CN116432860A - Short-term power load prediction method, system and medium based on EWT-TCN - Google Patents
Short-term power load prediction method, system and medium based on EWT-TCN Download PDFInfo
- Publication number
- CN116432860A CN116432860A CN202310518198.1A CN202310518198A CN116432860A CN 116432860 A CN116432860 A CN 116432860A CN 202310518198 A CN202310518198 A CN 202310518198A CN 116432860 A CN116432860 A CN 116432860A
- Authority
- CN
- China
- Prior art keywords
- tcn
- power load
- ewt
- prediction
- network
- 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 53
- 238000012549 training Methods 0.000 claims abstract description 49
- 238000012795 verification Methods 0.000 claims abstract description 30
- 238000012360 testing method Methods 0.000 claims abstract description 28
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 26
- 238000010606 normalization Methods 0.000 claims abstract description 19
- 238000001228 spectrum Methods 0.000 claims abstract description 15
- 238000007781 pre-processing Methods 0.000 claims abstract description 14
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 15
- 238000000354 decomposition reaction Methods 0.000 claims description 15
- 230000008569 process Effects 0.000 claims description 12
- 230000001364 causal effect Effects 0.000 claims description 9
- 239000013598 vector Substances 0.000 claims description 9
- 230000006870 function Effects 0.000 claims description 8
- 238000012216 screening Methods 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 5
- 238000001914 filtration Methods 0.000 claims description 4
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 238000004590 computer program Methods 0.000 claims description 3
- 238000010200 validation analysis Methods 0.000 claims 1
- 230000008859 change Effects 0.000 abstract description 5
- 238000004364 calculation method Methods 0.000 description 5
- 238000013527 convolutional neural network Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 230000014509 gene expression Effects 0.000 description 3
- 230000010339 dilation Effects 0.000 description 2
- 238000004880 explosion Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 238000000611 regression analysis Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 230000009466 transformation Effects 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
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Economics (AREA)
- General Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- Game Theory and Decision Science (AREA)
- Water Supply & Treatment (AREA)
- Public Health (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Development Economics (AREA)
- Primary Health Care (AREA)
- Quality & Reliability (AREA)
- Power Engineering (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a short-term power load prediction method, a short-term power load prediction system and a short-term power load prediction medium based on EWT-TCN, which comprise the following steps: acquiring power load information, and preprocessing the power load information; according to the preprocessed frequency spectrum of the power load information, decomposing the power load information into N subsequences by using an EWT algorithm, and dividing the N subsequences into a training set, a verification set and a test set; constructing a TCN (TCN) network, inputting training set and verification set data into the TCN network for training and verification, and obtaining optimized TCN network weight parameters; inputting the test set into the trained TCN network to obtain each subsequence prediction result; reconstructing the prediction results of the subsequences through an EWT algorithm, and performing inverse normalization to obtain the prediction results. The invention has accurate and reliable prediction results on the change of the load demand and the load characteristic, and plays an important role in a plurality of departments of an actual power system.
Description
Technical Field
The invention relates to the technical field of short-term power prediction, in particular to a short-term power load prediction method, a short-term power load prediction system and a short-term power load prediction medium based on EWT-TCN.
Background
The power load prediction is to search the influence of the change rule of the power load history data on the future load according to the history data of the power load, society, weather and the like, and search the inherent relation between the power load and various related factors so as to scientifically predict the future power load. Load prediction plays an important role in many departments of power systems. For example, a short-term load prediction using a daily load curve as a prediction target is a basis for making a daily power generation plan. For this reason, in power system planning, operation management, and power market trading, it is necessary to have an accurate prediction of the load demand variation and load characteristics.
However, the power load information is affected by many factors, and has the characteristics of strong nonlinearity, randomness and the like, which greatly increases the difficulty of prediction. The existing load prediction methods are mainly divided into three types, namely a statistical-based method, a machine learning-based method and a deep learning-based method. The traditional statistical method mainly comprises a time sequence method and a regression analysis method, and the method has the advantages of simple structural form, high calculation speed and insufficient precision, and has good prediction capability for stable and uniform change data. With the development of machine learning technology, a new method is brought to load prediction, such as common algorithms of support vector machines, XGboost, random forests, BP neural networks and the like, and the accuracy of the algorithms is improved to a certain extent, but the algorithms are shallow in structure and suitable for tasks with smaller data volume, and the accuracy and stability of the algorithms are difficult to be ensured when millions of load data of modern society are faced.
Disclosure of Invention
In order to solve the problems, the technical scheme provided by the invention is as follows:
a short-term power load prediction method based on EWT-TCN, comprising:
acquiring power load information, and preprocessing the power load information;
according to the preprocessed frequency spectrum of the power load information, decomposing the power load information into N subsequences by using an EWT algorithm, and dividing the N subsequences into a training set, a verification set and a test set;
constructing a TCN (TCN) network, inputting training set and verification set data into the TCN network for training and verification, and obtaining optimized TCN network weight parameters;
inputting the test set into the trained TCN network to obtain each subsequence prediction result;
reconstructing the prediction results of the subsequences through an EWT algorithm, and performing inverse normalization to obtain the prediction results.
The invention is further arranged for preprocessing said electrical load information comprising:
traversing the acquired power load information, and complementing the missing value;
converting the digital type of the complemented power load information;
feature screening is carried out on the digitized power load information according to the Pearson correlation coefficient, and influence factors with the number or the correlation degree meeting the requirements are selected;
and normalizing the influence factors obtained by screening.
The invention is further configured to decompose the power load information into N subsequences by using an EWT algorithm according to the preprocessed spectrum of the power load information, and dividing the N subsequences into a training set, a verification set and a test set includes:
performing Fourier transform on the power load original signal, finding out the number of maximum points on a signal frequency spectrum, and recording the number as N;
dividing the frequency domain into N consecutive frequency band intervals within the frequency range of [0, pi ];
constructing a filter bank adapted to process the power load raw signal;
filtering by using the filter bank to obtain N group components;
superposing the corresponding influence factors on each component to obtain N subsequences;
dividing the N subsequences according to a preset proportion to obtain a training set, a verification set and a test set.
The invention further provides that the constructing the TCN network, inputting the training set and the verification set data into the TCN network for training verification, and obtaining optimized TCN network weight parameters comprises:
the TCN network comprises a plurality of TCN residual blocks which are stacked, each TCN residual block comprises two groups of causal expansion convolution layers, a weight normalization layer, a ReLU activation layer and a Dropout regularization layer which are sequentially connected, and the output of the former TCN residual block is formed by adding data after 1*1 convolution and two groups of operations, and the input of the latter TCN residual block is formed;
setting parameters of a TCN network; inputting a training set and a verification set into the TCN network for training and verification to obtain the TCN network optimal weight parameters corresponding to each component, and respectively marking the TCN parameters as TCN 1 -TCN N 。
The invention further provides that the obtaining of the optimal weight parameters in the TCN network comprises the following steps:
initializing parameters and organizing data, calculating a predicted value through forward propagation within preset training times, comparing the difference between the predicted value and a true value, updating the parameters through reverse propagation until the preset training times are reached, ending training, and outputting the TCN network optimal weight parameters.
The invention is further arranged that an optimizer selects Adam, learning rate 0.001, loss function MAE and training round number 500 in the TCN network training process; the TCN network is then provided with a full connectivity layer to adjust the output length.
The invention further provides that the inputting the test set into the trained TCN network to obtain each subsequence prediction result comprises:
reconstructing a TCN network, loading the TCN network optimal weight parameters to obtain N TCN networks, inputting the N groups of subsequences of the test set into the corresponding TCN networks, and obtaining the prediction results of each group of the test set.
The invention further provides that the reconstructing the predicted result of each subsequence by using the EWT algorithm, and performing inverse normalization to obtain the predicted result comprises:
and superposing the prediction results of each group of the test set, reconstructing according to an EWT algorithm to obtain complete prediction data, and performing inverse normalization on the reconstructed prediction data to obtain the prediction result of the power load.
The short-term power load prediction system based on the EWT-TCN adopts the short-term power load prediction method based on the EWT-TCN, and comprises the following steps:
the data preprocessing module is used for preprocessing the collected power load information;
the EWT decomposition module is used for decomposing the original data, and each component after decomposition and relevant influence factors form an input vector of the TCN network;
the TCN prediction module predicts the input vector to obtain a prediction result corresponding to the component;
and the EWT reconstruction module is used for reconstructing a prediction result corresponding to the component and performing inverse normalization to obtain the prediction result of the power load.
A storage medium having stored thereon computer program instructions which, when executed by a processor, implement the EWT-TCN based short-term power load prediction method described above.
Compared with the prior art, the technical scheme provided by the invention has the following beneficial effects:
according to the technical scheme, the short-term power load prediction method based on the EWT decomposition and the TCN network utilizes the EWT decomposition of power load data, fully excavates the intrinsic information of the data in the time domain and the frequency domain, effectively solves the problem of prediction difficulty caused by the nonlinear randomness of the original data, and adds relevant influence factors on the power load data to form an input vector of the TCN network; the TCN network is constructed to predict, the advantages of stable gradient of the convolutional neural network, parallel calculation, operation speed improvement and the like are fully exerted, efficiency is improved, and prediction precision is improved. The short-term power load prediction method based on the EWT-TCN has accurate and reliable prediction results on the change of the load demand and the load characteristic, and plays an important role in a plurality of departments of an actual power system.
Drawings
Fig. 1 is a flowchart of a short-term power load prediction method based on EWT-TCN according to an embodiment of the present invention.
FIG. 2 is an exploded view of an EWT according to an embodiment of the present invention.
Fig. 3 is a schematic diagram of input vectors of a TCN network according to an embodiment of the present invention.
FIG. 4 is a schematic diagram of a one-dimensional causal dilation convolution of an embodiment of the present invention.
Fig. 5 is a schematic diagram of residual connection in an embodiment of the present invention.
Fig. 6 is a flowchart of training the TCN network optimal weight parameters according to an embodiment of the present invention.
Fig. 7 is a schematic diagram of a short-term power load prediction system based on EWT-TCN according to an embodiment of the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Example 1
Referring to fig. 1, the technical scheme of the invention is a short-term power load prediction method based on EWT-TCN, comprising:
s100, acquiring power load information, and preprocessing the power load information;
s200, according to the preprocessed frequency spectrum of the power load information, decomposing the power load information into N subsequences by using an EWT algorithm, and dividing the N subsequences into a training set, a verification set and a test set;
s300, constructing a TCN (TCN network), inputting training set and verification set data into the TCN network for training and verification, and obtaining optimized TCN network weight parameters;
s400, inputting the test set into the trained TCN network to obtain each subsequence prediction result;
s500, reconstructing the prediction results of the subsequences through an EWT algorithm, and performing inverse normalization to obtain the prediction results.
The invention discloses a short-term power load prediction method based on an EWT decomposition and TCN network, which utilizes the EWT decomposition of power load data to fully mine the intrinsic information of the data in the time domain and the frequency domain, and simultaneously effectively solves the problem of prediction difficulty caused by the nonlinear randomness of the original data, and adds relevant influence factors on the power load data to form an input vector of the TCN network; the TCN network is constructed to predict, the advantages of stable gradient of the convolutional neural network, parallel calculation, operation speed improvement and the like are fully exerted, efficiency is improved, and prediction precision is improved.
In the above embodiment, the EWT algorithm is a non-stationary signal processing method that integrates the adaptive decomposition concept and the wavelet variation theory, and the kernel of the EWT algorithm is to perform fourier transform on an original signal, then divide the transformed spectrum into a series of continuous intervals, construct a wavelet filter bank according to different intervals to perform filtering, and finally obtain a set of amplitude modulation and frequency modulation components. Assuming a signal f (t), the decomposition process is as follows:
firstly, dividing frequency bands, carrying out Fourier transform on an original signal, and normalizing the frequency spectrum of the original signal to [0, pi ] according to shannon criterion]. Then, a frequency band division boundary is calculated according to the angular frequency omega N (n=1, 2,) corresponding to the first N maximum value points on the signal spectrumThen pass through w n Dividing the spectrum into N consecutive intervals, w 0 =0,w N Pi. Each section is represented as follows,
to facilitate the construction of the filter, take w n Define a width T for the center n =2τ n Wherein
Then constructing a filter bank, and after division of the intervals, dividing lambda according to Meyer wavelet basis pairs n The interval constructs an empirical wavelet function ψn (w) and an empirical scale function phin (w), expressed as follows,
wherein the arbitrary function β (x) is defined as follows:
according to the classical wavelet transform method, the detail coefficients and approximation coefficients of the EWT algorithm can be defined as follows,
wherein the method comprises the steps ofRespectively is psi n (w) and phi 1 A complex conjugated form of (w); f (F) -1 [·]Is an inverse fourier transform. After EWT processing, the expressions of the components after signal decomposition are as follows,
finally, the signal is reconstructed, the reconstruction expression is as follows,
in the above embodiments, the TCN network is a new network architecture modified from CNN. Unlike traditional CNN networks, which are good at the field of picture processing, TCN is also applicable to the field of time series prediction by introducing one-dimensional causal expansion convolutional layer, residual block and other structures.
One-dimensional causal convolution means that the convolution kernel moves only in one direction, while for each layer t-moment value only depends on the previous layer t-moment and its preceding values, which is a strict time constraint model. As shown in fig. 4, the layer 1 is input to the hidden layer, which is a one-dimensional causal convolution with a convolution kernel size of k=1×3. And since the lengths of the input and output are kept uniform, the gray part is filled with 0. Due to the limitation of the convolution kernel, the simple one-dimensional causal convolution only stacks more layers to grasp longer historical information, which causes the problems of gradient explosion, disappearance and the like, and the increase of the parameters requires longer time to train. For this purpose,the length of the grabbing history information is increased while the number of network layers is kept relatively small, and expansion convolution is introduced. Expansion refers to the distance between two input elements, also referred to as the expansion coefficient, denoted by d. The coefficient of expansion of the input layer is 1 in fig. 4, i.e. each element is grabbed, and the coefficient of expansion of the hidden layer 1 is 2, i.e. every second element is grabbed. The expansion coefficient will generally increase exponentially with increasing number of layers, 2 being generally chosen as the base, so the expansion coefficient of the nth layer is d=2 n-1 。
The TCN network is stacked from a plurality of residual blocks. As shown in fig. 5, one residual block consists of two one-dimensional causal dilation convolutional layers. In order to not overly complicate the network model, it is necessary to add a ReLU activation function over the two convolutional layers, respectively, to introduce nonlinearities. To counteract the gradient explosion, etc., the input to the hidden layer is normalized, and a weight normalization is added after each convolution layer. To prevent the over-fitting problem, dropout regularization is introduced after each convolutional layer. Finally, to prevent the influence of different input and output channel widths on the residual network, a 1*1 convolution is introduced to the right in the figure to adjust the width of the residual tensor.
In this embodiment, preprocessing the power load information includes:
traversing the acquired power load information, and complementing the missing value;
converting the digital type of the complemented power load information;
feature screening is carried out on the digitized power load information according to the Pearson correlation coefficient, and influence factors with the number or the correlation degree meeting the requirements are selected;
and normalizing the influence factors obtained by screening.
In the above embodiment, the power load information is obtained by sampling power load data at certain intervals, for example, 96 power load data points in a day with 15 minutes as sampling intervals; meanwhile, meteorological factors are also required to be collected as important factors influencing the power load data, and comprise the highest, lowest, average temperature, relative humidity, rainfall data and the like.
Secondly, preprocessing the power load information and the meteorological factors, traversing the power load information and the meteorological factors, and if the missing value is found, complementing the missing value a by adopting a mean filling method i Let the
Some data, such as the type of week, as a big factor affecting the power load data, cannot be directly input into the TCN network and therefore needs to be converted to a digital type. Averaging the load data corresponding to each week type to form matrix W 1 W 2 W 3 W 4 W 5 W 6 W 7 ]Find the maximum value of which is marked as W max According toConversion of the digital type is performed. In other embodiments, other factors that affect the power load data may be used, such as morning, noon, evening time periods.
Because the acquired meteorological factors are too complex, in order to reduce the complexity of the model and avoid the influence of factors with weak correlation on the prediction result, the data needs to be screened, and the factors with strong correlation are reserved. The invention adopts the pearson correlation coefficient to screen the factors. The pearson correlation coefficient is used to measure the degree of correlation between two variables, denoted as r, and its value is between-1 and 1, with a greater absolute value indicating a stronger correlation. Assuming that two variables are X, Y, r is defined as follows:
it is generally considered that: when the r is more than or equal to 0.8, the two variables can be considered to be highly related; the two variables are considered to be moderately related by 0.5-r < 0.8; the two variables are considered to be low-correlated by 0.3 < r < 0.5; the two variables can be considered to be substantially uncorrelated. For this purpose, 0.3 is used as a threshold value, and the influence factors with |r| equal to or larger than 0.3 are screened out.
Finally, the data value is limited to [0,1 ] through normalization]The calculation formula is as followsWherein: x is x i ' is x i Normalized value, x max And x min Respectively representing peaks and valleys.
In this embodiment, according to the preprocessed spectrum of the power load information, decomposing the power load information into N subsequences by using an EWT algorithm, and dividing the N subsequences into a training set, a verification set and a test set includes:
performing Fourier transform on the power load original signal, finding out the number of maximum points on a signal frequency spectrum, and recording the number as N;
dividing the frequency domain into N consecutive frequency band intervals within the frequency range of [0, pi ];
constructing a filter bank adapted to process the power load raw signal;
filtering by using the filter bank to obtain N group components;
superposing the corresponding influence factors on each component to obtain N subsequences;
dividing the N subsequences according to a preset proportion to obtain a training set, a verification set and a test set.
In the above embodiment, as shown in fig. 2, the decomposition is performed in the form of a sliding window. First, the data of the previous T days is decomposed (1 day is taken by T) into N components, which are respectively marked as IMF1-IMFn, and the load of the T+1 days is predicted by inputting a model. The data were then decomposed for days 1 to t+1, and predicted for t+2. As known from the EWT algorithm principle, the determination of N depends on the number of maximum points on the spectrum, and for this purpose, the original signal is first subjected to fourier transformation, and the number of maximum points is found and recorded as N.
After the decomposition of the original power load data is completed, N groups of samples can be obtained, then input samples are constructed, each sample is composed of components plus the influence factors of the same day, as shown in fig. 3, N groups of samples are obtained, each group of samples is independently trained, data sets are divided according to the proportion, wherein the data sets are divided according to the ratio of 7:2:1, namely, training sets account for 70% of the total data, verification sets account for 20% of the total data, and test sets account for 10% of the total data.
In other embodiments, the ratio of the data set divisions may be other, such as 3:1:1 or 8:1:1, etc.
In this embodiment, the constructing a TCN network, inputting training set and verification set data into the TCN network for training verification, and obtaining optimized TCN network weight parameters includes:
the TCN network comprises a plurality of TCN residual blocks which are stacked, each TCN residual block comprises two groups of causal expansion convolution layers, a weight normalization layer, a ReLU activation layer and a Dropout regularization layer which are sequentially connected, and the output of the former TCN residual block is formed by adding data after 1*1 convolution and two groups of operations, and the input of the latter TCN residual block is formed;
setting parameters of a TCN network; inputting a training set and a verification set into the TCN network for training and verification to obtain the TCN network optimal weight parameters corresponding to each component, and respectively marking the TCN parameters as TCN 1 -TCN N 。
In the above embodiment, the TCN network expansion coefficient is selected from [1,2,4,8,16,32], the convolution kernel size is 1×5, and Dropout0.1. Because the TCN network has the same input and output length, the input length is 96+ and the number of the influence factors after screening, and the output length is 96 which is a daily load curve, in order to realize the process, a full connection layer is added behind the TCN network so as to adjust the output length, and the sigmoid is selected by the activation function of the full connection layer.
