CN109583944A - Gas Demand Forecast method based on one-dimensional Wavelet decomposing and recomposing and neural network - Google Patents
Gas Demand Forecast method based on one-dimensional Wavelet decomposing and recomposing and neural network Download PDFInfo
- Publication number
- CN109583944A CN109583944A CN201811346720.8A CN201811346720A CN109583944A CN 109583944 A CN109583944 A CN 109583944A CN 201811346720 A CN201811346720 A CN 201811346720A CN 109583944 A CN109583944 A CN 109583944A
- Authority
- CN
- China
- Prior art keywords
- neural network
- layer
- low frequency
- frequency component
- component
- 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
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 62
- 238000000034 method Methods 0.000 title claims abstract description 53
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 claims abstract description 42
- 239000007789 gas Substances 0.000 claims abstract description 26
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 25
- 239000003345 natural gas Substances 0.000 claims abstract description 22
- 238000009499 grossing Methods 0.000 claims description 25
- 230000008569 process Effects 0.000 claims description 13
- 210000004218 nerve net Anatomy 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims 1
- 238000010586 diagram Methods 0.000 description 32
- 230000006870 function Effects 0.000 description 18
- 238000012360 testing method Methods 0.000 description 14
- 238000012549 training Methods 0.000 description 7
- 230000009471 action Effects 0.000 description 2
- 235000013399 edible fruits Nutrition 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 210000005036 nerve Anatomy 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 230000001133 acceleration Effects 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
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
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
- G06Q30/0202—Market predictions or forecasting for commercial activities
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Economics (AREA)
- Finance (AREA)
- Development Economics (AREA)
- Accounting & Taxation (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- General Business, Economics & Management (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Entrepreneurship & Innovation (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Marketing (AREA)
- Game Theory and Decision Science (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Human Resources & Organizations (AREA)
- Primary Health Care (AREA)
- Tourism & Hospitality (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention relates to Forecast of Natural Gas Load method and technology fields, in particular to the Gas Demand Forecast method based on one-dimensional Wavelet decomposing and recomposing and neural network.Based on the data of city gas door station actual acquisition, the Natural Gas Demand time series of actual acquisition is decomposed using wavelet theory, the number of plies of Symlets wavelet decomposition is 4 layers, the order of decomposition is 8 ranks, the high fdrequency component come will be decomposited to predict using GRNN neural network, the low frequency component come is decomposited to be predicted using BP neural network, finally successively it is reconstructed, obtain final prediction result, and the result is compared with the prediction result that GRNN neural network and BP neural network are predicted is used alone, verify the validity and advance of the Gas Demand Forecast method proposed by the present invention based on one-dimensional Wavelet decomposing and recomposing and neural network.
Description
Technical field
The present invention relates to Forecast of Natural Gas Load method and technology fields, in particular to a kind of to be based on one-dimensional Wavelet decomposing and recomposing and mind
Gas Demand Forecast method through network.
Background technique
As the acceleration of Chinese Industrialization and urbanization process promotes, demand of all trades and professions to natural gas is growing day by day, due to
Gas Productivity wretched insufficiency and consumption figure rapid growth cause " gas is waste ", specific manifestation are as follows: when peak to urban
Section, demand is excessive, and air pressure is relatively low in pipe network, causes many users that cannot just commonly use gas；Low-valley interval, demand compared with
Few, gas ductwork air pressure inside is excessively high, equipment safety is threatened, so that pipe network operation inefficiency.Therefore, short-term natural gas load
The research of prediction, for guaranteeing that gas distributing system gas consumption, the scheduling for optimizing pipe network and maintenance of equipment have extremely important meaning
Justice.
However, natural gas load is in addition to having the characteristics that with all, day mechanical periodicities, it is also many by weather, season, festivals or holidays etc.
The features such as factor influences, and causes natural gas load fluctuation very frequent, is in nonlinearity, time variation, dispersibility and randomness,
Accurate Prediction difficulty is big.Traditional Forecast of Natural Gas Load method includes linear regression analysis, time series method and grey colour system
System is theoretical, but these methods are the model based on linear data prediction mostly, therefore are not suitable for complicated Forecast of Natural Gas Load.
Summary of the invention
It is a kind of based on one-dimensional Wavelet decomposing and recomposing and mind it is an object of the invention to provide aiming at the problems existing in the prior art
Gas Demand Forecast method through network.
The technical scheme is that
Gas Demand Forecast method based on one-dimensional Wavelet decomposing and recomposing and neural network, including following method and step:
(1) the when Load Time Series data for acquiring certain city's natural gas station, using Symlets small echo to the natural of actual acquisition
Load Time Series data are decomposed when gas, decomposite high fdrequency component and low frequency component, and building is based on one-dimensional wavelet decomposition weight
The Gas Demand Forecast model of structure and neural network；
(2) high fdrequency component come out through Symlets wavelet decomposition is predicted using GRNN neural network, with BP nerve net
Network predicts the low frequency component come out through Symlets wavelet decomposition；
(3) prediction result of the prediction result to GRNN neural network and BP neural network is reconstructed, and by reconstruction result with
The prediction result that GRNN neural network and BP neural network is used alone compares, to determine the prediction of the prediction model of building
Precision and validity.
Specifically, using the detailed process of Symlets wavelet decomposition in the step (1) are as follows: use Symlets wavelet basis letter
Load Time Series data are decomposed when several pairs of natural gases, and decomposition order is eight ranks, and Decomposition order is four layers, are respectively as follows: the
One layer of component, second layer component, third layer component and the 4th layer of component；It decomposites and comes four high fdrequency components and four low frequencies point
Amount, respectively first layer high fdrequency component, first layer low frequency component, second layer high fdrequency component, second layer low frequency component, third layer are high
Frequency component, third layer low frequency component, the 4th layer of high fdrequency component, the 4th layer of low frequency component.
The present invention uses Mallat Wavelet Fast Decomposition algorithm using Symlets wavelet decomposition, and Mallat algorithm is one kind by just
The decomposition algorithm and restructing algorithm of Wavelet Expansions time series are handed over, the algorithm is similar with quick Fourier transformation, has operation fast
Victory designs the features such as simple, is a kind of recursion fast algorithm of pure digi-tal, therefore is more and more applied, by signal point
Solution be different frequency bands component, so as to deeper into analysis signal the characteristics of, if by d0It is interpreted as discrete signal to be decomposed, root
Decomposable process such as Fig. 1 can be obtained according to Mallat decomposition algorithm, formula based on Fig. 1 decomposable process is as follows:
dj=ldj+1, j=1,2 ..., N
aj=haj+1, j=1,2 ..., N
In formula: l is low-pass filter, and h is high-pass filter, dj+1Indicate original signal 2-(j+1)Low frequency point under resolution ratio
Amount, aj+1Indicate original signal 2-(j+1)High fdrequency component under resolution ratio, by original discrete signal d0It is decomposed into ajAnd d1,
d2,…,dj, respectively indicate jth layer high fdrequency component, first layer low frequency component, second layer low frequency component ..., jth layer low frequency component,
Maximum decomposition level number is N.
Specifically, the detailed process of the step (2) predicted with two kinds of neural networks are as follows:
1. predicting using GRNN neural network the 4th layer of high fdrequency component come out through Symlets wavelet decomposition；
2. using BP neural network to first layer low frequency component, the second layer low frequency component, the come out through Symlets wavelet decomposition
Three layers of low frequency component and the 4th layer of low frequency component are predicted respectively.
GRNN neural network is a kind of general nonparametric Regression Model, one point as radial basis function neural network
Branch, is the feed-forward type neural network based on nonlinear regression theory, it is by activation neuron come approximating function, GRNN nerve net
The structure of network is divided into input layer, hidden layer and linear convergent rate layer, and network structure is similar to radial basis function network structure, such as specification
Shown in attached drawing Fig. 2.Fig. 2 is GRNN neural network structure schematic diagram, wherein 1 represent input layer, 2 represent hidden layer, 3 represent it is linear
Output layer, in formula: P is input vector；Q is the number of input vector；b1For hidden layer threshold value；| | dist | | it is distance function；
R is the element number of every group of vector；IW1,1For the weight of input layer；LW2,1For weight matrix；n2For output vector； a2It is linear
Transmission function.
BP neural network model includes input/output model, action function model, error calculating and self learning model.BP mind
It is a kind of Multilayer Network as made of input layer, output layer and one or more hidden layer node interconnection through network.BP neural network
Learning process is made of the forward-propagating of signal and two processes of reverse propagation of error, and when forward-propagating, model function is in defeated
Enter layer, after hidden layer is handled, be passed to the reverse propagation stage of error, output error is pressed into certain seed form, passes through hidden layer
It is successively returned to input layer, and " sharing " gives all units of each layer, to obtain the reference error of each layer unit, using as repairing
Change the foundation of each unit weight, the process that weight is constantly modified is also network learning procedure, and it is defeated that this process is performed until network
Error out be gradually reduced to acceptable degree or reach setting study number until.
Specifically, the detailed process of the step (3) are as follows:
1. the first layer of prediction result and BP neural network prediction to the 4th layer of high fdrequency component of GRNN neural network prediction is low
Frequency component, second layer low frequency component, third layer low frequency component and the 4th layer of low frequency component prediction result be reconstructed；According to weight
Structure algorithm can obtain restructuring procedure as shown in Figure of description Fig. 3, and formula based on Fig. 3 restructuring procedure is as follows:
d0=l*dj+1+h*aj+1
In formula: l*And h*A pair of of dual operator, j=N-1, N-2 ..., 0, using decomposite come a1,a2,…,ajAnd djRespectively
It is reconstructed, obtains A1,A2,…,AjAnd Dj, respectively first layer reconstructs low frequency signal, and the second layer reconstructs low frequency signal ..., and the
J layers of reconstruct low frequency signal and jth layer reconstructed high frequency signal；
2. the result after reconstruct is compared with the prediction result that GRNN neural network and BP neural network is used alone, with true
Surely the precision of prediction and validity of the prediction model constructed；
3. shown in the following formula of error assessment index:
Respectively RSME (root-mean-square error), MAE (mean absolute error), MAPE (average absolute percentage error), specific formula
Are as follows:
In formula: N is the total quantity of error comparison；Lactual is the natural gas load of actual acquisition；Lforecast is prediction
Natural gas load.
Specifically, in the step (1) to the natural gas of actual acquisition when Load Time Series data carry out decomposing preceding progress
Pretreatment, the preprocess method is Gaussian smoothing method.
The study amendment that GRNN is connected to the network weight uses BP algorithm, since the action function in network hidden layer node is using high
This function, so that there is partial approximation ability, further, since artificial adjustment parameter is seldom in GRNN, only one threshold value, network
Study all rely on data sample, this feature determines that network is able to avoid most possibly artificial subjective to assume to tie prediction
The influence of fruit.
BP neural network is a kind of Multi-layered Feedforward Networks by Back Propagation Algorithm training, is current most widely used nerve
One of network model, BP network can learn and store a large amount of input -- and output mode mapping relations are retouched without disclosing in advance
The math equation of this mapping relations is stated, there is strong nonlinearity capability of fitting.
In order to improve the precision of prediction of Natural Gas Demand, method provided by the invention be based on Symlets wavelet basis function,
In conjunction with the strong nonlinearity capability of fitting of GRNN neural network and BP neural network, and GRNN neural network and BP nerve is used alone
Network carries out prediction and compares, and is then reconstructed, can be improved the precision of prediction of Gas Demand Forecast.
Detailed description of the invention
Fig. 1 is that Mallat algorithm decomposed structural schematic diagram；
Fig. 2 is GRNN neural network structure schematic diagram；
Fig. 3 is the restructuring procedure structural schematic diagram according to restructing algorithm；
Fig. 4 is method and technology route map provided by the present invention；
Fig. 5 is the primary data sample schematic diagram of acquisition；
Fig. 6 is the sample schematic diagram carried out after Gaussian smoothing to data shown in Fig. 5；
Fig. 7 is that Symlets wavelet basis function decomposites and carrys out the 4th layer of high fdrequency component schematic diagram；
Fig. 8 is that Symlets wavelet basis function decomposites and carrys out first layer low frequency component schematic diagram；
Fig. 9 is that Symlets wavelet basis function decomposites and carrys out second layer low frequency component schematic diagram；
Figure 10 is that Symlets wavelet basis function decomposites and carrys out third layer low frequency component schematic diagram；
Figure 11 is that Symlets wavelet basis function decomposites and carrys out the 4th layer of low frequency component schematic diagram；
Figure 12 is the 4th layer of high fdrequency component schematic diagram after Gaussian smoothing；
Figure 13 is first layer low frequency component schematic diagram after Gaussian smoothing；
Figure 14 is second layer low frequency component schematic diagram after Gaussian smoothing；
Figure 15 is third layer low frequency component schematic diagram after Gaussian smoothing；
Figure 16 is the 4th layer of low frequency component schematic diagram after Gaussian smoothing；
Figure 17 is the 4th layer of high fdrequency component training set result schematic diagram after Gaussian smoothing；
Figure 18 is the 4th layer of high fdrequency component test set result schematic diagram after Gaussian smoothing；
Figure 19 is first layer low frequency component training set result after Gaussian smoothing；
Figure 20 is first layer low frequency component test set result schematic diagram after Gaussian smoothing；
Figure 21 is second layer low frequency component training set result schematic diagram after Gaussian smoothing；
Figure 22 is second layer low frequency component test set result schematic diagram after Gaussian smoothing；
Figure 23 is third layer low frequency component training set result schematic diagram after Gaussian smoothing；
Figure 24 is third layer low frequency component test set result schematic diagram after Gaussian smoothing；
Figure 25 is the 4th layer of low frequency component training set result schematic diagram after Gaussian smoothing；
Figure 26 is the 4th layer of low frequency component test set result schematic diagram after Gaussian smoothing；
Figure 27 is five component Gaussians smoothly rear reconstruction result schematic diagram；
Figure 28 is five component test set reconstruction result schematic diagrames.
Specific embodiment
Below with reference to example and attached drawing to the present invention is based on the Gas Demand Forecasts of one-dimensional Wavelet decomposing and recomposing and neural network
Method makes detailed explanation.
Fig. 4 is the Technology Roadmap of the method for the invention.The when load data at the natural valve station in certain city is acquired first,
Sample acquires 30 days data altogether, wherein daily 24 hours, it is acquired since 8 points of June 1, until 7 points of evening of June 30, often
It acquires 1 time within 1 hour, amounts to 720 data.When using Symlets wavelet basis function to natural gas Load Time Series data into
Row decomposes, and predicts using GRNN neural network the high fdrequency component come is decomposited, next to decompositing with BP neural network
Low frequency component predicted that be finally reconstructed, comparison is individually using the prediction essence of GRNN neural network and BP neural network
Degree.
Fig. 5 gives collected primary data sample, and Fig. 6 is that data sample after Gaussian smoothing is carried out to initial data.
In order to improve the precision of prediction of Natural Gas Demand time series, the preprocess method of use is Gaussian smoothing method, Gaussian smoothing method
The order used is 3 rank.
Fig. 7-Figure 11 gives the schematic diagram that Symlets wavelet basis function decomposites each component come, and the number of plies of decomposition is 4
Layer, the order used is 8 rank；Figure 12-Figure 16 gives the schematic diagram of each layer component after Gaussian smoothing, to through Symlets small echo
The high and low frequency component that basis function decomposition comes out carries out smoothly, and the order that Gaussian smoothing method uses decomposes for 3 ranks to the 4th layer
High fdrequency component, first layer low frequency component, second layer low frequency component, third layer low frequency component and the 4th layer of low frequency component carry out flat
It is sliding.
