CN114243695A - Power load prediction method based on bidirectional long-time and short-time memory neural network - Google Patents
Power load prediction method based on bidirectional long-time and short-time memory neural network Download PDFInfo
- Publication number
- CN114243695A CN114243695A CN202111561942.3A CN202111561942A CN114243695A CN 114243695 A CN114243695 A CN 114243695A CN 202111561942 A CN202111561942 A CN 202111561942A CN 114243695 A CN114243695 A CN 114243695A
- Authority
- CN
- China
- Prior art keywords
- date
- neural network
- ridge regression
- input time
- memory neural
- 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.)
- Granted
Links
- 230000002457 bidirectional effect Effects 0.000 title claims abstract description 48
- 230000015654 memory Effects 0.000 title claims abstract description 46
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 35
- 238000000034 method Methods 0.000 title claims abstract description 33
- 238000013507 mapping Methods 0.000 claims abstract description 6
- 238000012549 training Methods 0.000 claims description 61
- 238000012360 testing method Methods 0.000 claims description 44
- 238000003062 neural network model Methods 0.000 claims description 15
- 230000001419 dependent effect Effects 0.000 claims description 11
- 238000011156 evaluation Methods 0.000 claims description 8
- 238000012216 screening Methods 0.000 abstract description 2
- 238000011161 development Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 230000007774 longterm Effects 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/003—Load forecast, e.g. methods or systems for forecasting future load demand
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
-
- 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
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Economics (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Power Engineering (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention discloses a power load prediction method based on a bidirectional long-short time memory neural network, which adopts a double-layer model architecture combining the bidirectional long-short time memory neural network and a ridge regression algorithm to predict power loads for p hours in the future. The invention provides a method for screening an optimal input time scale combination, which is used for establishing a two-way long-time and short-time memory neural network according to the optimal input time scale combination and solving the problem that the input time scale combination cannot be determined; the ridge regression model is used for establishing a mapping relation between a predicted value and an actual value of the day before the predicted date and predicting the predicted date, so that the prediction precision of the power load is greatly improved.
Description
Technical Field
The invention relates to the field of power load prediction, in particular to a power load prediction method based on a bidirectional long-time and short-time memory neural network.
Background
The power load prediction is one of the key technologies for stable operation of the power system. Because the electric energy cannot be stored in large quantities, the production, transportation, distribution and consumption of the electric energy are all completed within the same time, which requires higher accuracy of electric load prediction to ensure dynamic balance of electric energy supply and consumption in the electric power system. Inaccurate power load prediction will result in waste of electric energy and loss of power transmission lines, which affects the stable development of national economy. Due to the influence of various complex factors on the power system, the power load prediction has various randomness, volatility and uncertainty, and great challenges are brought to the power load prediction, so that the improvement of the accuracy of the power load prediction has important significance on the reasonable distribution and use of electric energy and the development of national economy.
With the development and progress of information technology, the field of power systems generates massive data, and the massive data is fully mined and utilized to provide greater data support for distribution and scheduling of electric energy. The common power load prediction methods at present mainly comprise artificial intelligence repetition and statistical methods and the like. Most of the existing methods need to cooperate with external characteristic data such as climate factors to predict, and most of the existing methods adopt a single model to predict, so that the power load prediction cannot obtain satisfactory accuracy.
Disclosure of Invention
The invention provides a power load prediction method based on a bidirectional long-short time memory neural network, which adopts a double-layer model architecture combining the bidirectional long-short time memory neural network and a ridge regression algorithm to predict power loads for p hours in the future.
According to the method, big data analysis is carried out on historical data of the power load, and a bidirectional long-time and short-time memory neural network is used as a base model, so that the bidirectional long-time and short-time memory neural network has strong processing capacity on a time sequence with large data quantity; since the power load data has a long-term correlation and a short-term correlation, a plurality of input time scales are set, and useful information features are extracted from different time scales before the forecast date.
In order to achieve the purpose, the invention adopts the following technical scheme:
a power load prediction method based on a bidirectional long-short-term memory neural network is characterized in that a double-layer model architecture combining the bidirectional long-short-term memory neural network and a ridge regression algorithm is adopted to predict power loads for p hours in the future, an optimal input time scale combination and a ridge regression coefficient alpha are screened out from input time scale combinations, the bidirectional long-short-term memory neural network is used as a prediction model, and the power loads for p hours in the future are predicted for a prediction date, and the method comprises the following steps:
step S1: firstly, determining the optimal input time scale combination and ridge regression coefficient alpha of a first layer of bidirectional long-short time memory neural network model by adopting a double-layer model architecture combining a bidirectional long-short time memory neural network and a ridge regression algorithm;
step S2: reconstructing historical data of the date D according to the optimal input time scale combination to obtain a first sample set, dividing the first sample set into a training set and a testing set, inputting the training set into a bidirectional long-time memory neural network model, and taking an obtained predicted value as a testing set of a second sample set; reconstructing historical data of the date D-1 according to the optimal input time scale combination to obtain a first sample set, dividing the first sample set into a training set and a testing set, inputting the training set into a bidirectional long-time memory neural network model to obtain a predicted value of the date D-1, and taking the predicted value and an actual value of the date D-1 as a training set of a second sample set;
step S3: training the ridge regression model according to the training set of the second sample set obtained in the step S2 to obtain the mapping relation between the predicted value corresponding to the date D-1 and the actual value, and inputting the second sample set test set obtained in the step S2 into the trained ridge regression for prediction to obtain the final predicted value of the date D;
the step S1 specifically includes the following steps:
step S1.1: firstly, using collected historical power load data as a sample data set, determining m different input time scalesDetermining the output step number as p, reconstructing the sample data set according to the output step number p and m different input time scales, and generating m different sample data sets; taking m different reconstruction sample data sets as a first sample set, dividing a training set and a testing set, and respectively training a bidirectional long-time memory neural network by using the m training sets to obtain m prediction models. Taking the date T as a prediction object, dividing a training set and a test set of a first sample set before the date T, respectively inputting the test sets corresponding to the date T into a prediction model, wherein m groups of different predicted values are obtained by the date T, and each group of predicted values represents information features extracted from corresponding input time scales; fully combining m groups of different predicted values to obtain
Combining the seeds and using the combinations as a test set of a second sample set;
step S1.2: dividing a first sample set before the date T-1 into a training set and a test set, and respectively inputting the test set corresponding to the date T-1 into the m prediction models obtained in the step S1.1 to obtain m different prediction values of the date T-1;
step S1.3: knowing the actual value of the date T-1, taking the actual value of the date T-1 as a dependent variable, and fully combining the m different predicted values of the date T-1 obtained in step S1.2 to obtain
The independent variables of each group, namely the combination types of different input time scales, correspond to the same dependent variable after combination, each independent variable and dependent variable of each group are used as a training set of a second sample set to train ridge regression, the test set of the second sample set obtained in the step S1.1 is input into the trained ridge regression model, and the output value is used as a final predicted value corresponding to the date T;
step S1.4: carrying out accuracy evaluation on the final predicted value and the actual value of the date T obtained in the step S1.3, and carrying out fitting degree R2And the root mean square error RMSE is used as an evaluation index, and the optimal ridge regression coefficient alpha and the fitting degree R are determined by a grid search method2Determining the ridge regression coefficient alpha corresponding to the highest time as a final ridge regression coefficient; the input time scale combination corresponding to the independent variable combination with the highest fitting degree R2 is used as the optimal input time scale combination, and the optimal input time scale combination and the optimal ridge regression coefficient alpha are applied to the prediction model, so that the optimal input time scale combination and the ridge regression coefficient alpha of the first-layer bidirectional long-time and short-time memory neural network model are determined;
the invention has the advantages and beneficial effects that:
compared with the prior art, the invention discloses and provides a power load prediction method based on a bidirectional long-time and short-time memory neural network, and the method has the advantages that: aiming at the characteristics of long-term correlation and short-term correlation of power load data, a double-layer model architecture combining a bidirectional long-term and short-term memory neural network and a ridge regression algorithm is adopted, and a method for screening an optimal input time scale combination is provided. A bidirectional long-time and short-time memory neural network is established according to the optimal input time scale combination, so that the problem that the input time scale combination cannot be determined is solved; the ridge regression model is used for establishing a mapping relation between a predicted value and an actual value of the day before the predicted date and predicting the predicted date, so that the prediction precision of the power load is greatly improved.
Drawings
Fig. 1 is a flowchart of a power load prediction method based on a bidirectional long-and-short-term memory neural network disclosed by the invention.
FIG. 2 is a diagram of a two-layer model architecture combining a two-way long-short term memory neural network and a ridge regression algorithm.
Detailed Description
In order to more clearly illustrate the object and technical solution of the present invention, the present invention will be further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. 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 invention provides a power load prediction method based on a bidirectional long-short time memory neural network, which adopts a double-layer model architecture combining the bidirectional long-short time memory neural network and a ridge regression algorithm to predict power load for p hours in the future, firstly, selects an optimal input time scale combination and a ridge regression coefficient alpha from different input time scale combinations, then, takes the bidirectional long-short time memory neural network as a prediction model to predict the power load for p hours in the future for a prediction date, and comprises the following steps:
step S1: firstly, determining the optimal input time scale combination and ridge regression coefficient alpha of a first layer of bidirectional long-short time memory neural network model by adopting a double-layer model architecture combining a bidirectional long-short time memory neural network and a ridge regression algorithm;
step S2: reconstructing historical data of a date D to be predicted according to the optimal input time scale combination to obtain a first sample set, dividing the first sample set into a training set and a testing set, inputting the training set into a bidirectional long-time memory neural network model, and taking the obtained predicted value as a testing set of a second sample set; reconstructing historical data of the date D-1 according to the optimal input time scale combination to obtain a first sample set, dividing the first sample set into a training set and a testing set, inputting the training set into a bidirectional neural network model to obtain a predicted value of the date D-1, and taking the predicted value and an actual value of the date D-1 as a training set of a second sample set;
step S3: training the ridge regression model according to the training set of the second sample set obtained in the step S2 to obtain the mapping relation between the predicted value corresponding to the date D-1 and the actual value, and inputting the second sample set test set obtained in the step S2 into the trained ridge regression for prediction to obtain the final predicted value of the date D;
the step S1 specifically includes the following steps:
step S1.1: firstly, taking collected historical power load data as a sample data set, determining m different input time scales, determining an output step number as p, reconstructing the sample data set according to the output step number p and the m different input time scales, and generating m different sample data sets; will be provided withAnd taking m different reconstruction sample data sets as a first sample set, dividing a training set and a testing set, and respectively training a bidirectional long-time memory neural network by using the m training sets to obtain m prediction models. Taking the date T as a prediction object, dividing a training set and a test set of a first sample set before the date T, respectively inputting the test sets corresponding to the date T into a prediction model, wherein m groups of different predicted values are obtained by the date T, and each group of predicted values represents information features extracted from corresponding input time scales; fully combining m groups of different predicted values to obtain
Combining the seeds and using the combinations as a test set of a second sample set;
step S1.2: dividing a first sample set before the date T-1 into a training set and a test set, and respectively inputting the test set corresponding to the date T-1 into the m prediction models obtained in the step S1.1 to obtain m different prediction values of the date T-1;
step S1.3: knowing the actual value of the date T-1, taking the actual value of the date T-1 as a dependent variable, and fully combining the m different predicted values of the date T-1 obtained in step S1.2 to obtain
