CN111784043A - Accurate prediction method for power selling amount of power distribution station area based on modal GRU learning network - Google Patents

Accurate prediction method for power selling amount of power distribution station area based on modal GRU learning network Download PDF

Info

Publication number
CN111784043A
CN111784043A CN202010607362.2A CN202010607362A CN111784043A CN 111784043 A CN111784043 A CN 111784043A CN 202010607362 A CN202010607362 A CN 202010607362A CN 111784043 A CN111784043 A CN 111784043A
Authority
CN
China
Prior art keywords
modal
gru
data
electricity sales
model
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
Application number
CN202010607362.2A
Other languages
Chinese (zh)
Inventor
陈光宇
张仰飞
刘成
郝思鹏
刘海涛
吕干云
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing Institute of Technology
Original Assignee
Nanjing Institute of Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Nanjing Institute of Technology filed Critical Nanjing Institute of Technology
Priority to CN202010607362.2A priority Critical patent/CN111784043A/en
Publication of CN111784043A publication Critical patent/CN111784043A/en
Pending legal-status Critical Current

Links

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
    • G06N3/08—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—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Electricity, gas or water supply

Abstract

A method for accurately predicting the electricity selling amount of a power distribution station area based on a modal GRU learning network comprises the following steps: s1, obtaining historical data of electricity sales of the distribution room, and dividing a test set and a training set; s2, preprocessing data, complementing sampling time points to ensure continuity of the sampling time points, and filling missing data of sampling points by using an average interpolation method; s3, determining the optimal mode number K of Variational Mode Decomposition (VMD) according to the center frequency of each mode component by using an experimental method; s4, VMD decomposition is carried out on the historical data of the electricity sales volume of the distribution room, and low-frequency modal components and high-frequency modal components after decomposition are respectively extracted; s5, respectively predicting a low-frequency mode and a high-frequency mode by using a Prophet prediction model and a GRU learning network; and S6, reconstructing prediction results of each mode to obtain a predicted value of the power selling amount of the distribution room. The method and the device can improve the prediction precision of the power selling amount of the distribution room, and can provide theoretical and practical support for the accurate prediction and management of the power selling amount of the distribution room.

