, Supervision and Safety for Technical Processes (SAFEPROCESS)》;20201006;全文 * |
| SOM特征提取与ELM在电力负荷预测中的应用;马立新等;《电力科学与工程》;20150528(第05期);全文 * |
| 基于双图正则非负低秩分解的电力负荷短期预测;梁寿愚等;《计算机与现代化》;20200515(第05期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112488397A (en) | 2021-03-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wang et al. | A self-adaptive hybrid approach for wind speed forecasting | |
| CN112488397B (en) | Load prediction method under extreme scene based on modal decomposition and transfer learning | |
| CN110445646B (en) | Resource reservation method based on attention mechanism and flow prediction model | |
| CN110348624B (en) | Sand storm grade prediction method based on Stacking integration strategy | |
| WO2024077969A1 (en) | Lstm-svr subway station temperature prediction method based on characteristic of multiple periods | |
| CN103577694B (en) | Aquaculture water quality short-time combination forecast method on basis of multi-scale analysis | |
| CN111652425B (en) | River water quality prediction method based on rough set and long-short-term memory network | |
| CN104951836A (en) | Posting predication system based on nerual network technique | |
| CN108022014B (en) | Power system load prediction method and system | |
| CN107993012B (en) | Time-adaptive online transient stability evaluation method for power system | |
| CN109784692B (en) | Rapid safety constraint economic dispatching method based on deep learning | |
| CN112396234A (en) | User side load probability prediction method based on time domain convolutional neural network | |
| CN112232604A (en) | Prediction method for extracting network traffic based on Prophet model | |
| CN113360848A (en) | Time sequence data prediction method and device | |
| CN107145968A (en) | Photovoltaic apparatus life cycle cost Forecasting Methodology and system based on BP neural network | |
| CN116028315A (en) | Operation early warning method, device, medium and electronic equipment | |
| Li et al. | A deep learning framework for non-stationary time series prediction | |
| Parmar et al. | A Comparative study on different ANN techniques in wind speed forecasting for generation of electricity | |
| CN110991729A (en) | Load prediction method based on transfer learning and multi-head attention mechanism | |
| CN114648217A (en) | Coal mine safety risk early warning method based on incremental extreme learning machine | |
| CN113627685A (en) | Wind power generator power prediction method considering wind power on-grid load limitation | |
| CN104077231B (en) | Transformer maintenance optimization method based on symbol dynamics and LS-SVM | |
| CN116993002A (en) | Load electric quantity combination prediction method and device for realizing same | |
| Albelwi | A Robust Energy Consumption Forecasting Model using ResNet-LSTM with Huber Loss | |
| CN116468333A (en) | Coal port air quality prediction method based on data decomposition and deep learning |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |