CN115759474A - Multitask load prediction method and system based on Autoformer model and XGboost model - Google Patents

Multitask load prediction method and system based on Autoformer model and XGboost model Download PDF

Info

Publication number
CN115759474A
CN115759474A CN202211573438.XA CN202211573438A CN115759474A CN 115759474 A CN115759474 A CN 115759474A CN 202211573438 A CN202211573438 A CN 202211573438A CN 115759474 A CN115759474 A CN 115759474A
Authority
CN
China
Prior art keywords
model
load
predicted
prediction
xgboost
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
CN202211573438.XA
Other languages
Chinese (zh)
Inventor
蒋志鹏
张建宇
戴帅夫
李莉
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.)
Beijing Jiuqi Technology Co ltd
Original Assignee
Beijing Jiuqi Technology Co ltd
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 Beijing Jiuqi Technology Co ltd filed Critical Beijing Jiuqi Technology Co ltd
Priority to CN202211573438.XA priority Critical patent/CN115759474A/en
Publication of CN115759474A publication Critical patent/CN115759474A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a multitask load prediction method and a multitask load prediction system based on an auto-former model and an XGboost model, and the method comprises the following steps: acquiring historical data, and respectively extracting time characteristics, meteorological characteristics, population characteristics and resident load data so as to construct historical time sequence data; predicting the population quantity and the resident load based on an auto-former model to obtain a population quantity predicted value and a resident maximum load predicted value in a time period; predicting the maximum load of residents in the same time period based on the XGboost model by using the time characteristics, the meteorological characteristics and the population quantity prediction value predicted by the Autoformer model; and fusing the maximum load of the residents predicted by the Autoformer model and the XGboost model to obtain a final prediction result. The system comprises a feature extraction module, an auto-former model prediction module, an XGboost model prediction module and a prediction result fusion module; the invention solves the problem that the load prediction can not be carried out by effectively utilizing time sequence data in the prior art.

Description

Multitask load prediction method and system based on Autoformer model and XGboost model
Technical Field
The invention relates to the technical field of electric load prediction, in particular to a multitask load prediction method and a multitask load prediction system based on an auto-former model and an XGboost model.
Background
The power consumption prediction is an important research field, and the accurate circuit load consumption prediction can effectively help the power company to distribute and schedule electric energy and improve the use efficiency of resources.
The power consumption prediction is a typical timing problem, is influenced by multiple factors such as time, weather, holidays, population mobility and the like, and has the uncertainty and complexity of a typical complex system; especially in recent years, the limited population mobility brings more challenges to the prediction of the power consumption. For medium-and long-term prediction, a larger prediction time span further amplifies the uncertainty factor of the system.
In general, the timing prediction method can be classified into a conventional statistical method, a machine learning-based method, and a deep learning-based method. The traditional statistical representative algorithm comprises an ARMA (autoregressive moving average) model, an ARIMA (autoregressive integrated moving average) model, an AR model and the like, and the traditional statistical algorithm can only learn time sequence characteristics and cannot utilize additional characteristics; machine learning based algorithms can exploit more features, enabling learning the temporal dynamics of a time series in a data-driven manner. The XGboost algorithm is a typical representative of a machine learning algorithm, has the advantages of high training speed, high prediction accuracy and the like, and has excellent performance on classification and regression tasks, but the XGboost is a tree-based model which has the defect of weak extrapolation capability, and the prediction value of the XGboost algorithm does not exceed the range of the maximum and minimum values of data in a training set for prediction problems. In recent years, deep learning algorithms such as RNN, LSTM and Tranformer models are also beginning to be applied to the timing prediction problem, and the prediction algorithms based on deep learning can obtain strong learning representation capabilities from a large amount of data, and the generalization capability of the models is strong, wherein the auto-former model uses an autocorrelation mechanism to replace a point attention mechanism, tries to further find the similarity of subsequences, and is more advantageous in long-time timing data prediction. However, the main problem of the prediction algorithm based on deep learning is the black box characteristic of the model, and the time characteristic of time series data cannot be effectively utilized.
Therefore, an urgent need exists in the art for a method and a system for providing an auto former model and XGBoost model-based multitask load prediction method capable of effectively utilizing time series data to predict accurate data.
Disclosure of Invention
In view of the above, the invention provides a multitask load prediction method and a multitask load prediction system based on an auto-former model and an XGBoost model, which solve the problem that time series data cannot be effectively utilized to perform load prediction in the prior art.
In order to achieve the purpose, the invention adopts the following technical scheme:
a multitask load prediction method based on an Autoformer model and an XGboost model comprises the following steps:
acquiring historical data, and respectively extracting time characteristics, meteorological characteristics, population characteristics and resident load data so as to construct historical time sequence data;
predicting the population quantity and the resident load based on an auto-former model to obtain a population quantity predicted value and a resident maximum load predicted value in a time period;
predicting the maximum load of residents in the same time period by using the time characteristics, meteorological characteristics and population quantity prediction values predicted by an auto-former model based on an XGboost model;
and fusing the maximum load of the residents predicted by the Autoformer model and the XGboost model to obtain a final prediction result of the maximum load of the residents, and outputting the final prediction result of the maximum load of the residents and a predicted value of the population number.
Preferably, the step of predicting the population number and the load of the residents based on the auto former model includes:
s11, recording historical time sequence data as x, and performing sequence decomposition on the x to obtain a season item x des And a trend term x det ；
S12, coding x through N auto-provider coding layers of the auto-provider coder, and outputting final coded data
Figure BDA0003988399630000031
The l layer coding method comprises the following steps:
Figure BDA0003988399630000032
Figure BDA0003988399630000033
wherein the input of the coding layer is
Figure BDA0003988399630000034
Figure BDA0003988399630000035
Obtained by performing a convolution operation on x, _ is the trend part of the elimination,
Figure BDA0003988399630000036
represents the output of the l-th coding layer,
Figure BDA0003988399630000037
represents the ith sequence decomposition performed at the l-th layer; feed forward represents feed forward operation, and Auto-Correlation is the autocorrelation coefficient of the sequence;
s13. Tendency item x is decoded through M decoding layers of an auto-former decoder det And the season item x des Decoding is performed and final decoded data is output
Figure BDA0003988399630000038
And
Figure BDA0003988399630000039
the decoding method of the l layer comprises the following steps:
Figure BDA00039883996300000310
Figure BDA00039883996300000311
Figure BDA00039883996300000312
Figure BDA00039883996300000313
in the formula (I), the compound is shown in the specification,
Figure BDA00039883996300000314
the output of the l-th layer is represented,
Figure BDA00039883996300000315
by the season item x des Obtaining after convolution;
Figure BDA00039883996300000316
Figure BDA00039883996300000317
the ith sequence decomposition operation of the periodic term and the trend term in the ith layer is respectively represented, W l,i I e {1,2,3} represents
Figure BDA00039883996300000318
Projection of (2);
s14, according to the final decoded data
Figure BDA00039883996300000319
And
Figure BDA00039883996300000320
obtaining predicted values through multi-layer perceptron MLP
Figure BDA00039883996300000321
Preferably, the temporal features are subjected to one-hot encoding processing, and the remaining features are subjected to maximum and minimum normalization processing, so as to construct historical time series data.
Preferably, for a discrete time-sequential process { x t }, autocorrelation coefficients
Figure BDA00039883996300000322
The specific calculation method comprises the following steps:
Figure BDA0003988399630000041
in the formula, L is the length of the sequence, and L approaches infinity.
Preferably, the variance uncertainty is used as a reference in both the resident load prediction and population count tasks to weight the loss of the different tasks, with the predicted resident load and population count being expressed as y load And y pop The predicted target is represented as p (y) load ,y pop If (x | θ)), the loss function is:
Figure BDA0003988399630000042
where theta is the parameter of the learning,
Figure BDA0003988399630000043
respectively, variance of load of residents and population quantity, theta load And theta pop The parameters needed to learn for the load and population tasks, respectively.
Preferably, the specific contents of predicting the maximum load of residents in the same time period by using the time characteristic, the meteorological characteristic and the population quantity predicted by the auto-former model based on the XGboost model include:
s21, training an XGboost model according to the characteristic data;
s22, acquiring time characteristics and weather characteristics of a time period to be predicted and population number predicted values predicted by an auto-former model, and performing maximum and minimum normalization processing on the population number predicted values and the weather characteristics to obtain a test set;
s23, forecasting is carried out according to the XGboost training model and the characteristics of the test set, and the residential load forecasting value based on the XGboost model is obtained
Figure BDA0003988399630000044
Preferably, the concrete contents of fusing the maximum loads of the residents predicted by the auto model and the XGboost model include:
Figure BDA0003988399630000045
wherein, the first and the second end of the pipe are connected with each other,
Figure BDA0003988399630000051
for the resident maximum load predicted by the auto-former model,
Figure BDA0003988399630000052
for the resident maximum load, ε, predicted by the XGBoost model xgboost And ε autoformer The training errors for XGboost and Autoformer are expressed separately.
A multitask load prediction system based on an Autoformer model and an XGboost model comprises the following components: the system comprises a feature extraction module, an auto-former model prediction module, an XGboost model prediction module and a prediction result fusion module;
the characteristic extraction module is used for acquiring historical data and respectively extracting time characteristics, meteorological characteristics, population characteristics and resident load data so as to construct historical time sequence data;
the automatic former model prediction module is used for predicting the population quantity and the resident load based on the automatic former model to obtain a population quantity predicted value and a resident maximum load predicted value in a time period;
the XGboost model prediction module is used for predicting the maximum load of residents in the same time period by using the time characteristics, meteorological characteristics and population quantity prediction value predicted by the auto-former model based on the XGboost model;
and the prediction result fusion module is used for fusing the maximum load of the residents predicted by the Autoformer model and the XGboost model to obtain a final prediction result.
Compared with the prior art, the invention discloses a multitask load prediction method and a system based on an auto-former model and an XGboost model, and the invention has the advantages that:
(1) Based on an auto-former model, an autocorrelation mechanism is used for replacing an attention mechanism with point connection, and the similarity relation between sequence levels is fully learned, so that the periodicity of a time sequence can be responded, and the accuracy of long-term prediction can be effectively improved;
(2) The XGboost can automatically learn the most relevant characteristics of the prediction task, and the model has strong interpretability;
(3) The co-variance uncertainty combined multitasking loss function is used to simultaneously learn the population prediction task and the population load task. In an actual application scene, the establishment of a quantitative population-load conduction prediction model can be used for predicting the inflow peak value and the load pressure peak value level and the time period of festival population in the early stage of a major festival, so that quantitative support and experience reference are provided for festival protection power supply work in a key area.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.
Fig. 1 is a schematic flow diagram of a multitask load prediction method based on an auto-former model and an XGBoost model according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The embodiment of the invention discloses a multitask load prediction method based on an auto-former model and an XGboost model, which comprises the following steps as shown in figure 1:
acquiring historical data, and respectively extracting time characteristics, meteorological characteristics, population characteristics and resident load data so as to construct historical time sequence data;
predicting the population quantity and the resident load based on an auto-former model to obtain a predicted value of the population quantity and a predicted value of the maximum load of residents in a time period;
predicting the maximum load of residents in the same time period by using the time characteristics, meteorological characteristics and population quantity prediction values predicted by an auto-former model based on an XGboost model;
and fusing the maximum load of the residents predicted by the Autoformer model and the XGboost model to obtain a final prediction result of the maximum load of the residents, and outputting the final prediction result of the maximum load of the residents and a predicted value of the population number.
In the present embodiment, the temporal characteristics: the current day is the day of the year, week, month, quarter, weekend, holiday, year, etc.; meteorological features: the highest air temperature and the lowest air temperature of the current day and whether the current day rains or not; population characteristics: the number of people per day in the current area; resident load data: the current maximum load of the residential electricity in the area.
x pop And x load Are respectively historyPopulation number, maximum load of residents, and predicted load data y of residents load It is a one-dimensional vector with length O, and this embodiment relates to medium-long term prediction, so O is generally between 7 and 14.
In order to further implement the above solution, the step of predicting the population and the load of the residents based on the auto-former model includes:
s11, recording historical time sequence data as x, and performing sequence decomposition on the x to obtain a season item x des And a trend term x det ；
S12, coding x through N auto-provider coding layers of the auto-provider coder, and outputting final coded data
Figure BDA0003988399630000071
The l layer coding method comprises the following steps:
Figure BDA0003988399630000072
Figure BDA0003988399630000073
wherein the input of the coding layer is
Figure BDA0003988399630000074
Figure BDA0003988399630000075
Obtained by performing a convolution operation on x, _ is the trend part of the cancellation,
Figure BDA0003988399630000076
the output of the l-th coding layer is indicated,
Figure BDA0003988399630000077
indicates the ith sequence decomposition performed at the l-th layer; feed forward represents feed forward operation, and Auto-Correlation is the autocorrelation coefficient of the sequence;
S13.trend term x through M decoding layer pairs of auto-former decoder det And the season item x des Decoding and outputting final decoded data
Figure BDA0003988399630000081
And
Figure BDA0003988399630000082
the decoding method of the l layer comprises the following steps:
Figure BDA0003988399630000083
Figure BDA0003988399630000084
Figure BDA0003988399630000085
Figure BDA0003988399630000086
in the formula (I), the compound is shown in the specification,
Figure BDA0003988399630000087
the output of the l-th layer is represented,
Figure BDA0003988399630000088
by seasonal item x des Obtaining after convolution;
Figure BDA0003988399630000089
Figure BDA00039883996300000810
the ith sequence decomposition operation of the periodic term and the trend term in the ith layer is respectively represented, W l,i I e {1,2,3} represents
Figure BDA00039883996300000811
Projection of (2);
s14, according to the final decoded data
Figure BDA00039883996300000812
And
Figure BDA00039883996300000813
obtaining predicted values through multi-layer perceptron MLP
Figure BDA00039883996300000814
In this embodiment, the input of the coding layer