In this embodiment, as shown in fig. 6, the obtaining of the optimal weight parameter in the TCN network includes:
initializing parameters and organizing data, calculating predicted values by forward propagation within preset training times, comparing differences between the predicted values and the actual values, and further by backward propagationAnd finishing training until the preset training times are reached, and outputting the TCN network optimal weight parameters. For N groups of subsequences, N TCN network models are needed to correspond to the N groups of subsequences, a first group of subsequences are input into a network, and TCN network optimal weight parameters corresponding to the first group of component are obtained after training and recorded as TCN 1 The method comprises the steps of carrying out a first treatment on the surface of the Inputting the second group of subsequences into the network, training to obtain TCN network optimal weight parameters corresponding to the second group of components, and marking the TCN network optimal weight parameters as TCN 2 The method comprises the steps of carrying out a first treatment on the surface of the And so on to obtain N groups of optimal weight parameters which are respectively marked as TCN 1 -TCN N 。
In this embodiment, the optimizer selects Adam, learning rate 0.001, loss function MAE, training round number 500 in the TCN network training process; the TCN network is then provided with a full connectivity layer to adjust the output length.
In this embodiment, the inputting the test set into the trained TCN network to obtain each subsequence prediction result includes:
reconstructing a TCN network, loading the TCN network optimal weight parameters to obtain N TCN networks, inputting the N groups of subsequences of the test set into the corresponding TCN networks, and obtaining the prediction results of each group of the test set.
In the above embodiment, the TCN network is reconstructed, the same expansion coefficient is set as that of the TCN network, the expansion coefficient is selected to be 1×5, dropout0.1, and then the optimal weight parameters are loaded to obtain N TCN networks, and each group of subsequences of the test set is input into the corresponding TCN network, so as to obtain the prediction result of each group.
In this embodiment, reconstructing the predicted result of each subsequence by using an EWT algorithm, and performing inverse normalization to obtain the predicted result includes:
and superposing the prediction results of each group of the test set, reconstructing according to an EWT algorithm to obtain complete prediction data, and performing inverse normalization on the reconstructed prediction data to obtain the prediction result of the power load.
In the above embodiment, after each sub-sequence prediction result is obtained, it is necessary to recombine them together, thereby obtaining power load data that needs to be predicted. According to the EWT algorithm, the reconstruction expression of the signal f (t) is:
in short, the components are overlapped, and a complete prediction result can be obtained after reconstruction. Finally, carrying out inverse normalization on the reconstructed data to obtain a final prediction result, wherein the calculation formula is x i =x′ i (x max -x min ) Wherein: x's' i Is x i Normalized value, x max And x min Respectively representing peaks and valleys.
The short-term power load prediction method based on the EWT-TCN has accurate and reliable prediction results on the change of the load demand and the load characteristic, and plays an important role in guiding a plurality of departments of an actual power system.
Example 2
Referring to fig. 7, the technical scheme of the invention is a short-term power load prediction system based on EWT-TCN, and the short-term power load prediction method based on EWT-TCN according to embodiment 1 is adopted, which comprises the following steps:
the data preprocessing module is used for preprocessing the collected power load information;
the EWT decomposition module is used for decomposing the original data, and each component after decomposition and relevant influence factors form an input vector of the TCN network;
the TCN prediction module predicts the input vector to obtain a prediction result corresponding to the component;
and the EWT reconstruction module is used for reconstructing a prediction result corresponding to the component and performing inverse normalization to obtain the prediction result of the power load.
When the short-term power load prediction system based on the EWT-TCN is implemented, a process of establishing a prediction model according to the recent power load data is provided, and the recent data is required to be trained to obtain the prediction model suitable for the current scene.
Example 3
The technical solution of the present invention is a storage medium, on which computer program instructions are stored, which when executed by a processor implement the short-term electric load prediction method based on EWT-TCN as described in embodiment 1.
