CN113627658A - Short-term electricity sales amount prediction method based on generalized cross-correlation entropy gating circulation unit - Google Patents

Short-term electricity sales amount prediction method based on generalized cross-correlation entropy gating circulation unit Download PDF

Info

Publication number
CN113627658A
CN113627658A CN202110840949.2A CN202110840949A CN113627658A CN 113627658 A CN113627658 A CN 113627658A CN 202110840949 A CN202110840949 A CN 202110840949A CN 113627658 A CN113627658 A CN 113627658A
Authority
CN
China
Prior art keywords
data
prediction
correlation entropy
generalized
historical
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
CN202110840949.2A
Other languages
Chinese (zh)
Inventor
段建东
方帅
王鹏
马文涛
侯泽权
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.)
Xian University of Technology
Original Assignee
Xian University 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 Xian University of Technology filed Critical Xian University of Technology
Priority to CN202110840949.2A priority Critical patent/CN113627658A/en
Publication of CN113627658A publication Critical patent/CN113627658A/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
    • G06N3/04—Architecture, e.g. interconnection topology
    • 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
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201—Market modelling; Market analysis; Collecting market data
    • G06Q30/0202—Market predictions or forecasting for commercial activities
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Strategic Management (AREA)
  • General Physics & Mathematics (AREA)
  • Economics (AREA)
  • Health & Medical Sciences (AREA)
  • Development Economics (AREA)
  • General Business, Economics & Management (AREA)
  • Data Mining & Analysis (AREA)
  • Human Resources & Organizations (AREA)
  • Marketing (AREA)
  • Finance (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Accounting & Taxation (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Biophysics (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Game Theory and Decision Science (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Tourism & Hospitality (AREA)
  • Molecular Biology (AREA)
  • Water Supply & Treatment (AREA)
  • Primary Health Care (AREA)
  • Public Health (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a short-term electricity sales amount prediction method based on a generalized cross-correlation entropy gating circulation unit, which comprises the following steps: collecting data, and supplementing missing data in historical electricity consumption data; taking the historical hourly power consumption and the temperature corresponding to the historical hourly power consumption as a training sample set of a prediction model, and constructing feature input of the prediction model; data standardization; the method comprises the steps that a gated cycle unit model is selected for predicting the hourly power consumption, and a generalized maximum correlation entropy criterion corresponding to a generalized cross-correlation entropy is used for replacing a mean square error criterion in a gated cycle unit to serve as a cost function of a prediction model; optimizing parameters lambda and alpha of the gate control cycle unit model by a K-fold cross validation and grid optimization method; and predicting the electricity selling amount in the hour time scale, and evaluating the prediction result by selecting the RMSE index. The power selling amount prediction method has good prediction performance under the condition that the power selling amount data has outliers or noises and under the condition that the power selling amount data is distributed in a non-Gaussian manner.

Description

Short-term electricity sales amount prediction method based on generalized cross-correlation entropy gating circulation unit
Technical Field
The invention belongs to the technical field of electric quantity prediction of an electric power system, and particularly relates to a short-term electricity sales quantity prediction method based on a generalized cross-correlation entropy gating circulation unit.
Background
Under a new round of power system innovation, the transaction forms of electricity selling companies participating in the power market mainly include medium and long term transaction and spot transaction, if the deviation between the user power consumption declared in advance by the electricity selling companies and the actual user power consumption is large, deviation assessment needs to be accepted, and fine payment needs to be paid to a power transaction center. Therefore, the method for predicting the electric quantity to improve the prediction accuracy has important engineering application value.
The traditional electric quantity prediction problem mainly aims at large areas with more historical data, and the prediction technology is perfected; the problem of electric quantity prediction under new electricity change is mainly that for multiple or single users, the prediction area is small, the prediction time interval is small, the influence of random factors is large, and strong randomness, nonlinearity, time-varying property and non-Gaussian property of data distribution are reflected. For electric quantity prediction, a lot of research is carried out by scholars at home and abroad, and models of the scholars can be roughly divided into two types: traditional predictive models and machine learning models. The traditional prediction models comprise an AR comprehensive model, a moving average model, an autoregressive time series model, a gray prediction model and the like, the theoretical basis of the models is a linear model, the stationarity of the time series is required to be high, and the data regression capability is weak. The machine learning algorithm can well process various influence factors and nonlinear data, and is widely applied to electric quantity prediction. However, the prediction methods are designed based on a Mean Squared Error (MSE) cost function, which only considers the second moment of the Error distribution and is highly effective for predicting data with gaussian characteristics. However, data of actual electric quantity prediction often has non-gaussian and non-linear characteristics, so that it is very important to develop a prediction method suitable for the non-gaussian and non-linear working condition.
Disclosure of Invention
The invention aims to provide a short-term electricity sales amount prediction method based on a generalized cross-correlation entropy gating circulation unit, and solves the problems that the prediction precision of non-Gaussian non-linear electricity sales data is not high in the existing prediction technology, and the demand of electricity sales amount prediction precision when an electricity sales company conducts electricity transaction is difficult to meet.
The technical scheme adopted by the invention is as follows: the short-term electricity sales amount prediction method based on the generalized cross-correlation entropy gating circulation unit comprises the following specific operation steps:
step 1, data preprocessing
Collecting data, and supplementing missing data in historical electricity consumption data;
step 2, constructing a training sample set
Taking the historical hourly power consumption and the temperature corresponding to the historical hourly power consumption as a training sample set of a prediction model, and constructing characteristic input of the power prediction model;
step 3, standardizing data
The data mainly used in the electric quantity prediction model comprises historical hour power consumption data and temperature data corresponding to the historical hour power consumption data, and the two data are subjected to standardization processing in order to reduce the influence of larger magnitude difference of the two data on a prediction result;
step 4, a gated cyclic unit GRU model is selected to predict the hourly power consumption, and a generalized maximum correlation entropy criterion GMCC corresponding to the generalized cross-correlation entropy is used for replacing a mean square error criterion in the gated cyclic unit as a cost function of the power prediction model aiming at non-Gaussian characteristics of a power sales prediction error;
step 5, optimizing parameters lambda and alpha of the generalized cross-correlation entropy gating circulation unit model through a K-fold cross validation and grid optimization method;
and 6, predicting the electricity sales amount of the hour time scale by using the generalized cross-correlation entropy gating circulation unit prediction model to obtain a prediction result, and evaluating the prediction result by selecting an RMSE index.
The invention is also characterized in that:
the specific process of the data preprocessing in the step 1 is as follows:
collecting data, supplementing missing data in historical electricity consumption data of electricity selling users, and correcting the missing data by using an equation (1):
Figure BDA0003177550900000031
wherein, tiThe real value of the historical electricity selling data at the moment i, N is the number of samples,
Figure BDA0003177550900000032
the correction value for the missing data at time i,
Figure BDA0003177550900000033
is the average value of the historical electricity sales data,
Figure BDA0003177550900000034
xi and zeta are weight coefficients.
The specific process of constructing the training sample set in the step 2 is as follows:
taking historical hourly power consumption and temperature corresponding to the historical hourly power consumption as a training sample set of a prediction model, and constructing a characteristic input Pearson correlation coefficient calculation formula of the prediction model by utilizing the Pearson correlation coefficient as shown in formula (2):
Figure BDA0003177550900000035
wherein t is historical hour power consumption data, x is training sample set data comprising historical hour power consumption data and temperature data corresponding to the time, E (-) represents expectation, and D (-) represents variance.
Step 3, the specific process of data standardization is as follows:
the data mainly used in the electric quantity prediction model comprises historical hour power consumption data and temperature data corresponding to the historical hour power consumption data, and in order to reduce the influence of larger magnitude difference of the two data on the prediction result, the two data are subjected to standardization treatment, and the standardization formula is as shown in formula (3):
Figure BDA0003177550900000036
wherein x isminIs the minimum value of this class of data, xmaxIs the maximum value, x, of this class of dataiIs the true value of the data.
The step 4 is as follows:
the method comprises the steps that a gated cyclic unit GRU model is selected to predict the hourly power consumption, and a generalized maximum correlation entropy criterion GMCC corresponding to generalized cross-correlation entropy is used for replacing a mean square error criterion in the gated cyclic unit as a cost function of a prediction model aiming at non-Gaussian characteristics of a prediction error of the power sales;
the GRU network is a deep learning model, and the network feed-forward process is as follows:
Figure BDA0003177550900000041
wherein x istInput at time t; w and U are corresponding weights; sigma is a Sigmoid activation function; tan h (·) is a hyperbolic tangent function; z is a radical oft,rtOutputs of the refresh gate and the reset gate, respectively; h istThe network hidden state at the moment t is also input at the moment t +1, and the short-term memory function is realized;
Figure BDA0003177550900000042
respectively in cell state, and has long-term memory function; y istPredicting and outputting for the time t;
using a generalized maximum correlation entropy criterion GMCC corresponding to the generalized cross-correlation entropy as a cost function of a gating cycle unit GRU, establishing a GMCCGRU prediction model, wherein an expression of the generalized maximum correlation entropy criterion corresponding to the generalized maximum correlation entropy is shown in an expression (5):
Figure BDA0003177550900000043
where N is the number of training samples, yiAs model predicted value, tiIs an actual value, γα,λIs a normalization constant, λ>0 is the kernel parameter of the generalized Gaussian density function, α>0 is a shape parameter and the formula has a maximum value when the predicted value is equal to the actual value;
therefore, an updating formula of each weight parameter in the gated cyclic unit network can be obtained:
Figure BDA0003177550900000051
wherein x istFor the input at time t, W, U are the corresponding weights, zt,rtOutputs of the refresh gate and the reset gate, respectively, htFor the implicit state of the network at time t,
Figure BDA0003177550900000052
for the cell state, J is the cost penalty of the network, i.e., the generalized maximum correlation entropy criterion, ytPredicted output for time t, ttFor the desired output at time t, tanh (·), tanh' (·) is the hyperbolic tangent function and its derivatives, C is a penalty factor, and
Figure BDA0003177550900000053
the specific process of the step 5 is as follows:
optimizing key parameters lambda and alpha of the generalized cross-correlation entropy gated cyclic unit model by a K-fold cross validation and grid optimization method
And (3) searching parameters by using grid optimization: giving possible value ranges of the parameters lambda and alpha according to the sample, then taking a plurality of values in the value range of each parameter, and combining every two values to form a parameter pair related to the parameters lambda and alpha, thereby selecting the parameter pair which enables the generalized maximum correlation entropy to be maximum; measuring the generalization ability of the parameters by using a K-fold cross validation method: dividing the existing data set into K subsets, making each subset respectively perform a test set, and making other subsets perform a training set to verify the generalization capability of the parameters for K times.
The step 6 is as follows:
predicting the electricity sales amount of the hour time scale by using a generalized cross-correlation entropy gating circulation unit prediction model to obtain a prediction result, and evaluating the prediction result by selecting an RMSE index:
Figure BDA0003177550900000061
the invention has the beneficial effects that:
aiming at the characteristics of small prediction range, strong randomness, time sequence and the like of the electric quantity prediction, a gate control cycle unit model is selected to predict the electric quantity of a user. The traditional gating circulation unit adopts a root mean square error cost function with global measurement characteristics, only the second-order distance of prediction error distribution is considered, the prediction effectiveness of stable data with Gaussian characteristics is high, and the electricity sales data are usually in non-Gaussian distribution. Aiming at the problem, the generalized maximum correlation entropy criterion with locality measurement characteristics is adopted to replace the cost function root mean square error criterion of a classical gating circulation unit, and a prediction model based on the generalized cross-correlation entropy gating circulation unit is established to improve the effectiveness of the prediction of the electricity sales quantity. The method is applied to the power selling amount prediction of the power system for the first time, can effectively predict the power selling amount of the user, and has key theoretical significance and actual engineering value.
Drawings
FIG. 1 is a general flow diagram of the prediction method of the present invention;
FIG. 2 is a diagram illustrating the results of an embodiment of the prediction method of the present invention.
Detailed Description
The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.
The invention relates to a short-term electricity sales amount prediction method based on a generalized cross-correlation entropy gating circulation unit, which is implemented according to the following steps with reference to FIG. 1:
step 1, data preprocessing
And collecting data and supplementing missing data in the historical electricity consumption data of the electricity selling users. The missing data is corrected by the formula (1).
Figure BDA0003177550900000071
Wherein, tiThe real value of the historical electricity selling data at the moment i, N is the number of samples,
Figure BDA0003177550900000072
the correction value for the missing data at time i,
Figure BDA0003177550900000073
is the average value of the historical electricity sales data,
Figure BDA0003177550900000074
xi and zeta are weight coefficients.
Step 2, constructing a training sample set
Taking historical hourly power consumption and temperature corresponding to the historical hourly power consumption as a training sample set of a prediction model, and constructing characteristic input of the prediction model by utilizing a Pearson correlation coefficient, wherein a Pearson correlation coefficient calculation formula is as follows (2):
Figure BDA0003177550900000075
wherein t is historical hour power consumption data, x is training sample set data comprising historical hour power consumption data and temperature data corresponding to the historical hour power consumption data, E (-) represents expectation, and D (-) represents variance;
step 3, standardizing data
The data mainly used in the electric quantity prediction model comprises historical hour power consumption data and temperature data corresponding to the historical hour power consumption data, and in order to reduce the influence of larger magnitude difference of the two data on the prediction result, the two data are subjected to standardization treatment, and the standardization formula is as shown in formula (3):
Figure BDA0003177550900000076
wherein x isminIs the minimum value of this class of data, xmaxIs the maximum value, x, of this class of dataiIs the true value of the data.
And 4, aiming at short-term electricity quantity prediction of an hour time scale, the prediction range is small, the nonlinear randomness is strong, the characteristics of time sequence and the like are realized, and a Gated Recurrent Unit (GRU) model is selected to predict the hour electricity consumption. Aiming at non-Gaussian characteristics of the prediction error of the electricity sales quantity, a Generalized Maximum correlation entropy Criterion (GMCC) corresponding to the Generalized cross-correlation entropy is used for replacing a mean square error Criterion in a gating cycle unit as a cost function of the prediction model.
The GRU network is a deep learning model, and the network feed-forward process is as follows:
Figure BDA0003177550900000081
wherein x istInput at time t; w and U are corresponding weights; sigma is a Sigmoid activation function; tan h (·) is a hyperbolic tangent function; z is a radical oft,rtOutputs of the refresh gate and the reset gate, respectively; h istThe network hidden state at the moment t is also input at the moment t +1, and the short-term memory function is realized;
Figure BDA0003177550900000082
respectively in cell state, and has long-term memory function; y istPredicting and outputting for the time t;
the cost function mean square error criterion of the traditional gating circulation unit only considers the second-order distance of prediction error distribution, is a global similarity measurement criterion, has higher prediction effectiveness on stable data with Gaussian characteristics, but the short-term electric quantity prediction is greatly influenced by random factors, and the error distribution of the short-term electric quantity prediction often has characteristics of non-Gaussian, non-linearity and strong fluctuation, so the algorithm has certain limitation when being applied to the short-term electric quantity prediction problem. Aiming at the problem, a GMCCGRU prediction model is established by using a generalized maximum correlation entropy criterion GMCC corresponding to the generalized cross-correlation entropy as a cost function of a gated cyclic unit GRU. The expression of the generalized maximum correlation entropy criterion corresponding to the generalized maximum correlation entropy is shown in (5):
Figure BDA0003177550900000083
where N is the number of training samples, yiAs model predicted value, tiIs an actual value, γα,λIs a normalization constant, λ>0 is the kernel parameter of the generalized Gaussian density function, α>0 is a shape parameter, and the equation has a maximum value when the predicted value is equal to the actual value.
Therefore, an updating formula of each weight parameter in the gated cyclic unit network can be obtained:
Figure BDA0003177550900000091
wherein x istFor the input at time t, W, U are the corresponding weights, zt,rtOutputs of the refresh gate and the reset gate, respectively, htFor the implicit state of the network at time t,
Figure BDA0003177550900000092
in a cellular state, ytFor prediction output at time t, J is cost loss of network, namely generalized maximum correlation entropy criterion, ttFor the desired output at time t, tanh (·), tanh' (·) is the hyperbolic tangent function and its derivative, C is a penalty factor, and
Figure BDA0003177550900000093
and 5, optimizing key parameters lambda and alpha of the generalized cross-correlation entropy gating circulation unit model by a K-fold cross validation and grid optimization method.
And (3) searching parameters by using grid optimization: and giving possible value ranges of the parameters lambda and alpha according to the sample, then taking a plurality of values in the value range of each parameter, and combining every two values to form a parameter pair related to the parameters lambda and alpha, thereby selecting the parameter pair which maximizes the generalized maximum correlation entropy. Measuring the generalization ability of the parameters by using a K-fold cross validation method: dividing the existing data set into K subsets, making each subset respectively perform a test set, and making other subsets perform a training set to verify the generalization capability of the parameters for K times.
Step 6, predicting the electricity sales amount of the hour time scale by using a generalized cross-correlation entropy gating circulation unit prediction model to obtain a prediction result, and evaluating the prediction result by selecting an RMSE index:
Figure BDA0003177550900000101
examples
Step 1, data preprocessing. Collecting electricity consumption data of a user of an electricity selling company in the calendar history hour from 5 months and 1 day in 2020 to 6 months and 19 months in 2020, and supplementing missing data in the electricity consumption data of the electricity selling user. The missing data is corrected by the formula (1).
And 2, constructing a training sample set. The electricity consumption and the temperature corresponding to 18 days of 1/2020/6/2020/18 are used as the training sample set of the prediction model, and the feature input of the prediction model is constructed by using the formula (2), and in this case, the electricity consumption and the temperature corresponding to the prediction hour 24 hours before the prediction hour are selected as the feature input.
And step 3, standardizing data. The data mainly used in the electric quantity prediction model comprises historical hour power consumption data and temperature data corresponding to the hour, and in order to reduce the influence of large magnitude difference of the two data on the prediction result, the two data are subjected to standardization processing, and the historical electricity selling quantity data and the temperature data are subjected to normalization processing by using an application formula (3).
And 4, training the generalized cross-correlation entropy gating circulation unit prediction model by using a training sample set, and selecting the power consumption 24 hours before the prediction hour and the temperature corresponding to the prediction hour as characteristic inputs during training. The prediction model network weights W, U are updated using equation (6).
And 5, optimizing key parameters lambda and alpha of the generalized cross-correlation entropy gating circulation unit model through a K-fold cross validation and grid optimization method to obtain a parameter value enabling the generalized maximum correlation entropy value to be maximum.
And 6, testing the trained model by using the data of the test set. And predicting 24-hour electricity sales of 6, 19 and 6 in 2020 by using the determined parameters and the updated network weight, and evaluating the effectiveness of the prediction model by using an indicator RMSE.
And 7, comparing the generalized cross-correlation entropy gated cycle unit prediction model with the gated cycle unit prediction model, and comparing the prediction result with the prediction error shown in the table 1.
TABLE 1
Figure BDA0003177550900000111
As can be seen from Table 1, the method (i.e., the generalized cross-correlation entropy gated cyclic unit method) of the invention has higher prediction precision and is more effective, and the prediction accuracy can meet the actual requirements.
The method disclosed by the invention is used for predicting the electricity sales amount by a short-term electricity sales amount prediction method based on the generalized cross-correlation entropy gating circulation unit, and the gating circulation unit model is selected to predict the electricity sales amount of the user aiming at the characteristics of small prediction range, strong randomness, time sequence and the like of the electricity prediction. The cost function of the gating cycle unit is root mean square error, the root mean square error is a global measurement criterion, only the second-order distance of prediction error distribution is considered, the prediction effectiveness of stable data with Gaussian characteristics is high, the electricity selling data is usually non-Gaussian distribution, and a classic gating cycle unit prediction model has certain limitation on the problem of electricity prediction. The generalized maximum correlation entropy criterion corresponding to the generalized cross-correlation entropy is a local similarity measurement criterion, is insensitive to outliers and noise in electricity selling data, and is high in effectiveness page of non-Gaussian distribution data prediction. The method is applied to the power selling amount prediction of the power system for the first time, can effectively predict the power selling amount of the user, and has key theoretical significance and actual engineering value.