The independent variables of each group, namely the combination types of different input time scales, correspond to the same dependent variable after combination, each independent variable and dependent variable of each group are used as a training set of a second sample set to train ridge regression, the test set of the second sample set obtained in the step S1.1 is input into the trained ridge regression model, and the output value is used as a final predicted value corresponding to the date T;
step S1.4: and (4) carrying out accuracy evaluation on the final predicted value and the actual value of the date T obtained in the step (S1.3), and determining an optimal ridge regression coefficient alpha and a fitting degree R through a grid search method by taking the fitting degree R2 and the root mean square error RMSE as evaluation indexes2Determining the ridge regression coefficient alpha corresponding to the highest time as a final ridge regression coefficient; the combination of arguments corresponding to the highest fitness R2 will be the best input timescale combination,applying the input time scale combination and the optimal ridge regression coefficient alpha in the prediction model, thereby determining the optimal input time scale combination and the ridge regression coefficient alpha of the first-layer bidirectional long-time memory neural network model;
in the embodiment, a certain power load historical data is used as a sample data, the data acquisition frequency of the sample is small, the time span is from 1/2007 to 10/2010 and 31/2010, the power load prediction is performed by using the bidirectional long-short term memory neural network, and the specific implementation steps are as follows:
step S1: firstly, determining the optimal input time scale combination and ridge regression coefficient alpha of a first layer of bidirectional long-short time memory neural network model by adopting a double-layer model architecture combining a bidirectional long-short time memory neural network and a ridge regression algorithm;
step S1.1: firstly, determining 7 different input time scales, setting the lengths of time sliding windows to be 168, 144, 120, 96, 72, 48 and 24 hours respectively, determining the output step number to be 24 hours, and generating 7 different reconstruction sample data sets in total; taking 7 different reconstruction sample data sets as a first sample set and dividing a training set and a test set; taking 12-month-31-year 2009 as a prediction object, respectively taking 168, 144, 120, 96, 72, 48 and 24 hours before the date as a test set of a first sample set, and taking the rest reconstruction data as a training set of the first sample set; inputting a training set corresponding to a prediction date into a bidirectional long-time memory neural network for training to obtain 7 different prediction models in total, inputting a test set into the trained prediction models, and obtaining 7 groups of different prediction values in total by predicting the date, wherein each group of prediction values represents information characteristics extracted from a corresponding input time scale; fully combining 7 groups of different predicted values to obtain
127 in total;
step S1.1 obtains 7 different sets of predicted values, each set of predicted values representing 24 hours of power load data on a predicted day, respectively, as:
Y168＝[[Y1681，Y1682，Y168_3，...，Y168_24]a predicted value representing a predicted day when the input time scale is 168 hours;
Y144＝[Y144_1，Y144_2，Y144_3，...，Y144_24]a predicted value representing a predicted day when the input time scale is 144 hours;
Y120＝[Y120_1，Y120_2，Y120_3，...，Y120_24]a predicted value representing a predicted day when the input time scale is 120 hours;
Y96＝[Y96_1，Y96_2，Y96_3，...，Y96_24]a predicted value representing a predicted day when the input time scale is 96 hours;
Y72＝[Y72_1，Y72_2，Y72_3，...，Y72_24]a predicted value representing a predicted day when the input time scale is 72 hours;
Y48＝[Y48_1，Y48_2，Y48_3，...，Y48_24]a predicted value representing a predicted day when the input time scale is 48 hours;
Y24＝[Y24_1，Y24_2，Y24_3，...，Y24_24]a predicted value representing a predicted day when the input time scale is 24 hours;