Description

Accurate prediction method for power selling amount of power distribution station area based on modal GRU learning network
Technical Field
The invention relates to the field of electric quantity prediction, in particular to a method for accurately predicting the electric quantity sold in a power distribution station area based on a modal GRU learning network.
Background
The power supply enterprise can adjust the power supply plan and optimize the power supply structure by analyzing and predicting the power sale amount of the distribution room, and the development concept of constructing a conservation-oriented society and promoting energy conservation and emission reduction is met. Therefore, establishing an effective power sales prediction model in the distribution area has been a research hotspot in the power field.
The power selling amount of the distribution room is generally influenced by the superposition of a plurality of factors such as power utilization behaviors of users, load changes, seasonal changes, holidays and the like, so that the time sequence of the power selling amount of the distribution room shows an unstable change trend, and common prediction models comprise: support vector machines, random forest algorithms, neural networks and the like, but because reasonable refinement and decomposition are not carried out on the electric quantity data, the influence of various superposition factors on the electric quantity data cannot be considered, and therefore the prediction effect is usually poor.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides a power distribution station area electricity sales amount accurate prediction method based on a modal GRU learning network.
In order to achieve the purpose, the invention adopts the following technical scheme:
a method for accurately predicting the electricity selling amount of a power distribution station area based on a modal GRU learning network is characterized by comprising the following steps:
s1: acquiring historical data of electricity sales in a distribution room, and dividing a test set and a training set;
s2: data preprocessing, namely complementing the sampling time points to ensure continuity of the sampling time points, and filling missing data of the sampling time points by using an average interpolation method to obtain a time sequence of power selling amount of the distribution room;
s3: determining the optimal modal number K of the variational modal decomposition according to the central frequency of each modal component by using an experimental method;
s4: at the optimum number of modesVMD decomposition is carried out on the time sequence of the electricity sales amount of the distribution room under the condition of K, and the decomposed low-frequency modal components IMF are respectively extractedLAnd high frequency modal component IMFHA time series;
s5: respectively predicting a low-frequency mode and a high-frequency mode by using a Prophet prediction model and a GRU learning network;
s6: and reconstructing prediction results of all modes to obtain a predicted value of the electricity sales amount of the distribution room.
In order to optimize the technical scheme, the specific measures adopted further comprise:
further, the data preprocessing process in step S2 is specifically as follows:
s21: filling a small number of missing sampling time points in the sample time sequence data acquired in the S1, and if the sampling time points and the data are missing in a large area, filling by using the data of other years in the same period;
s22: filling missing data T at sampling time point by using average interpolation method, and if T is calculatediAnd Ti+1At any point in between, the average value of the sums is taken directly as:
Figure BDA0002559344030000021
in the formula, Ti、Ti+1Respectively representing the data before and after the missing data.
Further, the process of determining the optimal mode number K in step S3 is specifically as follows:
s31, setting the mode number K to be 2, and initializing a VMD parameter;
s32, VMD decomposition is carried out on the time sequence of the electricity sales volume of the distribution room to obtain the center frequency of each mode, and the constraint variational equation of the K-order decomposition process is as follows:
Figure BDA0002559344030000022
where f (t) is the original signal, t is a time variable, ukAnd ωkRespectively representing the resulting set of K-order modes and their center frequencies, ukEquivalent to uk(t); (t) is(ii) a Rake distribution; l ({ u)k}，{ωkH, λ) is an augmented lagrange function, α is a balance parameter, λ is a lagrange multiplier, λ is equal to λ (t), denotes convolution;
Figure BDA0002559344030000023
are respectively
Figure BDA0002559344030000024