Figure 17-Figure 26 gives each layer component training set and test set result schematic diagram after Gaussian smoothing, and BP neural network uses
Structure be (5,5), error result, the first layer of the 4th layer of high fdrequency component G4 test set and forecast set is set forth in the following table 1
The error result of low frequency component D1 test set and forecast set, second layer low frequency component D2 test set and forecast set error result,
The error of the error result of third layer low frequency component D3 test set and forecast set, the 4th layer of low frequency component D4 test set and forecast set
As a result.
Table 1
Figure 27 gives reconstruction result schematic diagram after each layer component Gaussian smoothing processing, and Figure 28 is each layer component test set reconstruct knot
Three errors of RMSE, MAE and MAPE value of fruit schematic diagram, prediction technique of the present invention are respectively 178.5540,24.2252 and
0.0301, individually using GRNN neural network predicted when three errors of RMSE, MAE and MAPE value be respectively
754.0844,25.1171 and 0.0315, three mistakes of RMSE, MAE and MAPE value when individually being predicted using BP neural network
Difference other 893.7062,26.4208 and 0.0315 carries out prediction ratio using GRNN neural network and BP neural network with independent
Compared with the present invention reduces 575.5304,0.8919 and in three smooth indexs of error of RMSE, MAE and MAPE value respectively
0.0014,715.1522,2.1956 and 0.0054, it can be seen that, the present invention is improved on precision of prediction, demonstrates this
The validity of the prediction technique provided is provided.
In conclusion the present invention is relatively individually improved using GRNN neural network and BP neural network on precision of prediction, test
The validity of inventive algorithm is demonstrate,proved.Prediction technique provided by the invention has higher precision of prediction to Gas Demand Forecast,
It is a kind of efficient gas dissipation method.
Finally it should be noted that: the above embodiments are merely illustrative of the technical scheme of the present invention and are not intended to be limiting thereof；Although ginseng
According to preferred embodiment, invention is explained in detail, it should be understood by those ordinary skilled in the art that: still can be with
It modifies to a specific embodiment of the invention or some technical features can be equivalently replaced；Without departing from skill of the present invention
The spirit of art scheme should all cover within the scope of the technical scheme claimed by the invention.
Claims (5)
1. the Gas Demand Forecast method based on one-dimensional Wavelet decomposing and recomposing and neural network, which is characterized in that including as follows
Method and step:
(1) the when Load Time Series data for acquiring certain city's natural gas station, using Symlets small echo to the natural of actual acquisition
Load Time Series data are decomposed when gas, decomposite high fdrequency component and low frequency component, and building is based on one-dimensional wavelet decomposition weight
The Gas Demand Forecast model of structure and neural network；
(2) high fdrequency component come out through Symlets wavelet decomposition is predicted using GRNN neural network, with BP nerve net
Network predicts the low frequency component come out through Symlets wavelet decomposition；
(3) prediction result of the prediction result to GRNN neural network and BP neural network is reconstructed, and by reconstruction result with
The prediction result that GRNN neural network and BP neural network is used alone compares, to determine the prediction of the prediction model of building
Precision and validity.
2. the Gas Demand Forecast method based on one-dimensional Wavelet decomposing and recomposing and neural network according to claim 1,
It is characterized in that, using the detailed process of Symlets wavelet decomposition in the step (1) are as follows: use Symlets wavelet basis function pair
Load Time Series data are decomposed when natural gas, and decomposition order is eight ranks, and Decomposition order is four layers, are respectively as follows: first layer
Component, second layer component, third layer component and the 4th layer of component；It decomposites and comes four high fdrequency components and four low frequency components, point
It Wei not first layer high fdrequency component, first layer low frequency component, second layer high fdrequency component, second layer low frequency component, the high frequency division of third layer
Amount, third layer low frequency component, the 4th layer of high fdrequency component, the 4th layer of low frequency component.