Figure BDA00039883996300000815
The convolution operation is carried out on x by utilizing a Conv1d interface of the Pythroch, the feed forward represents the feed forward operation and is obtained through a Linear interface of the Pythroch,
Figure BDA00039883996300000816
is a Conv1d interface pair x using Pythrch des And performing convolution operation to obtain the target.
In order to further implement the scheme, the time characteristics are subjected to one-hot coding processing, and the rest characteristics are subjected to maximum and minimum normalization processing, so that historical time series data are constructed.
To further implement the above scheme, for a discrete time-sequential process { x } t }, autocorrelation coefficients
Figure BDA00039883996300000817
The specific calculation method comprises the following steps:
Figure BDA00039883996300000818
in the formula, L is the length of the sequence, and L approaches infinity.
In this embodiment, the autocorrelation coefficient may be calculated according to a fast fourier transform fft interface and an inverse fourier transform irfft interface of the Pytorch.
To further implement the above scheme, the variance uncertainty is used as a reference in both the resident load prediction and population count tasks, denoted as y for the predicted resident load and population count, respectively, to weight the loss for the different tasks load And y pop The predicted target is represented as p (y) load ,y pop If (x | θ)), the loss function is:
Figure BDA0003988399630000091
where theta is the parameter of the learning,
Figure BDA0003988399630000092
respectively, variance of load of residents and population quantity, theta load And theta pop The parameters needed to learn for the load and population tasks, respectively.
In order to further implement the above scheme, the specific contents of predicting the maximum load of residents in the same time period by using the population quantity predicted value predicted by the XGboost model using the time characteristic, the meteorological characteristic and the auto-former model include:
s21, training an XGboost model according to the characteristic data;
s22, acquiring time characteristics and weather characteristics of a time period to be predicted and population number predicted values predicted by an auto-former model, and performing maximum and minimum normalization processing on the population number predicted values and the weather characteristics to obtain a test set;
s23, forecasting is carried out according to the characteristics of the XGboost training model and the test set to obtain the residential load forecasting value based on the XGboost model
Figure BDA0003988399630000093
In order to further implement the above solution, the concrete contents of fusing the maximum loads of residents predicted by the auto-former model and the XGboost model include:
Figure BDA0003988399630000094
wherein, the first and the second end of the pipe are connected with each other,
Figure BDA0003988399630000095
for the resident maximum load predicted by the auto-former model,
Figure BDA0003988399630000096
is the resident maximum load, ε, predicted by the XGBOST model xgboost And ε autoformer The training errors for XGboost and Autoformer are expressed separately.
In this embodiment, the number of layers of the auto encoder is 2, the number of decoding layers is 1, and the dimension of the hidden layer is 512. The optimizer is ADAM, the learning rate is 0.0001, the size of batch is 16, and the main parameters of the XGboost model are as follows: maximum depth max _ depth is 10, learning rate learning _ rate is 0.01, n _ estimators =1100, enhancer is gbtree, objective learning function reg: and (4) linear.
A multitask load prediction system based on an Autoformer model and an XGboost model comprises the following components: the system comprises a feature extraction module, an auto-former model prediction module, an XGboost model prediction module and a prediction result fusion module;
the characteristic extraction module is used for acquiring historical data and respectively extracting time characteristics, meteorological characteristics, population characteristics and resident load data so as to construct historical time sequence data;
the automatic former model prediction module is used for predicting the population quantity and the resident load based on the automatic former model to obtain a predicted value of the population quantity and a predicted value of the maximum load of residents in a time period;
the XGboost model prediction module is used for predicting the maximum load of residents in the same time period based on the XGboost model by utilizing time characteristics, meteorological characteristics and the population quantity prediction value predicted by the auto-former model;
and the prediction result fusion module is used for fusing the maximum load of the residents predicted by the Autoformer model and the XGboost model to obtain a final prediction result.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. The device disclosed by the embodiment corresponds to the method disclosed by the embodiment, so that the description is simple, and the relevant points can be referred to the method part for description.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (8)