f(t)、ut(t), a Fourier transform of λ (t); ω represents the modal mid-frequency and n represents the fourier expansion order;
s33, comparing whether each adjacent modal component is similar under the modal number K, if so, taking the K as the optimal modal number, and turning to S4; if not, the mode number K is set to K +1, and the process goes to S32.
Furthermore, Prophet is a time series decomposition model, which is mainly used for researching time series data characteristics and time series change rules and predicting future trends. The model can make up the defects that the traditional time sequence model needs to be filled with missing values, the model is lack of flexibility and the like. The Prophet prediction model in step S5 takes the following basic form:
p(t)＝g(t)+s(t)+h(t)+t
wherein, in the model p (t):
1) and g (t) is a sales electricity quantity growth trend model, and the component represents the variation trend of the sales electricity quantity time series in a non-periodic manner. In some cases, such as changing the month of season (changepoint), the trend of each data segment of the electricity selling time series changes with the changing point. Adding S mutation points in the time sequence, wherein the time stamp position of each mutation point is Sj(j is more than or equal to 1 and less than or equal to S), and realizing piecewise linear fitting by adopting a piecewise logistic regression growth model:
Figure BDA0002559344030000031
Figure BDA0002559344030000032
in the formulaAnd a (t) is a vector of the index function; r (t) is expected capacity (bearing capacity) of the model, and for simple calculation, the non-abnormal maximum selling electricity quantity of the historical selling electricity quantity is taken as the value of R (t); k + a (t)TThe increase rate of the electricity sales volume along with the change of time is the change value of the increase rate at the mutation point; m is an offset parameter, γ is a parameter vector that makes the function continuous, and t is a time variable;
2) s (t) is a seasonal trend model of the power sales, which usually shows corresponding periodic changes with seasons such as week, month, year, etc., and therefore needs to be modeled and predicted separately. The seasonal components are modeled based on a fourier series:
Figure BDA0002559344030000033
Figure BDA0002559344030000034
where T is the period length of the time series (e.g., T-7 indicates a period of a cycle), 2N indicates the number of periods expected to be used in the model, N is the order of the Fourier transform, and β ═ a1，b1，...，aN，bN]，an、bnIs a parameter to be estimated;
3) h (t) is a holiday trend model of the electricity sales, and if M holidays exist, a holiday component model is expressed as:
Figure BDA0002559344030000035
z(t)＝[1(t∈Di)，...，1(t∈DM)]
κ＝(κ1，...，κM)T
wherein, κ to Normal (o, v)2) Kappa represents the influence of holiday i and conforms to normal distribution, the index kappa is influenced by a holiday intensity index v, the value of the kappa is in direct proportion to the influence of holidays on electricity sales, and the kappa isMShows the shadow of the Mth holiday on the predicted valueSounding; diIndicating the time t contained in the window, 1 if t belongs to the vacation i (t ∈ D)i) 1 otherwise (t ∈ D)i)＝0；
4)tIs an error component, indicating an unpredicted fluctuation.
Furthermore, the GRU neural network can mine the self characteristics of the time sequence, improve the accuracy of the prediction result and is suitable for the condition that the rule is unknown or uncertain. The GRU comprises an update gate and a reset gate, and has fewer structural parameters and faster convergence speed compared with other neural networks. The degree of the state information of the previous moment of the door control is kept in the current state is updated, and the larger the value is, the more the state information of the previous moment is kept. Resetting the gate controls the extent to which the current state is combined with previous information, with smaller values indicating more information to ignore. In step S5, the GRU neural network includes an update gate and a reset gate, and has the following structure:
zt＝σ(W(z)xt+U(z)ht-1)
rt＝σ(W(r)xt+U(r)ht-1)
in the formula, ztTo refresh the door, rtTo reset the gate, xtFor input, htIs the output of the hidden layer; the GRU-based unit calculates h by the following formulat：
Figure BDA0002559344030000041
Figure BDA0002559344030000042
In the formula (I), the compound is shown in the specification,
Figure BDA0002559344030000043
is to input xtAnd past hidden state ht-1Summarizing; u shape(z)、W(z)、U(r)、W(r)U and W are trainable parameter matrices.
Further, the GRU neural network is trained by adopting a training set, and the prediction effect of the GRU neural network is checked by adopting a testing set.
The invention has the beneficial effects that: the invention provides a power distribution area electricity selling amount accurate prediction method based on a modal GRU learning network, which comprises the steps of firstly decomposing an area electricity selling amount time sequence into a low-frequency mode and a high-frequency mode by utilizing Variational Mode Decomposition (VMD), thereby reducing the non-stationarity of the time sequence; secondly, predicting the decomposed low-frequency mode and high-frequency mode by using a Prophet time series prediction model and a GRU learning network respectively; and finally reconstructing the prediction results of the high-frequency and low-frequency modes to obtain the prediction result of the power selling amount of the distribution room. The method and the device can improve the accuracy of the prediction of the power selling amount of the distribution room, have good applicability and higher accuracy, and can provide theoretical and practical support for the accurate prediction and management of the power selling amount of the distribution room.
Drawings
FIG. 1 is a general flow diagram of the present invention.
Fig. 2 is a diagram of a Prophet model architecture.
Fig. 3 is a diagram of a GRU learning network architecture.
Fig. 4 is a trend graph of sales power for bays 2016.1-2019.5.
Fig. 5 is a VMD decomposition result diagram.
FIGS. 6a and 6b are a Prophet prediction analysis result graph and a composition trend analysis graph, respectively.
Fig. 7 is a GRU learning network test set prediction graph.
FIG. 8 is a comparison of the prediction results of different prediction models.
FIG. 9 is a graph of error versus predicted point for different models.
Detailed Description
The present invention will now be described in further detail with reference to the accompanying drawings.
The method for accurately predicting the power selling amount of the power distribution station area based on the modal GRU learning network, as shown in FIG. 1, comprises the following steps:
step 1: historical data of electricity sales in the distribution room is obtained, and a test set and a training set are divided.
Step 2: and (3) preprocessing data, complementing sampling time points to ensure continuity of the sampling time points, and filling missing data of sampling points by using an average interpolation method.
And step 3: determining the optimal modal number K of Variational Modal Decomposition (VMD) according to the central frequency of each modal component by using an experimental method; the method specifically comprises the following steps:
step 3.1: initializing a VMD parameter by setting the mode number K to be 2;
step 3.2: VMD decomposition is carried out on the time sequence of the electricity sales quantity to obtain the center frequency of each mode;
step 3.3: comparing whether each adjacent modal component is similar under the modal number K, if so, taking the K as the optimal modal number, and turning to the step 4; if not, let the mode number K be K +1, go to step 3.2.
And 4, step 4: VMD decomposition is carried out on the time sequence of the electricity sales volume of the distribution room under the condition of the optimal modal number K, and the decomposed low-frequency modal components IMF are respectively extractedLAnd high frequency modal component IMFHTime series.
And 5, respectively predicting the low-frequency mode and the high-frequency mode by utilizing a Prophet prediction model and a GRU learning network, wherein the structure of the Prophet model is shown in fig. 2, the structure of the GRU is shown in fig. 3, the direction indicated by an arrow in the diagram is the data flow direction, × is the number multiplication of a matrix, sigma is an activation function Sigmoid function, tanh is an activation function, and 1-represents that the forward propagation data of the link is 1-zt。
Step 6: and summing and reconstructing prediction results of all modes to obtain a predicted value of the electricity sales amount of the distribution room.
In order to further prove the effectiveness and the advantages of the method provided by the invention, the electricity sales data of a certain region in the eastern region of China from 2016, 1 and 1 month and 2019, 5 and 30 months are selected for experimental simulation analysis, the sampling time period of the data set is 24h, 1247 sampling points are provided in total, the monitoring data of 1155 sampling points from 2016, 1 and 1 month and 2019, 2 and 28 months are selected as a training set, and the monitoring data of 92 points from 2019, 3 and 1 month and 2019, 5 and 31 months are selected as a testing set.
The historical data of the electricity selling amount of the selected sample platform area has the condition of a small amount of data loss, and the VMD decomposition is influenced by the loss of the sampled data, so the invention adopts an average interpolation method to complement the lost data. The time series of the electricity sales after the station area is processed is shown in fig. 4.
As can be seen from fig. 4, on one hand, the time series of electricity sales in the distribution area has obvious periodic fluctuation, presents the characteristics of low electricity sales in spring and autumn and high electricity sales in summer and winter, and belongs to the time series with seasonal regularity; on the other hand, the time sequence has obvious non-stationarity and can generate larger error when being directly predicted, so that the time sequence is decomposed into modes with higher regularity by using the VMD, and low-frequency modes and high-frequency modes are respectively predicted, summed and reconstructed, so that a more accurate prediction result is obtained.
Before decomposition, the mode number K needs to be determined, and too much mode number can cause repetition or noise, and too little mode number can cause under-decomposition. The center frequencies at which K takes different values are shown in table 1.
Table 12016-Summit and holiday list 2019
Figure BDA0002559344030000061
As can be seen from table 1, when K is 3, two similar modes of 0.142 and 0.298 appear in the center frequency, so that K is 2, and the decomposition result is shown in fig. 5.
As can be seen from fig. 5, the low-frequency mode IMF1 bears the long-term trend of the district sales power, and the high-frequency mode IMF2 reveals the paroxysmal and irregular fluctuations of the district sales power. On the basis, the prediction is carried out by utilizing Prophet and GRU respectively.
1) Predicting low frequency modal IMF1
The prediction result of the low-frequency mode IMF1 of the district electricity sales volume and the prediction trend of each decomposed component are shown in fig. 6a and 6 b.
2) Predicting high frequency mode IMF2
The first 1155 data of the high-frequency mode IMF2 are selected as the training set of the GRU learning network, and the last 92 data are selected as the test set, so that the test set prediction curve is obtained as shown in fig. 7.
3) Analysis of reconstructed prediction results
And reconstructing the prediction results of the IMF1 and the IMF2 by the Prophet and the GRU to obtain the predicted value of the station district electricity sales amount. In order to illustrate the superiority of the method provided by the invention in the aspect of predicting the electricity sales amount of the platform area, Prophet and GRU single models before combination and SVR and ARIMA are directly used for predicting the electricity sales amount of the platform area between 3 and 1 days in 2019 and 5 and 31 days in 2019 by comparison analysis, the result of each model prediction is compared with the method provided by the invention, the method uses Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) evaluation indexes to test the prediction effect of the model, the Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) evaluation index results of each model prediction value and observation value are calculated and shown in table 2, and the electricity sales amount prediction result of each method is shown in figure 8.
TABLE 2 error comparison of different model predictions
Figure BDA0002559344030000071
As can be seen from the evaluation results in table 2: compared with the effect of predicting the electricity sales amount directly by a Prophet and GRU single model before combination and SVR and ARIMA, the VMD-Prophet-GRU method provided by the invention has the minimum value in the indexes of RMSE and MAE.
From FIG. 8, it can be seen that the VMD-Prophet-GRU method proposed by the present invention is closer to the true value than other methods.
Fig. 9 shows the error comparison results (all the errors are absolute values) of the predicted points of different models, and it is obvious from the figure that the VMD-Prophet-GRU method provided by the present invention has the advantage of a small error (the red point is closer to 0) compared with other methods, which also shows that the prediction of the line loss rate trend by the method provided by the present invention is more accurate and has a better prediction effect compared with other methods.
The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above-mentioned embodiments, and all technical solutions belonging to the idea of the present invention belong to the protection scope of the present invention. It should be noted that modifications and embellishments within the scope of the invention may be made by those skilled in the art without departing from the principle of the invention.

Claims (6)

1. A method for accurately predicting the electricity selling amount of a power distribution station area based on a modal GRU learning network is characterized by comprising the following steps:
s1: acquiring historical data of electricity sales in a distribution room, and dividing a test set and a training set;
s2: data preprocessing, namely complementing the sampling time points to ensure continuity of the sampling time points, and filling missing data of the sampling time points by using an average interpolation method to obtain a time sequence of power selling amount of the distribution room;
s3: determining the optimal modal number K of the variational modal decomposition according to the central frequency of each modal component by using an experimental method;
s4: VMD decomposition is carried out on the time sequence of the electricity sales amount of the distribution room under the condition of the optimal mode number K, and low-frequency mode components and high-frequency mode components after decomposition are respectively extracted;
s5: respectively predicting a low-frequency mode and a high-frequency mode by using a Prophet prediction model and a GRU learning network;
s6: and reconstructing prediction results of all modes to obtain a predicted value of the electricity sales amount of the distribution room.
2. The accurate prediction method of power distribution station power selling amount based on modal GRU learning network as claimed in claim 1, characterized in that: the data preprocessing process in step S2 is specifically as follows:
s21: filling a small number of missing sampling time points in the sample time sequence data acquired in the S1, and if the sampling time points and the data are missing in a large area, filling by using the data of other years in the same period;
s22: filling missing data T at sampling time point by using average interpolation method, and if T is calculatediAnd Ti+1At any point in between, the average value of the sums is taken directly as:
Figure FDA0002559344020000011
in the formula, Ti、Ti+1Respectively representing the data before and after the missing data.
3. The accurate prediction method of power distribution station power selling amount based on modal GRU learning network as claimed in claim 1, characterized in that: the process of determining the optimal mode number K in step S3 is specifically as follows:
s31, setting the mode number K to be 2, and initializing a VMD parameter;
s32, VMD decomposition is carried out on the time sequence of the electricity sales volume of the distribution room to obtain the center frequency of each mode, and the constraint variational equation of the K-order decomposition process is as follows:
Figure FDA0002559344020000021
where f (t) is the original signal, t is a time variable, ukAnd ωkRespectively representing the resulting set of K-order modes and their center frequencies, ukEquivalent to uk(t); (t) is the dirac distribution; l ({ u)k}，{ωkH, λ) is an augmented lagrange function, α is a balance parameter, λ is a lagrange multiplier, λ is equal to λ (t), denotes convolution;
Figure FDA0002559344020000022
are respectively
Figure FDA0002559344020000023
f(t)、ut(t), a Fourier transform of λ (t); ω represents the modal mid-frequency and n represents the fourier expansion order;
s33, comparing whether each adjacent modal component is similar under the modal number K, if so, taking the K as the optimal modal number, and turning to S4; if not, the mode number K is set to K +1, and the process goes to S32.
4. The accurate prediction method of power distribution station power selling amount based on modal GRU learning network as claimed in claim 1, characterized in that: the Prophet prediction model in step S5 takes the following basic form:
p(t)＝g(t)+s(t)+h(t)+t
wherein, in the model p (t):
1) g (t) isThe model of the increase trend of the electricity sales volume represents the change trend of the electricity sales volume time sequence in the non-period, the trend of each data segment of the electricity sales volume time sequence in the distribution area changes along with the situation of the mutation points, S mutation points are added into the time sequence, and the position of a timestamp of each mutation point is Sj(j is more than or equal to 1 and less than or equal to S), and realizing piecewise linear fitting by adopting a piecewise logistic regression growth model:
Figure FDA0002559344020000024
Figure FDA0002559344020000025
wherein a (t) is a vector of the index function; r (t) is the expected capacity of the model; k + a (t)TThe increase rate of the electricity sales volume along with the change of time is the change value of the increase rate at the mutation point; m is an offset parameter, γ is a parameter vector that makes the function continuous, and t is a time variable;
2) s (t) is a seasonal trend model of electricity sales, and a model of a seasonal component is established based on a Fourier series:
Figure FDA0002559344020000031
Figure FDA0002559344020000032
where T is the period length of the time series, 2N represents the number of periods expected to be used in the model, N is the order of the Fourier transform, β ═ a1，b1，...，aN，bN]，an、bnIs a parameter to be estimated;
3) h (t) is a holiday trend model of the electricity sales, and if M holidays exist, a holiday component model is expressed as:
Figure FDA0002559344020000033
z(t)＝[1(t∈Di)，...，1(t∈DM)]
κ＝(κ1，...，κM)T
wherein, κ to Normal (o, v)2) Kappa represents the influence of holiday i and conforms to normal distribution, the index kappa is influenced by a holiday intensity index v, the value of the kappa is in direct proportion to the influence of holidays on electricity sales, and the kappa isMRepresenting the influence of the Mth holiday on the predicted value; diIndicating the time t contained in the window, 1 if t belongs to the vacation i (t ∈ D)i) 1 otherwise (t ∈ D)i)＝0；
4)tIs an error component, indicating an unpredicted fluctuation.
5. The accurate prediction method of power distribution station power selling amount based on modal GRU learning network as claimed in claim 1, characterized in that: in step S5, the GRU neural network includes an update gate and a reset gate, and has the following structure:
zt＝σ(W(z)xt+U(z)ht-1)
rt＝σ(W(r)xt+U(r)ht-1)
in the formula, ztTo refresh the door, rtTo reset the gate, xtFor input, htIs the output of the hidden layer; the GRU-based unit calculates h by the following formulat：
Figure FDA0002559344020000041
Figure FDA0002559344020000042
In the formula (I), the compound is shown in the specification,
Figure FDA0002559344020000043
is to input xtAnd past hidingLayer state ht-1Summarizing; u shape(z)、W(z)、U(r)、W(r)U and W are trainable parameter matrices.
6. The accurate prediction method of power distribution station power selling amount based on modal GRU learning network as claimed in claim 1, characterized in that: and training the GRU neural network by adopting the training set, and checking the prediction effect of the GRU neural network by adopting the testing set.
CN202010607362.2A 2020-06-29 2020-06-29 Accurate prediction method for power selling amount of power distribution station area based on modal GRU learning network Pending CN111784043A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010607362.2A CN111784043A (en) 2020-06-29 2020-06-29 Accurate prediction method for power selling amount of power distribution station area based on modal GRU learning network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010607362.2A CN111784043A (en) 2020-06-29 2020-06-29 Accurate prediction method for power selling amount of power distribution station area based on modal GRU learning network

Publications (1)

Publication Number Publication Date
CN111784043A true CN111784043A (en) 2020-10-16

Family

ID=72759886

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010607362.2A Pending CN111784043A (en) 2020-06-29 2020-06-29 Accurate prediction method for power selling amount of power distribution station area based on modal GRU learning network

Country Status (1)

Country Link
CN (1) CN111784043A (en)

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112883649A (en) * 2021-02-26 2021-06-01 润联软件系统(深圳)有限公司 Power load prediction method, system, computer equipment and storage medium
CN112990556A (en) * 2021-02-24 2021-06-18 江苏大学 User power consumption prediction method based on Prophet-LSTM model
CN113052398A (en) * 2021-04-21 2021-06-29 广州高谱技术有限公司 Power consumption prediction method and system based on variational modal decomposition
CN113450141A (en) * 2021-06-09 2021-09-28 重庆锦禹云能源科技有限公司 Intelligent prediction method and device based on electricity selling quantity characteristics of large-power customer groups
CN113554466A (en) * 2021-07-26 2021-10-26 国网四川省电力公司电力科学研究院 Short-term power consumption prediction model construction method, prediction method and device
CN113625164A (en) * 2021-08-02 2021-11-09 南京航空航天大学 Aviation generator fault feature extraction method, system, medium and computing device
CN115049169A (en) * 2022-08-16 2022-09-13 国网湖北省电力有限公司信息通信公司 Regional power consumption prediction method, system and medium based on combination of frequency domain and spatial domain
CN115759886A (en) * 2023-01-10 2023-03-07 佰聆数据股份有限公司 Three-dimensional factor-based electricity sales analysis method, device, equipment and medium
CN116204760A (en) * 2023-01-16 2023-06-02 海南师范大学 Drilling strain data anomaly extraction method based on GRU network
CN117495435A (en) * 2023-12-29 2024-02-02 国网浙江省电力有限公司营销服务中心 FIG-IRELM-based electricity sales interval prediction method and device

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110414724A (en) * 2019-07-10 2019-11-05 东软集团股份有限公司 For predicting method, apparatus, readable storage medium storing program for executing and the electronic equipment of power consumption
EP3567479A1 (en) * 2018-05-10 2019-11-13 Guang Dong Oppo Mobile Telecommunications Corp., Ltd. Method for establishing application prediction model, storage medium, and terminal
CN111193256A (en) * 2020-01-07 2020-05-22 上海电力大学 Power load prediction method based on variational modal decomposition and gated cyclic unit

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP3567479A1 (en) * 2018-05-10 2019-11-13 Guang Dong Oppo Mobile Telecommunications Corp., Ltd. Method for establishing application prediction model, storage medium, and terminal
CN110414724A (en) * 2019-07-10 2019-11-05 东软集团股份有限公司 For predicting method, apparatus, readable storage medium storing program for executing and the electronic equipment of power consumption
CN111193256A (en) * 2020-01-07 2020-05-22 上海电力大学 Power load prediction method based on variational modal decomposition and gated cyclic unit

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
YU LI ET AL.: "Prophet model and Gaussian process regression based user traffic prediction in wireless networks", 《SCIENCE CHINA PRESS AND SPRINGER-VERLAG GMBH GERMANY, PART OF SPRINGER NATURE 2020》 *
孟凡煦: "基于变分模态分解和极限学习机的短期负荷预测方法研究", 《中国优秀硕士学位论文全文数据库 工程科技II辑》 *
许东 等: "《基于MATLAB 6.x的系统分析与设计——神经网络》", 30 September 2008, 西安电子科技大学出版社 *