3. the Gas Demand Forecast method based on one-dimensional Wavelet decomposing and recomposing and neural network according to claim 2,
It is characterized in that, the detailed process of the step (2) predicted with two kinds of neural networks are as follows:
1. predicting using GRNN neural network the 4th layer of high fdrequency component come out through Symlets wavelet decomposition；
2. using BP neural network to first layer low frequency component, the second layer low frequency component, the come out through Symlets wavelet decomposition
Three layers of low frequency component and the 4th layer of low frequency component are predicted respectively.
4. the Gas Demand Forecast method based on one-dimensional Wavelet decomposing and recomposing and neural network according to claim 1,
It is characterized in that, the detailed process of the step (3) are as follows:
1. the first layer of prediction result and BP neural network prediction to the 4th layer of high fdrequency component of GRNN neural network prediction is low
Frequency component, second layer low frequency component, third layer low frequency component and the 4th layer of low frequency component prediction result be reconstructed；
2. the result after reconstruct is compared with the prediction result that GRNN neural network and BP neural network is used alone, with true
Surely the precision of prediction and validity of the prediction model constructed；
3. shown in the following formula of error assessment index:
Respectively RSME (root-mean-square error), MAE (mean absolute error), MAPE (average absolute percentage error), specific formula
Are as follows:
In formula: N is the total quantity of error comparison；Lactual is the natural gas load of actual acquisition；Lforecast is prediction
Natural gas load.
5. the Gas Demand Forecast method based on one-dimensional Wavelet decomposing and recomposing and neural network according to claim 1,
It is characterized in that, is carried out before Load Time Series data are decomposed when in the step (1) to the natural gas of actual acquisition pre-
Processing, the preprocess method is Gaussian smoothing method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811346720.8A CN109583944A (en) | 2018-11-13 | 2018-11-13 | Gas Demand Forecast method based on one-dimensional Wavelet decomposing and recomposing and neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811346720.8A CN109583944A (en) | 2018-11-13 | 2018-11-13 | Gas Demand Forecast method based on one-dimensional Wavelet decomposing and recomposing and neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109583944A true CN109583944A (en) | 2019-04-05 |
Family
ID=65922213
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811346720.8A Pending CN109583944A (en) | 2018-11-13 | 2018-11-13 | Gas Demand Forecast method based on one-dimensional Wavelet decomposing and recomposing and neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109583944A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110390429A (en) * | 2019-07-15 | 2019-10-29 | 重庆邮电大学 | A kind of quality of water environment forecasting system neural network based and method |
| CN114325057A (en) * | 2021-12-31 | 2022-04-12 | 重庆大学 | PT primary voltage reconstruction method based on inverse black box and inverse electromagnetic dual model |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103136598A (en) * | 2013-02-26 | 2013-06-05 | 福建省电力有限公司 | Monthly electrical load computer forecasting method based on wavelet analysis |
| CN104123591A (en) * | 2014-07-14 | 2014-10-29 | 上海师范大学 | Smart grid classification and fuzzy neural network based natural gas load prediction method |
| CN104408529A (en) * | 2014-11-21 | 2015-03-11 | 广东工业大学 | Short-term load predicting method of power grid |
| WO2016106373A1 (en) * | 2014-12-22 | 2016-06-30 | Robert Bosch Gmbh | Method for adaptive demand charge reduction |
| CN105894113A (en) * | 2016-03-31 | 2016-08-24 | 中国石油天然气股份有限公司规划总院 | Natural gas short-period demand prediction method |
-
2018