In the description, each embodiment is described in a progressive manner, and each embodiment is mainly described by the differences from other embodiments, so that the same similar parts among the embodiments are mutually referred.
It should also be noted that in this specification, relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (10)
1. A short-term power load prediction method based on EWT-TCN, comprising:
acquiring power load information, and preprocessing the power load information;
according to the preprocessed frequency spectrum of the power load information, decomposing the power load information into N subsequences by using an EWT algorithm, and dividing the N subsequences into a training set, a verification set and a test set;
constructing a TCN (TCN) network, inputting training set and verification set data into the TCN network for training and verification, and obtaining optimized TCN network weight parameters;
inputting the test set into the trained TCN network to obtain each subsequence prediction result;
reconstructing the prediction results of the subsequences through an EWT algorithm, and performing inverse normalization to obtain the prediction results.
2. The EWT-TCN based short-term power load prediction method of claim 1, wherein preprocessing the power load information comprises:
traversing the acquired power load information, and complementing the missing value;
converting the digital type of the complemented power load information;
feature screening is carried out on the digitized power load information according to the Pearson correlation coefficient, and influence factors with the number or the correlation degree meeting the requirements are selected;
and normalizing the influence factors obtained by screening.
3. The EWT-TCN based short-term power load prediction method of claim 2, wherein decomposing the power load information into N subsequences using an EWT algorithm according to the pre-processed spectrum of the power load information, and dividing the N subsequences into a training set, a validation set, and a test set comprises:
performing Fourier transform on the power load original signal, finding out the number of maximum points on a signal frequency spectrum, and recording the number as N;
dividing the frequency domain into N consecutive frequency band intervals within the frequency range of [0, pi ];
constructing a filter bank adapted to process the power load raw signal;
filtering by using the filter bank to obtain N group components;
superposing the corresponding influence factors on each component to obtain N subsequences;
dividing the N subsequences according to a preset proportion to obtain a training set, a verification set and a test set.
4. The method for short-term power load prediction based on EWT-TCN according to claim 3, wherein the constructing a TCN network, inputting training set and verification set data into the TCN network for training verification, and obtaining optimized TCN network weight parameters includes:
the TCN network comprises a plurality of TCN residual blocks which are stacked, each TCN residual block comprises two groups of causal expansion convolution layers, a weight normalization layer, a ReLU activation layer and a Dropout regularization layer which are sequentially connected, and the output of the former TCN residual block is formed by adding data after 1*1 convolution and two groups of operations, and the input of the latter TCN residual block is formed;
setting parameters of a TCN network; inputting a training set and a verification set into the TCN network for training and verification to obtain the TCN network optimal weight parameters corresponding to each component, and respectively marking the TCN parameters as TCN 1 -TCN N 。
5. The method for short-term power load prediction based on EWT-TCN according to claim 4, wherein the obtaining of the optimal weight parameters in the TCN network comprises:
initializing parameters and organizing data, calculating a predicted value through forward propagation within preset training times, comparing the difference between the predicted value and a true value, updating the parameters through reverse propagation until the preset training times are reached, ending training, and outputting the TCN network optimal weight parameters.
6. The short-term power load prediction method based on EWT-TCN according to claim 5, wherein the optimizer selects Adam, learning rate 0.001, loss function MAE, training round number 500 in the TCN network training process; the TCN network is then provided with a full connectivity layer to adjust the output length.
7. The method for short-term power load prediction based on EWT-TCN according to claim 6, wherein inputting the test set into the trained TCN network to obtain each subsequence prediction result comprises:
reconstructing a TCN network, loading the TCN network optimal weight parameters to obtain N TCN networks, inputting the N groups of subsequences of the test set into the corresponding TCN networks, and obtaining the prediction results of each group of the test set.
8. The method for short-term power load prediction based on EWT-TCN according to claim 7, wherein reconstructing each of the subsequences prediction results by EWT algorithm and performing inverse normalization to obtain the prediction results comprises:
and superposing the prediction results of each group of the test set, reconstructing according to an EWT algorithm to obtain complete prediction data, and performing inverse normalization on the reconstructed prediction data to obtain the prediction result of the power load.