Claims (7)

1. The short-term electricity sales amount prediction method based on the generalized cross-correlation entropy gating circulation unit is characterized by comprising the following steps of:
step 1, data preprocessing
Collecting data, and supplementing missing data in historical electricity consumption data;
step 2, constructing a training sample set
Taking the historical hourly power consumption and the temperature corresponding to the historical hourly power consumption as a training sample set of a prediction model, and constructing characteristic input of the power prediction model;
step 3, standardizing data
The data mainly used in the electric quantity prediction model comprises historical hour power consumption data and temperature data corresponding to the historical hour power consumption data, and the two data are subjected to standardization processing in order to reduce the influence of larger magnitude difference of the two data on a prediction result;
step 4, a gated cyclic unit GRU model is selected to predict the hourly power consumption, and a generalized maximum correlation entropy criterion GMCC corresponding to the generalized cross-correlation entropy is used for replacing a mean square error criterion in the gated cyclic unit as a cost function of the power prediction model aiming at non-Gaussian characteristics of a power sales prediction error;
step 5, optimizing parameters lambda and alpha of the generalized cross-correlation entropy gating circulation unit model through a K-fold cross validation and grid optimization method;
and 6, predicting the electricity sales amount of the hour time scale by using the generalized cross-correlation entropy gating circulation unit prediction model to obtain a prediction result, and evaluating the prediction result by selecting an RMSE index.
2. The method for predicting the short-term electricity sales amount based on the generalized cross-correlation entropy gated cyclic unit according to claim 1, wherein the specific process of the data preprocessing in the step 1 is as follows:
collecting data, supplementing missing data in historical electricity consumption data of electricity selling users, and correcting the missing data by using an equation (1):
Figure FDA0003177550890000011
wherein, tiThe real value of the historical electricity selling data at the moment i, N is the number of samples,
Figure FDA0003177550890000021
the correction value for the missing data at time i,
Figure FDA0003177550890000022
is the average value of the historical electricity sales data,
Figure FDA0003177550890000025
xi and zeta are weight coefficients.
3. The method for predicting the short-term electricity sales amount based on the generalized cross-correlation entropy gated cyclic unit according to claim 1, wherein the specific process of constructing the training sample set in the step 2 is as follows:
taking historical hourly power consumption and temperature corresponding to the historical hourly power consumption as a training sample set of a prediction model, and constructing a characteristic input Pearson correlation coefficient calculation formula of the prediction model by utilizing the Pearson correlation coefficient as shown in formula (2):
Figure FDA0003177550890000023
wherein t is historical hour power consumption data, x is training sample set data comprising historical hour power consumption data and temperature data corresponding to the time, E (-) represents expectation, and D (-) represents variance.
4. The method for predicting the short-term electricity sales amount based on the generalized cross-correlation entropy gated cyclic unit as claimed in claim 1, wherein the specific process of step 3 data standardization is as follows:
the data mainly used in the electric quantity prediction model comprises historical hour power consumption data and temperature data corresponding to the historical hour power consumption data, and in order to reduce the influence of larger magnitude difference of the two data on the prediction result, the two data are subjected to standardization treatment, and the standardization formula is as shown in formula (3):
Figure FDA0003177550890000024
wherein x isminIs the minimum value of this class of data, xmaxIs the maximum value, x, of this class of dataiIs the true value of the data.
5. The method for predicting the short-term electricity sales amount based on the generalized cross-correlation entropy gated cyclic unit according to claim 1, wherein the step 4 is as follows:
the method comprises the steps that a gated cyclic unit GRU model is selected to predict the hourly power consumption, and a generalized maximum correlation entropy criterion GMCC corresponding to generalized cross-correlation entropy is used for replacing a mean square error criterion in the gated cyclic unit as a cost function of a prediction model aiming at non-Gaussian characteristics of a prediction error of the power sales;
the GRU network is a deep learning model, and the network feed-forward process is as follows:
Figure FDA0003177550890000031
wherein x istInput at time t; w and U are corresponding weights; sigma is a Sigmoid activation function; tan h (·) is a hyperbolic tangent function; z is a radical oft,rtOutputs of the refresh gate and the reset gate, respectively; h istThe network hidden state at the moment t is also input at the moment t +1, and the short-term memory function is realized;
Figure FDA0003177550890000032
respectively in cell state, and has long-term memory function; y istPredicting and outputting for the time t;
using a generalized maximum correlation entropy criterion GMCC corresponding to the generalized cross-correlation entropy as a cost function of a gating cycle unit GRU, establishing a GMCCGRU prediction model, wherein an expression of the generalized maximum correlation entropy criterion corresponding to the generalized maximum correlation entropy is shown in an expression (5):
Figure FDA0003177550890000033
where N is the number of training samples, yiAs model predicted value, tiIs an actual value, γα,λIs a normalization constant, λ>0 is the kernel parameter of the generalized Gaussian density function, α>0 is a shape parameter and the formula has a maximum value when the predicted value is equal to the actual value;
therefore, an updating formula of each weight parameter in the gated cyclic unit network can be obtained:
Figure FDA0003177550890000041
wherein x istFor the input at time t, W, U are the corresponding weights, zt,rtOutputs of the refresh gate and the reset gate, respectively, htFor the implicit state of the network at time t,
Figure FDA0003177550890000042
in a cellular state, ytFor prediction output at time t, J is cost loss of network, namely generalized maximum correlation entropy criterion, ttFor the desired output at time t, tanh (·), tanh' (·) is the hyperbolic tangent function and its derivative, C is a penalty factor, and
Figure FDA0003177550890000043
6. the method for predicting the short-term electricity sales amount based on the generalized cross-correlation entropy gated cyclic unit according to claim 5, wherein the specific process of the step 5 is as follows:
optimizing key parameters lambda and alpha of the generalized cross-correlation entropy gated cyclic unit model by a K-fold cross validation and grid optimization method
And (3) searching parameters by using grid optimization: giving possible value ranges of the parameters lambda and alpha according to the sample, then taking a plurality of values in the value range of each parameter, and combining every two values to form a parameter pair related to the parameters lambda and alpha, thereby selecting the parameter pair which enables the generalized maximum correlation entropy to be maximum; measuring the generalization ability of the parameters by using a K-fold cross validation method: dividing the existing data set into K subsets, making each subset respectively perform a test set, and making other subsets perform a training set to verify the generalization capability of the parameters for K times.
7. The method for predicting the short-term electricity sales amount based on the generalized cross-correlation entropy gated cyclic unit of claim 1, wherein: the step 6 is as follows:
predicting the electricity sales amount of the hour time scale by using a generalized cross-correlation entropy gating circulation unit prediction model to obtain a prediction result, and evaluating the prediction result by selecting an RMSE index:
Figure FDA0003177550890000051
CN202110840949.2A 2021-07-23 2021-07-23 Short-term electricity sales amount prediction method based on generalized cross-correlation entropy gating circulation unit Pending CN113627658A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110840949.2A CN113627658A (en) 2021-07-23 2021-07-23 Short-term electricity sales amount prediction method based on generalized cross-correlation entropy gating circulation unit

