CN113487062A - Power load prediction method based on periodic automatic encoder - Google Patents
Power load prediction method based on periodic automatic encoder Download PDFInfo
- Publication number
- CN113487062A CN113487062A CN202110597175.5A CN202110597175A CN113487062A CN 113487062 A CN113487062 A CN 113487062A CN 202110597175 A CN202110597175 A CN 202110597175A CN 113487062 A CN113487062 A CN 113487062A
- Authority
- CN
- China
- Prior art keywords
- power load
- sequence
- data
- time
- load data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 57
- 230000000737 periodic effect Effects 0.000 title claims abstract description 47
- 238000007781 pre-processing Methods 0.000 claims abstract description 5
- 230000008569 process Effects 0.000 claims description 20
- 230000002159 abnormal effect Effects 0.000 claims description 19
- 238000012545 processing Methods 0.000 claims description 15
- 238000004364 calculation method Methods 0.000 claims description 6
- 239000006185 dispersion Substances 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 6
- 238000004140 cleaning Methods 0.000 claims description 5
- 238000012937 correction Methods 0.000 claims description 3
- 210000002569 neuron Anatomy 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 238000002474 experimental method Methods 0.000 description 10
- 230000008859 change Effects 0.000 description 7
- 238000009826 distribution Methods 0.000 description 6
- 230000005611 electricity Effects 0.000 description 5
- 238000013507 mapping Methods 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 238000012549 training Methods 0.000 description 3
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 description 2
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification 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
- 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
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Marketing (AREA)
- General Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Public Health (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Water Supply & Treatment (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention relates to a power load prediction method based on a periodic automatic encoder, which comprises the following steps: collecting power load data of each transformer area to form a historical power time sequence; preprocessing the historical power time sequence, reconstructing the historical power time sequence through a trained periodic automatic encoder, and generating an embedded sequence of the historical power time sequence; and inputting the embedded sequence of the historical power time sequence into a trained predictor to obtain the predicted power load sequence data. Compared with the prior art, the method has the advantages of high prediction precision and the like.
Description
Technical Field
The invention relates to a power load prediction method, in particular to a power load prediction method based on a periodic automatic encoder.
Background
With the development of economic technology and the improvement of living standard of people, electric energy becomes essential secondary energy in production and life of people, and brings endless convenience to production and life of people.
With the frequent occurrence of extreme weather and the higher and higher requirements of people on living standard and living quality, the demand of people on electricity is increased. With the increase of household electricity consumption of residents, the distribution transformer area is heavily overloaded, especially, the overload condition of the electricity load of the distribution transformer area in winter is serious, the load peak-valley difference is large, especially, the electricity consumption level of the residents reaches the highest peak within one year during the spring festival, and the serious overload condition is easy to occur. The resident is greatly influenced by electricity limit, power failure and tripping in daily life.
The power distribution and utilization global full time sequence load data prediction is a technical means capable of effectively solving the problems and is an important basis for power grid planning, operation and energy-saving management, and the power distribution and utilization global full time sequence load data prediction enables a power system to be prepared in advance by predicting the load of the power system in advance, so that heavy overload of a transformer area is avoided.
However, when individual-level power load prediction is performed, because the global full load of the level has high volatility and uncertainty, the existing prediction method cannot overcome the problem, and the prediction accuracy is not high.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide a power load prediction method based on a periodic automatic encoder, which has high prediction precision.
The purpose of the invention can be realized by the following technical scheme:
a power load prediction method based on a periodic automatic encoder comprises the following steps:
collecting power load data of each transformer area to form a historical power time sequence;
preprocessing the historical power time sequence, reconstructing the historical power time sequence through a trained periodic automatic encoder, and generating an embedded sequence of the historical power time sequence;
inputting the generated embedded sequence into a trained predictor to obtain predicted power load sequence data;
the historical power time sequence xnIs as follows;
wherein N is the number of the station areas, M is the time range value, Qk,tPower load data representing k zones at time t;
the reconstruction process of the periodic automatic encoder comprises the following steps:
z(n,1)=sigmoid(W1xn+b1)
z(n,2)=sigmoid(W2z(n,1)+b2)
......
z(n,l)=sigmoid(Wlz(n,l-1)+bl)
......
z(n,L)=sigmoid(WLz(n,L-1)+bL)
wherein L is the number of hidden layers of the periodic automatic encoder, and WlAnd blRespectively the weight and the threshold of the hidden layer of the l layer,is the output of the first hidden layer, HlThe number of neurons in the first hidden layer,is the vector space of the first hidden layer, the output z of the last hidden layer(n,L)Namely an embedded sequence generated by a periodic automatic encoder according to a historical power time sequence;
since the individual level power load data has a periodic characteristic, the periodic automatic encoder is used for generating a historical power time sequence xnThe vector space is embedded, the embedded sequence is generated, the generated embedded sequence can capture the inherent periodic characteristics of the power load data, the uncertainty and the volatility of the power load data are filtered, the embedded sequence is input into a predictor, accurate prediction of the power load data of an individual level is achieved, and the prediction accuracy of the power load data is greatly improved.
Further, since the historical power time series share one period auto-encoder, i.e. share the same mapping, to mitigate overfitting, the loss function L of the period auto-encoder is trainedPAEThe root mean square error RMSE is adopted, and the calculation formula is as follows:
wherein T is the period of the historical power time sequence, and N is the stationThe number of the zones is such that,for the load value sequence of the nth station at time t,reconstruction of time t for periodic autoencoder output The expression of (a) is:
wherein, WreconstructAnd breconstructRespectively, the weights and thresholds of the output layers of the periodic auto-encoder.
Further, the formula of the predictor is as follows:
Xt,k=(Weekt,Montht,ADt,Weathert,k,Weatherdif(t,k),Loadbf(t,k))
wherein,for the load prediction value at the kth point of the prediction date t, S is a trained prediction model, thetat,kModel parameters, X, used for predicting the kth time point of day tt,kAs a vector of influencing factors, WeektAs week type influencing factors, MonthtFor month influencing factors, ADtWeather, a factor affecting the overall time trendt,kWeather as a factor influencing meteorological valuesdif(t,k)Relative to the previous dayWeather change influencing factor, Loadbf(t,k)The load value influence factor of the previous day;
the relevant factors influencing the load size are mainly the week type, month, overall time trend, weather value of the day, weather change of the day relative to the previous day, and load value of the previous day.
Further, the prediction model S is a multilayer perceptron MLP, and the reconstruction process is as follows:
h(n,1)=sigmoid(W'1z(n,L)+b'1)
h(n,2)=sigmoid(W'2h(n,1)+b'2)
......
h(n,l)=sigmoid(W'lh(n,l-1)+bl')
......
h(n,L)=sigmoid(W'L'h(n,L′-1)+b'L')
wherein L' is the number of hidden layers of the prediction model, h(n,l)Is the output of the l-th hidden layer, W'lAnd bl' are the weight and threshold of the l-th hidden layer, respectively.
Further, since the historical power time series share one period of the autoencoder, i.e. share the same mapping, to mitigate overfitting, the loss function L of the prediction model S is trainedMLPThe root mean square error RMSE is adopted, and the calculation formula is as follows:
wherein T is the period of the historical power time series, N is the number of the transformer areas,for the embedding sequence of the nth station zone at time t,output for predictive model SReconstruction at time t The expression of (a) is:
wherein, WforecastAnd bforecastRespectively, the weight and the threshold of the output layer of the prediction model.
Further, the pretreatment process comprises the following steps:
and sequentially performing data cleaning processing, data standardization processing and data noise reduction processing on the historical power time sequence.
Further, the data cleaning process includes:
judging whether the power load data meet an identification inequality, if so, judging that the power load data are normal, otherwise, judging that the power load data are abnormal;
correcting abnormal power load data by using an interpolation method;
the identification inequality is as follows:
P(|Qk,t-μk|>3σk)≤0.003
wherein, mukIs the mean value, σ, of the power load data of the kth station areakIs the standard deviation of the kth power load data.
Further, the interpolation method comprises:
correcting abnormal power load data by the following formula:
wherein,for the load of the abnormal point after interpolation correction, n1Selecting range value n of power load data for time period near abnormal power load data2Selection range value of the area near the area where the abnormal power load data is located, CiIs an interpolation coefficient of the time interval, BjAs interpolation coefficient of the station area, CiAnd BjSatisfies the following conditions:
Further, the data normalization processing procedure includes:
the method comprises the following steps of normalizing the power load data by using a dispersion normalization formula, wherein the dispersion normalization formula is as follows:
wherein, Q'k,tNamely, the normalized power load data at the time point of the k station zone t.
Further, the data denoising processing process includes:
and denoising the power load data by adopting wavelet transformation.
Compared with the prior art, the invention has the following beneficial effects:
the invention collects the power load data of each station area to form a historical power time sequence, preprocesses the historical power time sequence, reconstructs the historical power time sequence through a trained period automatic encoder to generate an embedded sequence of the historical power time sequence, inputs the generated embedded sequence into a trained predictor to obtain predicted power load sequence data and realizes the prediction of power load data of individual levels, because the power load data of the individual levels have obvious period characteristics, the period automatic encoder embeds the historical power time sequence into a vector space and generates the embedded sequence, the generated embedded sequence can capture the inherent period characteristics of the power load data and filter the uncertainty and the fluctuation of the power load data, and then inputs the embedded sequence into the predictor, the predictor is not easy to learn the characteristics of random fluctuating noise in the historical power time sequence, the training process is more stable and not easy to over-fit, and the prediction precision is greatly improved.
Drawings
FIG. 1 is a schematic diagram of a periodic auto-encoder;
FIG. 2 is a graph of the change in RMSE for the test set at 48 points predicted at 48 points;
FIG. 3 is a graph of the change in RMSE for the test set at 96 points predicted 48 points;
FIG. 4 is a flow chart of the method of the present invention.
Detailed Description
The invention is described in detail below with reference to the figures and specific embodiments. The present embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation manner and a specific operation process are given, but the scope of the present invention is not limited to the following embodiments.
A power load prediction method based on a periodic automatic encoder, as shown in fig. 4, includes:
1) collecting power load data of each transformer area to form a historical power time sequence;
2) preprocessing the historical power time sequence, reconstructing the historical power time sequence through a trained periodic automatic encoder PAE, and generating an embedded sequence of the historical power time sequence;
3) inputting the generated embedded sequence into a trained predictor to obtain predicted power load sequence data;
historical power time series xnIs as follows;
wherein N is the number of the station areas, M is the time range value, Qk,tPower load data representing k zones at time t;
as shown in fig. 1, the reconstruction process of the periodic automatic encoder is as follows:
z(n,1)=sigmoid(W1xn+b1)
z(n,2)=sigmoid(W2z(n,1)+b2)
......
z(n,l)=sigmoid(Wlz(n,l-1)+bl)
......
z(n,L)=sigmoid(WLz(n,L-1)+bL)
wherein L is the number of hidden layers of the periodic automatic encoder, and WlAnd blRespectively the weight and the threshold of the hidden layer of the l layer,is the output of the first hidden layer, HlThe number of neurons in the first hidden layer,is the vector space of the first hidden layer, the output z of the last hidden layer(n,L)Namely an embedded sequence generated by a periodic automatic encoder according to a historical power time sequence;
since the power load data has a periodic characteristic, the periodic autoencoder will history the power time series xnThe vector space is embedded, the embedded sequence is generated, the generated embedded sequence can capture the inherent periodic characteristics of the power load data, the uncertainty and the volatility of the power load data are filtered, the embedded sequence is input into a predictor, accurate prediction of the power load data of an individual level is achieved, and the prediction accuracy of the power load data is greatly improved.
One cycle automatic due to historical power time series sharingEncoders, i.e. sharing the same mapping, train the loss function L of periodic auto-encoders to mitigate overfittingPAEThe root mean square error RMSE is adopted, and the calculation formula is as follows:
wherein T is the period of the historical power time series, N is the number of the transformer areas,for the load value sequence of the nth station at time t,reconstruction of time t for periodic autoencoder output The expression of (a) is:
wherein, WreconstructAnd breconstructRespectively, the weights and thresholds of the output layers of the periodic auto-encoder.
The formula of the predictor is:
Xt,k=(Weekt,Montht,ADt,Weathert,k,Weatherdif(t,k),Loadbf(t,k))
wherein,for the load prediction value at the kth point of the prediction date t, S is a trained prediction model, thetat,kModel parameters, X, used for predicting the kth time point of day tt,kAs a vector of influencing factors, WeektAs week type influencing factors, MonthtFor month influencing factors, ADtWeather, a factor affecting the overall time trendt,kWeather as a factor influencing meteorological valuesdif(t,k)Load as a factor influencing weather changes from the previous daybf(t,k)The load value influence factor of the previous day;
the relevant factors influencing the load size are mainly the week type, month, overall time trend, weather value of the day, weather change of the day relative to the previous day, and load value of the previous day.
The prediction model S is a multilayer perceptron MLP, and the reconstruction process is as follows:
h(n,1)=sigmoid(W'1z(n,L)+b'1)
h(n,2)=sigmoid(W'2h(n,1)+b'2)
......
h(n,l)=sigmoid(W'lh(n,l-1)+bl')
......
h(n,L)=sigmoid(W'L'h(n,L′-1)+b'L')
wherein L' is the number of hidden layers of the prediction model, h(n,l)Is the output of the l-th hidden layer, W'lAnd bl' are the weight and threshold of the l-th hidden layer, respectively.
Since the historical power time series share one period of the automatic encoder, i.e. share the same mapping, to mitigate overfitting, the loss function L of the prediction model S is trainedMLPThe root mean square error RMSE is adopted, and the calculation formula is as follows:
wherein T isThe period of the historical power time series, N being the number of zones,for the embedding sequence of the nth station zone at time t,reconstruction of time t for prediction model S output The expression of (a) is:
wherein, WforecastAnd bforecastRespectively, the weight and the threshold of the output layer of the prediction model.
The pretreatment process comprises the following steps:
and sequentially performing data cleaning processing, data standardization processing and data noise reduction processing on the historical power time sequence.
The data cleaning processing process comprises the following steps:
judging whether the power load data meet an identification inequality, if so, judging that the power load data are normal, otherwise, judging that the power load data are abnormal;
correcting abnormal power load data by using an interpolation method;
assuming that the historical power time series obeys normal distribution, the identification inequality is as follows:
P(|Qk,t-μk|>3σk)≤0.003
wherein, mukIs the mean value, σ, of the power load data of the kth station areakThe standard deviation of the power load data of the kth station area is shown;
abnormal values and missing values in the historical power time series may be identified and corrected by the data cleansing process.
The interpolation method comprises the following steps:
correcting abnormal power load data by the following formula:
wherein,for the load of the abnormal point after interpolation correction, n1Selecting range value n of power load data for time period near abnormal power load data2Selection range value of the area near the area where the abnormal power load data is located, CiIs an interpolation coefficient of the time interval, BjAs interpolation coefficient of the station area, CiAnd BjSatisfies the following conditions:
The data standardization processing process comprises the following steps:
the method comprises the following steps of (1) carrying out standardization processing on power load data by using a dispersion standardization formula, wherein the dispersion standardization formula is as follows:
wherein Q isk′,tNamely, the normalized power load data at the time point of the k station zone t.
The data denoising processing process comprises the following steps:
and denoising the power load data by adopting wavelet transformation.
Experiment one
The original historical power time series and the embedded series produced by the periodic automatic encoder are respectively input into a predictor to carry out a comparison experiment, in the experiment, for each household user, 48 points of the day 1 are firstly used for predicting 48 points of the day 2, and then 96 points of the day 1 and the day 2 are used for predicting 48 points of the day 3, and the results of the experiment are shown in the following table:
table 1 experimental results of experiment one
Wherein RMSE is the root mean square error, MAPE is the mean absolute percentage error, MAPE is defined as:
wherein, YtAndrespectively the actual value and the predicted value at the time t, n3Representing a total number of predicted power load data;
the graph of the change of the RMSE of the test set at 48 points is predicted as shown in fig. 2, the graph of the change of the RMSE of the test set at 48 points is predicted as shown in fig. 3 at 96 points, the horizontal axes of fig. 2 and 3 are the number of training rounds, and the vertical axis is the value of the RMSE, as can be seen from table 1, fig. 2 and fig. 3, the embedded sequence generated by the periodic automatic encoder can filter the uncertainty and the fluctuation of the power load data, therefore, compared with the original historical power time sequence, the embedded sequence generated by the periodic automatic encoder is used for predicting the future power load data, the predictor is not easy to learn the characteristics of the noise which fluctuates randomly in the historical power time sequence, the training process is more stable and not easy to over-fit, and the prediction accuracy is greatly improved.
Experiment two
In order to show that the embedded sequence generated by the periodic automatic encoder can capture the inherent periodic characteristics of the global full-consumption power distribution load and further improve the prediction precision,
the prediction was performed using a periodic autoencoder and a traditional embedding method, respectively, and comparative experiments were performed on the EDRP dataset:
the period automatic encoder embeds the historical power time sequence into a vector space according to periods, the traditional embedding method does not generate the embedded sequence according to periods, and other influence factors are kept consistent;
for each family user, firstly, a periodic automatic encoder uses 48 points on the 1 st day to predict 48 points on the 2 nd day, and the traditional embedding method uses the first 10 points to predict 48 points; the periodic autoencoder then used the 96 points for day 1 and day 2 of the user to predict 48 points for day 3, and the conventional embedding method used the first 100 points to predict the last 48 points, with the results of the experiment shown in the table below.
TABLE 2 Experimental results of experiment two
It can be seen that the conventional embedding method does not capture the cycle characteristics which are important for power load prediction, and although the historical power utilization information is increased, the prediction accuracy is not high, compared with the conventional embedding method, the embedded sequence generated by the cycle automatic encoder can capture the cycle characteristics inherent to the power load, and further can obtain higher accuracy.
According to the power load prediction method based on the periodic automatic encoder, the periodic automatic encoder PAE is used for embedding the power time sequence into the vector space according to the period, the embedded sequence generated by the PAE can capture the inherent periodic characteristics of the power load, and the prediction precision is greatly improved.
The foregoing detailed description of the preferred embodiments of the invention has been presented. It should be understood that numerous modifications and variations could be devised by those skilled in the art in light of the present teachings without departing from the inventive concepts. Therefore, the technical solutions available to those skilled in the art through logic analysis, reasoning and limited experiments based on the prior art according to the concept of the present invention should be within the scope of protection defined by the claims.
Claims (10)
1. A power load prediction method based on a periodic automatic encoder is characterized by comprising the following steps:
collecting power load data of each transformer area to form a historical power time sequence;
preprocessing the historical power time sequence, reconstructing the historical power time sequence through a trained periodic automatic encoder, and generating an embedded sequence of the historical power time sequence;
inputting the embedded sequence of the historical power time sequence into a trained predictor to obtain predicted power load sequence data;
the historical power time sequence is as follows;
wherein N is the number of the station areas, M is the time range value, Qk,tPower load data representing k zones at time t;
the reconstruction process of the periodic automatic encoder comprises the following steps:
z(n,1)=sigmoid(W1xn+b1)
z(n,2)=sigmoid(W2z(n,1)+b2)
......
z(n,l)=sigmoid(Wlz(n,l-1)+bl)
......
z(n,L)=sigmoid(WLz(n,L-1)+bL)
wherein x isnFor historical power time series, L is the number of hidden layers of the periodic automatic encoder, WlAnd blRespectively the weight and the threshold of the hidden layer of the l layer,is the output of the first hidden layer, HlThe number of neurons in the first hidden layer and the output z of the last hidden layer(n,L)I.e., an embedded sequence generated by a periodic autoencoder from a historical power time sequence.
2. The method according to claim 1, wherein the loss function L of the periodic automatic encoder is trainedPAEThe root mean square error RMSE is adopted, and the calculation formula is as follows:
wherein T is the period of the historical power time series, N is the number of the transformer areas,for the load value sequence of the nth station at time t,reconstruction of time t for periodic autoencoder outputThe expression of (a) is:
wherein, WreconstructAnd breconstructRespectively, the weights and thresholds of the output layers of the periodic auto-encoder.
3. The method of claim 1, wherein the predictor has the formula:
Xt,k=(Weekt,Montht,ADt,Weathert,k,Weatherdif(t,k),Loadbf(t,k))
wherein,for the load prediction value at the kth point of the prediction date t, S is a trained prediction model, thetat,kModel parameters, X, used for predicting the kth time point of day tt,kAs a vector of influencing factors, WeektAs week type influencing factors, MonthtFor month influencing factors, ADtWeather, a factor affecting the overall time trendt,kWeather as a factor influencing meteorological valuesdif(t,k)Load as a factor influencing weather changes from the previous daybf(t,k)Is the influence factor of the load value of the previous day.
4. The method according to claim 3, wherein the prediction model S is a multi-layer perceptron MLP, and the reconstruction process is as follows:
h(n,1)=sigmoid(W'1z(n,L)+b'1)
h(n,2)=sigmoid(W'2h(n,1)+b'2)
......
h(n,l)=sigmoid(W'lh(n,l-1)+b'l)
......
h(n,L)=sigmoid(W'L'h(n,L′-1)+b'L')
wherein L' is the number of hidden layers of the prediction model, h(n,l)Is the output of the l-th hidden layer, W'lAnd b'lThe weight and the threshold of the hidden layer of the l-th layer are respectively.
5. The method according to claim 3, wherein the loss function L of the prediction model S is trainedMLPThe root mean square error RMSE is adopted, and the calculation formula is as follows:
wherein T is the period of the historical power time series, N is the number of the transformer areas,for the embedding sequence of the nth station zone at time t,reconstruction of time t for prediction model S outputThe expression of (a) is:
wherein, WforecastAnd bforecastRespectively, the weight and the threshold of the output layer of the prediction model.
6. The method of claim 1, wherein the preprocessing comprises:
and sequentially performing data cleaning processing, data standardization processing and data noise reduction processing on the historical power time sequence.
7. The method of claim 6, wherein the data cleansing process comprises:
judging whether the power load data meet an identification inequality, if so, judging that the power load data are normal, otherwise, judging that the power load data are abnormal;
correcting abnormal power load data by using an interpolation method;
the identification inequality is as follows:
P(|Qk,t-μk|>3σk)≤0.003
wherein, mukIs the mean value, σ, of the power load data of the kth station areakIs the standard deviation of the kth power load data.
8. The method of claim 7, wherein the interpolation comprises:
correcting abnormal power load data by the following formula:
wherein,for the load of the abnormal point after interpolation correction, n1Selecting range value n of power load data for time period near abnormal power load data2Selection range value of the area near the area where the abnormal power load data is located, CiIs an interpolation coefficient of the time interval, BjAs interpolation coefficient of the station area, CiAnd BjSatisfies the following conditions:
9. The method of claim 6, wherein the data normalization process comprises:
the method comprises the following steps of normalizing the power load data by using a dispersion normalization formula, wherein the dispersion normalization formula is as follows:
wherein, Q'k,tNamely, the normalized power load data at the time point of the k station zone t.
10. The method of claim 6, wherein the data de-noising process comprises:
and denoising the power load data by adopting wavelet transformation.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110597175.5A CN113487062A (en) | 2021-05-31 | 2021-05-31 | Power load prediction method based on periodic automatic encoder |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110597175.5A CN113487062A (en) | 2021-05-31 | 2021-05-31 | Power load prediction method based on periodic automatic encoder |
Publications (1)
Publication Number | Publication Date |
---|---|
CN113487062A true CN113487062A (en) | 2021-10-08 |
Family
ID=77933687
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110597175.5A Pending CN113487062A (en) | 2021-05-31 | 2021-05-31 | Power load prediction method based on periodic automatic encoder |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113487062A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114511159A (en) * | 2022-04-20 | 2022-05-17 | 广东电网有限责任公司佛山供电局 | Power load probability prediction method and system based on conditional variational self-encoder |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106778841A (en) * | 2016-11-30 | 2017-05-31 | 国网上海市电力公司 | The method for building up of abnormal electricity consumption detection model |
CN109470946A (en) * | 2018-09-18 | 2019-03-15 | 中国电力科学研究院有限公司 | A kind of generating equipment fault detection method and system |
CN111578444A (en) * | 2019-02-19 | 2020-08-25 | 珠海格力电器股份有限公司 | Air conditioner fault prediction method and device, storage medium and air conditioner |
CN111985701A (en) * | 2020-07-31 | 2020-11-24 | 国网上海市电力公司 | Power utilization prediction method based on power supply enterprise big data model base |
-
2021
- 2021-05-31 CN CN202110597175.5A patent/CN113487062A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106778841A (en) * | 2016-11-30 | 2017-05-31 | 国网上海市电力公司 | The method for building up of abnormal electricity consumption detection model |
CN109470946A (en) * | 2018-09-18 | 2019-03-15 | 中国电力科学研究院有限公司 | A kind of generating equipment fault detection method and system |
CN111578444A (en) * | 2019-02-19 | 2020-08-25 | 珠海格力电器股份有限公司 | Air conditioner fault prediction method and device, storage medium and air conditioner |
CN111985701A (en) * | 2020-07-31 | 2020-11-24 | 国网上海市电力公司 | Power utilization prediction method based on power supply enterprise big data model base |
Non-Patent Citations (1)
Title |
---|
田英杰等: "基于时间序列嵌入的电力负荷预测方法", 计算机应用与软件, pages 55 - 60 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114511159A (en) * | 2022-04-20 | 2022-05-17 | 广东电网有限责任公司佛山供电局 | Power load probability prediction method and system based on conditional variational self-encoder |
CN114511159B (en) * | 2022-04-20 | 2022-07-12 | 广东电网有限责任公司佛山供电局 | Power load probability prediction method and system based on conditional variational self-encoder |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111080032A (en) | Load prediction method based on Transformer structure | |
CN102478584B (en) | Wind power station wind speed prediction method based on wavelet analysis and system thereof | |
CN106709816B (en) | Non-parametric regression analysis-based power load abnormal data identification and correction method | |
CN114792156B (en) | Photovoltaic output power prediction method and system based on curve characteristic index clustering | |
CN113554466A (en) | Short-term power consumption prediction model construction method, prediction method and device | |
CN111915083B (en) | Wind power prediction method and prediction system based on time layered combination | |
CN116073436B (en) | Capacity optimization control method for photovoltaic new energy power system | |
CN115065078B (en) | Energy storage capacity configuration method and system in micro-grid environment | |
CN109636054A (en) | Solar energy power generating amount prediction technique based on classification and error combination prediction | |
CN112288140A (en) | Keras-based short-term power load prediction method, storage medium and equipment | |
Liu | FDI and employment by industry: A co-integration study | |
Wei et al. | Short-term load forecasting using detrend singular spectrum fluctuation analysis | |
CN114140176A (en) | Adjustable capacity prediction method and device for load aggregation platform | |
CN112183877A (en) | Photovoltaic power station fault intelligent diagnosis method based on transfer learning | |
Sun et al. | The Short‐Term Load Forecasting Using an Artificial Neural Network Approach with Periodic and Nonperiodic Factors: A Case Study of Tai′ an, Shandong Province, China | |
CN113487062A (en) | Power load prediction method based on periodic automatic encoder | |
CN113935244A (en) | Method and system for predicting short-term power load of urban rural distribution transformer | |
Grandon et al. | Electricity demand forecasting with hybrid statistical and machine learning algorithms: Case study of ukraine | |
Wu et al. | Forecast of short‐term electricity price based on data analysis | |
CN113449933B (en) | Regional medium-term load prediction method and device based on clustering electric quantity curve decomposition | |
CN116191412A (en) | Power load prediction method | |
CN107563573B (en) | Solar power generation capacity prediction method based on self-adaptive learning hybrid model | |
CN114004405B (en) | Photovoltaic power prediction method and system based on Elman neural network and satellite cloud image | |
CN118395384B (en) | Multi-dimensional decomposition and intelligent fusion power load prediction method and related equipment | |
Hu et al. | Research on intelligent prediction of time-of-use price on power sale |
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 |