- 2018-11-13 CN CN201811346720.8A patent/CN109583944A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103136598A (en) * | 2013-02-26 | 2013-06-05 | 福建省电力有限公司 | Monthly electrical load computer forecasting method based on wavelet analysis |
| CN104123591A (en) * | 2014-07-14 | 2014-10-29 | 上海师范大学 | Smart grid classification and fuzzy neural network based natural gas load prediction method |
| CN104408529A (en) * | 2014-11-21 | 2015-03-11 | 广东工业大学 | Short-term load predicting method of power grid |
| WO2016106373A1 (en) * | 2014-12-22 | 2016-06-30 | Robert Bosch Gmbh | Method for adaptive demand charge reduction |
| CN105894113A (en) * | 2016-03-31 | 2016-08-24 | 中国石油天然气股份有限公司规划总院 | Natural gas short-period demand prediction method |
Non-Patent Citations (3)
| Title |
|---|
| 乔伟彪: "基于小波变换和LSSVM-DE的天然气日负荷组合预测模型", 《天然气工业》 * |
| 刘光中等: "组合神经网络模型对电力需求的预测", 《数量经济技术经济研究》 * |
| 孔繁萍: "基于神经网络的组合模型在物联网产业发展预测中的应用研究", 《南方农机》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110390429A (en) * | 2019-07-15 | 2019-10-29 | 重庆邮电大学 | A kind of quality of water environment forecasting system neural network based and method |
| CN114325057A (en) * | 2021-12-31 | 2022-04-12 | 重庆大学 | PT primary voltage reconstruction method based on inverse black box and inverse electromagnetic dual model |
| CN114325057B (en) * | 2021-12-31 | 2023-09-19 | 重庆大学 | PT primary voltage reconstruction method based on inverse black box and inverse electromagnetic dual model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Vamsidhar et al. | Prediction of rainfall using backpropagation neural network model | |
| CN110503251B (en) | Non-holiday load prediction method based on Stacking algorithm | |
| CN111178616B (en) | Wind speed prediction method based on negative correlation learning and regularization extreme learning machine integration | |
| CN112598180A (en) | Distributed regional wind power prediction method | |
| CN109242265B (en) | Urban water demand combined prediction method based on least square sum of errors | |
| CN104899653B (en) | Lake storehouse blue-green alga bloom Forecasting Methodology based on expert system and blue algae growth mechanism temporal model | |
| CN109214607A (en) | Short-term Forecast of Natural Gas Load model based on wavelet theory and neural network | |
| CN109583944A (en) | Gas Demand Forecast method based on one-dimensional Wavelet decomposing and recomposing and neural network | |
| CN114036736A (en) | Cause and effect network learning method based on local granger cause and effect analysis | |
| CN110046697A (en) | A kind of Kp index based on K index shows reporting method | |
| CN105515857B (en) | A kind of metropolitan data network flow measuring method based on RBF | |
| Ivakhnenko | Past, present, and future of GMDH | |
| CN110796303A (en) | Short-term power load prediction method based on EWT and ODBSCAN | |
| CN110110921A (en) | A kind of river level prediction technique considering time-lag effect | |
| Kisi et al. | Wind speed prediction by using different wavelet conjunction models | |
| CN115860797B (en) | Electric quantity demand prediction method suitable for new electricity price reform situation | |
| CN109214606A (en) | Load forecasting method when a kind of city gas | |
| Tao et al. | A combined model of wavelet and neural network for short term load forecasting | |
| CN114862032B (en) | XGBoost-LSTM-based power grid load prediction method and device | |
| CN115952915A (en) | Energy consumption prediction optimization method using fuzzy entropy classification | |
| CN114841412A (en) | Method for predicting pH value of sea cucumber growing water | |
| He et al. | Application of neural network model based on combination of fuzzy classification and input selection in short term load forecasting | |
| Roshan et al. | A new approach to technique for order-preference by similarity to ideal solution (TOPSIS) method for determining and ranking drought: A case study of Shiraz station | |
| CN110728051A (en) | Continuous model of two-stage method | |
| Johnes | Up around the bend: linear and nonlinear models of the UK economy compared |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190405 |
|
| RJ01 | Rejection of invention patent application after publication |