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110840949.2A CN113627658A (en) 2021-07-23 2021-07-23 Short-term electricity sales amount prediction method based on generalized cross-correlation entropy gating circulation unit

Publications (1)

Publication Number Publication Date
CN113627658A true CN113627658A (en) 2021-11-09

Family

ID=78380955

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110840949.2A Pending CN113627658A (en) 2021-07-23 2021-07-23 Short-term electricity sales amount prediction method based on generalized cross-correlation entropy gating circulation unit

Country Status (1)

Country Link
CN (1) CN113627658A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114511137A (en) * 2022-01-04 2022-05-17 西安理工大学 Short-term electric quantity prediction method based on cross-correlation entropy gating circulation unit
CN118226489A (en) * 2024-04-08 2024-06-21 北京航空航天大学 GNSS/INS integrated navigation method based on generalized maximum correlation entropy under GNSS short-time interruption

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111950697A (en) * 2020-07-01 2020-11-17 燕山大学 Cement product specific surface area prediction method based on gated cycle unit network
CN112348271A (en) * 2020-11-12 2021-02-09 华北电力大学 Short-term photovoltaic power prediction method based on VMD-IPSO-GRU
CN112435054A (en) * 2020-11-19 2021-03-02 西安理工大学 Nuclear extreme learning machine electricity sales amount prediction method based on generalized maximum correlation entropy criterion

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111950697A (en) * 2020-07-01 2020-11-17 燕山大学 Cement product specific surface area prediction method based on gated cycle unit network
CN112348271A (en) * 2020-11-12 2021-02-09 华北电力大学 Short-term photovoltaic power prediction method based on VMD-IPSO-GRU
CN112435054A (en) * 2020-11-19 2021-03-02 西安理工大学 Nuclear extreme learning machine electricity sales amount prediction method based on generalized maximum correlation entropy criterion

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114511137A (en) * 2022-01-04 2022-05-17 西安理工大学 Short-term electric quantity prediction method based on cross-correlation entropy gating circulation unit
CN118226489A (en) * 2024-04-08 2024-06-21 北京航空航天大学 GNSS/INS integrated navigation method based on generalized maximum correlation entropy under GNSS short-time interruption

Similar Documents

Publication Publication Date Title
CN110598929B (en) Wind power nonparametric probability interval ultrashort term prediction method
CN113627658A (en) Short-term electricity sales amount prediction method based on generalized cross-correlation entropy gating circulation unit
CN112435054B (en) Nuclear extreme learning machine electricity sales prediction method based on generalized maximum correlation entropy criterion
Dang et al. A novel multivariate grey model for forecasting periodic oscillation time series
CN113537469B (en) Urban water demand prediction method based on LSTM network and Attention mechanism
CN106446440A (en) Short-term photovoltaic generation power prediction method based on online sequential extreme learning machine
Sun et al. Enhancing financial risk management through lstm and extreme value theory: A high-frequency trading volume approach
Xiong et al. The housing demand analysis and prediction of the real estate based on the AWGM (1, N) model
Zhao et al. Point and interval forecasting for carbon trading price: a case of 8 carbon trading markets in China
CN107704944B (en) Construction method of stock market fluctuation interval prediction model based on information theory learning
CN117291069A (en) LSTM sewage water quality prediction method based on improved DE and attention mechanism
CN112232570A (en) Forward active total electric quantity prediction method and device and readable storage medium
Zang et al. The credit risk assessment of P2P lending based on BP neural network
CN112581311B (en) Method and system for predicting long-term output fluctuation characteristics of aggregated multiple wind power plants
CN115936791A (en) Short-term and medium-term price prediction method for electric power market
Liu et al. Forecasting China’s per capita living energy consumption by employing a novel DGM (1, 1, tα) model with fractional order accumulation
Ogihara et al. A study of risk reduction for daily peak load demand forecasting
Alamsyah et al. Comparison of Predictions Foreign Currency Exchange Rates (USD, JPY, EUR, GBP, CHF, CAD) to Rupiah (IDR) using Artificial Neural Network and Long Short-Term Memory
CN112529268A (en) Medium-short term load prediction method and device based on manifold learning
Wang et al. Modeling and forecasting of coal price based on influencing factors and time series
CN114511137A (en) Short-term electric quantity prediction method based on cross-correlation entropy gating circulation unit
Yan et al. Short-term Electricity Price Prediction Based on CEEMD-TCN-ATTENTION
Rasyad et al. Gum Rosin Price Forecasting Using A Hybrid ARIMA—LSTM Model
Zhang et al. Optimizing time-series prediction on China’s green trade economy
Zhou et al. Prediction of Fund Net Value Based on ARIMA-LSTM Hybrid Model

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