1. A multitask load prediction method based on an Autoformer model and an XGboost model is characterized by comprising the following steps:
acquiring historical data, and respectively extracting time characteristics, meteorological characteristics, population characteristics and resident load data so as to construct historical time sequence data;
predicting the population quantity and the resident load based on an auto-former model to obtain a predicted value of the population quantity and a predicted value of the maximum load of residents in a time period;
predicting the maximum load of residents in the same time period based on the XGboost model by using the time characteristics, the meteorological characteristics and the population quantity prediction value predicted by the Autoformer model;
and fusing the maximum load of the residents predicted by the Autoformer model and the XGboost model to obtain a final prediction result of the maximum load of the residents, and outputting the final prediction result of the maximum load of the residents and a predicted value of the population number.
2. The multitask load prediction method based on the auto-former model and the XGBoost model according to claim 1, wherein the step of predicting the population quantity and the residential load based on the auto-former model comprises:
s11, recording historical time sequence data as x, and performing sequence decomposition on the x to obtain a season item x des And a trend term x det ；
S12, coding x through N auto-former coding layers of the auto-former coder, and outputting final coded data
Figure FDA0003988399620000011
The l layer coding method comprises the following steps:
Figure FDA0003988399620000012
Figure FDA0003988399620000013
in the formula, the input of the coding layer is
Figure FDA0003988399620000014
Figure FDA0003988399620000015
Obtained by performing a convolution operation on x, _ is the trend part of the cancellation,
Figure FDA0003988399620000016
represents the output of the l-th coding layer,
Figure FDA0003988399620000017
indicates the ith sequence decomposition performed at the l-th layer; feed forward represents feed forward operation, and Auto-Correlation is the autocorrelation coefficient of the sequence;
s13. Tendency item x is decoded through M decoding layers of an auto-former decoder det And the season item x des Decoding and outputting the final decodingData of
Figure FDA0003988399620000021
And
Figure FDA0003988399620000022
the decoding method of the l layer comprises the following steps:
Figure FDA0003988399620000023
Figure FDA0003988399620000024
Figure FDA0003988399620000025
Figure FDA0003988399620000026
in the formula (I), the compound is shown in the specification,
Figure FDA0003988399620000027
the output of the l-th layer is represented,
Figure FDA0003988399620000028
by the season item x des Obtaining after convolution;
Figure FDA0003988399620000029
Figure FDA00039883996200000210
the ith sequence decomposition operation of the periodic term and the trend term in the ith layer is respectively represented, W l,i I ∈ {1,2,3} represents
Figure FDA00039883996200000211
Projection of (2);
s14, according to the final decoded data
Figure FDA00039883996200000212
And
Figure FDA00039883996200000213
obtaining predicted values through multi-layer perceptron MLP
Figure FDA00039883996200000214
3. The multitask load prediction method based on the Autoformer model and the XGboost model according to claim 2, wherein the time characteristics are subjected to one-hot coding processing, and the other characteristics are subjected to maximum and minimum normalization processing, so that historical time sequence data are constructed.
4. The method of claim 2, wherein { x } is a discrete time sequence process t }, autocorrelation coefficients
Figure FDA00039883996200000215
The specific calculation method comprises the following steps:
Figure FDA00039883996200000216
in the formula, L is the length of the sequence, and L approaches infinity.
5. The method as claimed in claim 1, wherein the variance uncertainty is used as a reference in the two tasks of the resident load prediction and the population quantity to weight the loss of different tasks, and the resident load and the population are predictedThe quantities being respectively denoted by y load And y pop The predicted target is represented as p (y) load ,y pop If (x | θ)), the loss function is:
Figure FDA0003988399620000031
where theta is the parameter of the learning,
Figure FDA0003988399620000032
respectively, variance of load of residents and population quantity, theta load And theta pop The parameters needed to learn for the load and population tasks, respectively.
6. The multitask load prediction method based on the Autoformer model and the XGboost model as claimed in claim 1, wherein the concrete contents of predicting the maximum load of residents in the same time period based on the XGboost model by using the time characteristics, the meteorological characteristics and the population quantity predicted by the Autoformer model comprise:
s21, training an XGboost model according to the characteristic data;
s22, acquiring time characteristics and weather characteristics of a time period to be predicted and population number predicted values predicted by an auto-former model, and carrying out maximum and minimum normalization processing on the population number predicted values and the weather characteristics to obtain a test set;
s23, forecasting is carried out according to the characteristics of the XGboost training model and the test set to obtain the residential load forecasting value based on the XGboost model
Figure FDA0003988399620000033
7. The multitask load prediction method based on the Autoformer model and the XGboost model according to claim 1, wherein the concrete contents of fusing the maximum loads of residents predicted by the Autoformer model and the XGboost model comprise:
Figure FDA0003988399620000034
wherein the content of the first and second substances,
Figure FDA0003988399620000035
for the resident maximum load predicted by the auto-former model,
Figure FDA0003988399620000036
for the resident maximum load, ε, predicted by the XGBoost model xgboost And ε autoformer The training errors for XGboost and auto former are expressed separately.
8. A multitask load prediction system based on an Autoformer model and an XGboost model is characterized by comprising the following steps: the system comprises a feature extraction module, an auto-former model prediction module, an XGboost model prediction module and a prediction result fusion module;
the characteristic extraction module is used for acquiring historical data and respectively extracting time characteristics, meteorological characteristics, population characteristics and resident load data so as to construct historical time sequence data;
the automatic former model prediction module is used for predicting the population quantity and the resident load based on the automatic former model to obtain a population quantity predicted value and a resident maximum load predicted value in a time period;
the XGboost model prediction module is used for predicting the maximum load of residents in the same time period based on the XGboost model by utilizing time characteristics, meteorological characteristics and population quantity prediction values predicted by an auto-former model;
and the prediction result fusion module is used for fusing the maximum load of the residents predicted by the auto-former model and the XGboost model to obtain a final prediction result.
CN202211573438.XA 2022-12-08 2022-12-08 Multitask load prediction method and system based on Autoformer model and XGboost model Pending CN115759474A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211573438.XA CN115759474A (en) 2022-12-08 2022-12-08 Multitask load prediction method and system based on Autoformer model and XGboost model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211573438.XA CN115759474A (en) 2022-12-08 2022-12-08 Multitask load prediction method and system based on Autoformer model and XGboost model