9. An EWT-TCN based short-term power load prediction system, characterized in that it employs the EWT-TCN based short-term power load prediction method according to any one of claims 1 to 8, comprising:
the data preprocessing module is used for preprocessing the collected power load information;
the EWT decomposition module is used for decomposing the original data, and each component after decomposition and relevant influence factors form an input vector of the TCN network;
the TCN prediction module predicts the input vector to obtain a prediction result corresponding to the component;
and the EWT reconstruction module is used for reconstructing a prediction result corresponding to the component and performing inverse normalization to obtain the prediction result of the power load.
10. A storage medium having stored thereon computer program instructions which, when executed by a processor, implement the EWT-TCN based short-term power load prediction method of any one of claims 1 to 8.
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310467491X | 2023-04-25 | ||
CN202310467491 | 2023-04-25 |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116432860A true CN116432860A (en) | 2023-07-14 |
Family
ID=87094531
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310518198.1A Pending CN116432860A (en) | 2023-04-25 | 2023-05-09 | Short-term power load prediction method, system and medium based on EWT-TCN |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116432860A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116632841A (en) * | 2023-07-26 | 2023-08-22 | 国网山东省电力公司信息通信公司 | Power distribution area short-term electricity load prediction method and system integrating multiple time sequence characteristics |
-
2023
- 2023-05-09 CN CN202310518198.1A patent/CN116432860A/en active Pending
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116632841A (en) * | 2023-07-26 | 2023-08-22 | 国网山东省电力公司信息通信公司 | Power distribution area short-term electricity load prediction method and system integrating multiple time sequence characteristics |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN113159361A (en) | Short-term load prediction method and system based on VDM and Stacking model fusion | |
CN112364975A (en) | Terminal operation state prediction method and system based on graph neural network | |
CN111950805B (en) | Medium-and-long-term power load prediction method and device, computer equipment and storage medium | |
US6560586B1 (en) | Multiresolution learning paradigm and signal prediction | |
CN110909928B (en) | Energy load short-term prediction method and device, computer equipment and storage medium | |
CN116432860A (en) | Short-term power load prediction method, system and medium based on EWT-TCN | |
CN111105321A (en) | Power load prediction method based on EEMD secondary decomposition | |
Zhang et al. | Hybrid short-term load forecasting method based on empirical wavelet transform and bidirectional long short-term memory neural networks | |
Osogami et al. | Bidirectional learning for time-series models with hidden units | |
CN116933175A (en) | Electric automobile charging load prediction method and device | |
CN114091768A (en) | STL (Standard template library) and LSTM (local Scale TM) with attention mechanism based tourism demand prediction method | |
CN117251705A (en) | Daily natural gas load prediction method | |
Wang et al. | TATCN: time series prediction model based on time attention mechanism and TCN | |
CN111523644A (en) | Intermediate-long term runoff prediction method based on LSFL combined model | |
Chen et al. | Forecasting daily electricity load by wavelet neural networks optimized by cuckoo search algorithm | |
CN113344245A (en) | Hybrid deep learning short-term prediction model, method, storage medium, and computing device | |
CN112163710A (en) | Short-term reactive load prediction method and device based on ensemble empirical mode | |
Kuang et al. | Reconstructing signal from quantized signal based on singular spectral analysis | |
CN110990766A (en) | Data prediction method and storage medium | |
CN111126659A (en) | Power load prediction method and system | |
Luo et al. | Frequency-division combination forecasting of stock market based on wavelet multiresolution analysis | |
Ahmad et al. | Wavelet-based multiresolution forecasting | |
Wang et al. | CEEMDAN-CatBoost-SATCN-based short-term load forecasting model considering time series decomposition and feature selection | |
Wang et al. | Short-term electricity sales forecasting model based on wavelet decomposition and LSTM | |
Park | Multiresolution-based bilinear recurrent neural network |
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 |