Cited By (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112990556A (en) * 2021-02-24 2021-06-18 江苏大学 User power consumption prediction method based on Prophet-LSTM model
CN112883649A (en) * 2021-02-26 2021-06-01 润联软件系统(深圳)有限公司 Power load prediction method, system, computer equipment and storage medium
CN112883649B (en) * 2021-02-26 2023-08-11 华润数字科技有限公司 Power load prediction method, system, computer equipment and storage medium
CN113052398A (en) * 2021-04-21 2021-06-29 广州高谱技术有限公司 Power consumption prediction method and system based on variational modal decomposition
CN113450141A (en) * 2021-06-09 2021-09-28 重庆锦禹云能源科技有限公司 Intelligent prediction method and device based on electricity selling quantity characteristics of large-power customer groups
CN113450141B (en) * 2021-06-09 2023-09-01 重庆锦禹云能源科技有限公司 Intelligent prediction method and device based on electricity sales quantity characteristics of large power customer group
CN113554466B (en) * 2021-07-26 2023-04-28 国网四川省电力公司电力科学研究院 Short-term electricity consumption prediction model construction method, prediction method and device
CN113554466A (en) * 2021-07-26 2021-10-26 国网四川省电力公司电力科学研究院 Short-term power consumption prediction model construction method, prediction method and device
CN113625164A (en) * 2021-08-02 2021-11-09 南京航空航天大学 Aviation generator fault feature extraction method, system, medium and computing device
CN115049169B (en) * 2022-08-16 2022-10-28 国网湖北省电力有限公司信息通信公司 Regional power consumption prediction method, system and medium based on combination of frequency domain and spatial domain
CN115049169A (en) * 2022-08-16 2022-09-13 国网湖北省电力有限公司信息通信公司 Regional power consumption prediction method, system and medium based on combination of frequency domain and spatial domain
CN115759886A (en) * 2023-01-10 2023-03-07 佰聆数据股份有限公司 Three-dimensional factor-based electricity sales analysis method, device, equipment and medium
CN116204760A (en) * 2023-01-16 2023-06-02 海南师范大学 Drilling strain data anomaly extraction method based on GRU network
CN116204760B (en) * 2023-01-16 2023-10-24 海南师范大学 Drilling strain data anomaly extraction method based on GRU network
CN117495435A (en) * 2023-12-29 2024-02-02 国网浙江省电力有限公司营销服务中心 FIG-IRELM-based electricity sales interval prediction method and device

Similar Documents

Publication Publication Date Title
CN111784043A (en) Accurate prediction method for power selling amount of power distribution station area based on modal GRU learning network
CN108256697B (en) Prediction method for short-term load of power system
WO2018214629A1 (en) Electricity sales projection method, device, and computer storage medium
Ahmad et al. A data-driven deep sequence-to-sequence long-short memory method along with a gated recurrent neural network for wind power forecasting
CN110705743B (en) New energy consumption electric quantity prediction method based on long-term and short-term memory neural network
Feng et al. The practical research on flood forecasting based on artificial neural networks
CN107622329A (en) The Methods of electric load forecasting of Memory Neural Networks in short-term is grown based on Multiple Time Scales
Agarwal et al. Runoff modelling through back propagation artificial neural network with variable rainfall-runoff data
CN112434787B (en) Terminal space energy consumption prediction method, medium and equipment based on total energy consumption of building
CN111182564A (en) Wireless link quality prediction method based on LSTM neural network
CN110570034B (en) Bus load prediction method based on multi-XGboost model fusion
CN112381673B (en) Park electricity utilization information analysis method and device based on digital twin
CN111709109A (en) Photovoltaic absorption capacity calculation method and device considering source-load time sequence correlation
CN110807508B (en) Bus peak load prediction method considering complex weather influence
Liu et al. Short‐term load forecasting based on LSTNet in power system
CN115186923A (en) Photovoltaic power generation power prediction method and device and electronic equipment
CN107909202B (en) Time sequence-based oil well liquid production integrated prediction method
CN109540089B (en) Bridge deck elevation fitting method based on Bayes-Kriging model
CN113393034A (en) Electric quantity prediction method of online self-adaptive OSELM-GARCH model
Li et al. A novel method of creating machine learning-based time series meta-models for building energy analysis
Zhang et al. Enhancing hourly heat demand prediction through artificial neural networks: a national level case study
CN113128666A (en) Mo-S-LSTMs model-based time series multi-step prediction method
CN114943328A (en) SARIMA-GRU time sequence prediction model based on BP neural network nonlinear combination
Yang et al. Probabilistic modeling of renewable energy source based on Spark platform with large‐scale sample data
CN114626581A (en) TCN-SVM combined electrical load prediction method based on VMD decomposition mechanism

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: 20201016

RJ01 Rejection of invention patent application after publication