Publications (1)

Publication Number Publication Date
CN115759474A true CN115759474A (en) 2023-03-07

Family

ID=85344598

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211573438.XA Pending CN115759474A (en) 2022-12-08 2022-12-08 Multitask load prediction method and system based on Autoformer model and XGboost model

Country Status (1)

Country Link
CN (1) CN115759474A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116128168A (en) * 2023-04-17 2023-05-16 南京信息工程大学 Weather prediction method based on causal expansion convolution and Autoformer

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116128168A (en) * 2023-04-17 2023-05-16 南京信息工程大学 Weather prediction method based on causal expansion convolution and Autoformer

Similar Documents

Publication Publication Date Title
CN110610280B (en) Short-term prediction method, model, device and system for power load
CN111079989B (en) DWT-PCA-LSTM-based water supply amount prediction device for water supply company
CN110619420B (en) Attention-GRU-based short-term residential load prediction method
CN110490385A (en) The unified prediction of electric load and thermic load in a kind of integrated energy system
CN109934395B (en) Multi-hydropower-region short-term power load prediction method based on time-sharing and regional meteorological data
CN110380444B (en) Capacity planning method for distributed wind power orderly access to power grid under multiple scenes based on variable structure Copula
WO2024046137A1 (en) Power prediction model construction method for multi-energy combined power generation system and power prediction method for multi-energy combined power generation system
CN110309984B (en) Non-invasive load identification and short-term user behavior prediction method
CN115759474A (en) Multitask load prediction method and system based on Autoformer model and XGboost model
CN116865261B (en) Power load prediction method and system based on twin network
CN115271253B (en) Method and device for constructing water-wind-solar power generation power prediction model and storage medium
CN115222106A (en) User day-ahead load prediction method of self-adaptive model
KR20220019376A (en) Power consumption prediction system of residential and commercial building using hybrid convolution neural network and method thereof
CN111652422A (en) Heat supply system load prediction method, device and system based on building classification
CN108846505B (en) Multidimensional checking method and equipment for renewable energy grid-connected consumption information
Tuyen et al. A combination of novel hybrid deep learning model and quantile regression for short‐term deterministic and probabilistic PV maximum power forecasting
CN116404637A (en) Short-term load prediction method and device for electric power system
CN113837486B (en) RNN-RBM-based distribution network feeder long-term load prediction method
CN109034453A (en) A kind of Short-Term Load Forecasting Method based on multiple labeling neural network
Haque et al. A hybrid intelligent framework for wind power forecasting engine
Xu et al. A combination model based on EEMD-PE and echo state network for chaotic time series prediction
CN113487062A (en) Power load prediction method based on periodic automatic encoder
CN111915070B (en) Prediction method for k-nearest neighbor day-ahead wind speed by adopting atmospheric pressure spatial correlation
Liu et al. SFCL: electricity consumption forecasting of CNN-LSTM based on similar filter
CN117407681B (en) Time sequence data prediction model establishment method based on vector clustering

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