the 7 groups of different predicted values are fully combined to represent that the information features extracted by different input time scales are combined to find the optimal input time scale combination, the optimal combination represents that the most effective information is contained when the power load of 24 hours in the future is predicted, the accuracy of predicting the power load of 24 hours in the future is favorably improved, and the combination type is as follows:
X1＝[Y168，Y144，Y120，Y96，Y72，Y48，Y24]the expression is that each group of predicted values in 7 groups of predicted values are independently used as a group of independent variables, namely, information features extracted by different input time scales are independently used as a group of independent variables to obtain 7 groups of independent variables and independent variablesQuantities are labeled in turn; x1_1，X1_2，X1_3，X1_4，X1_5，X1_6，X1_7；
X2＝[[Y168，Y144]，[Y168，Y120]，[Y168，Y96]，[Y168，Y72]，[Y168，Y48]，[Y168，Y24]，[Y144，Y120]，[Y144，Y96]，[Y144，Y72]，[Y144，Y48]，[Y144，Y24]，[Y120，Y96]，[Y120，Y72]，[Y120，Y48]，[Y120，Y24]，[Y96，Y72]，[Y96，Y48]，[Y96，Y24]，[Y72，Y48]，[Y72，Y24][Y48，Y24]]The method comprises the steps of combining every two groups of predicted values in 7 groups of predicted values to serve as a group of independent variables, namely combining information characteristics extracted from every two different input time scales to serve as the independent variables to obtain 21 groups of independent variables, wherein the independent variables are sequentially marked as X2_1，X2_2，X2_3，X2_4，X2_5，X2_6，X2_7，X2_8，X2_9，X2_10，X2_11，X2_12，X2_13，X2_14，X2_15，X2_16，X2_17，X2_18，X2_19，X2_20，X2_21；
X3＝[[Y168，Y144，Y120]，[Y168，Y144，Y96]，[Y168，Y144，Y72]，…，[Y72，Y48，Y24]]The expression combines each three groups of predicted values in 7 groups of predicted values to be used as a group of independent variables, namely, combines information characteristics extracted from each three different input time scales to be used as independent variables to obtain 35 sets of independent variables, the independent variables being successively marked X3_1，X3_2，X3_3，...，X3_35；
X4＝[[Y168，Y144，Y120，Y96]，[Y168，Y144，Y120，Y72]，[Y168，Y144，Y120，Y48]，…，[Y96，Y72，Y48，Y24]]The method comprises the steps of combining every four groups of predicted values in 7 groups of predicted values to be used as a group of independent variables, namely combining information characteristics extracted from every four different input time scales to be used as the independent variables to obtain 35 groups of independent variables, wherein the independent variables are sequentially marked as X4_1，X4_2，X4_3，...，X4_35；
X5＝[[Y168，Y144，Y120，Y96，Y72]，[Y168，Y144，Y120，Y96，Y48]，[Y168，Y144，Y120，Y96，Y24]，…，[Y120，Y96，Y72，Y48，Y24]]The method comprises the steps of combining every five predicted values in 7 groups of predicted values to serve as a group of independent variables, namely combining information characteristics extracted from every five different input time scales to serve as the independent variables to obtain 21 groups of independent variables, wherein the independent variables are sequentially marked as X5_1，X5_2，X5_3，...，X5_21；
X6＝[[Y168，Y144，Y120，Y96，Y72，Y48]，[Y168，Y144，Y120，Y96，Y72，Y24]，[Y168，Y144，Y120，Y96，Y48，Y24]，[Y168，Y144，Y120，Y72，Y48，Y24]，[Y168，Y144，Y96，Y72，Y48，Y24]，[Y168，Y120，Y96，Y72，Y48，Y24]，[Y144，Y120，Y96，Y72，Y48，Y24]]The method comprises the steps of combining every six groups of predicted values in 7 groups of predicted values to serve as a group of independent variables, namely combining information features extracted from every six different input time scales to serve as the independent variables to obtain 7 groups of independent variables, wherein the independent variables are sequentially marked as X6_1，X6_2，X6_3，X6_4，X6_5，X6_6，X6_7；
X7＝[[Y168，Y144，Y120，Y96，Y72，Y48，Y24]The method comprises the steps of combining 7 groups of predicted values to serve as a group of independent variables, namely combining information characteristics extracted by 7 different input time scales to serve as the independent variables to obtain 1 group of independent variables, wherein the independent variables are marked as X7_1；
All arguments are put into the set of arguments and are expressed as:
taking the obtained self-variable set X as a second sample test set of the prediction date of 2009, 12 and 31;
dividing a first sample set before the forecast date, namely, 2009, 12, month and 30 days into a training set and a test set, and respectively inputting the test sets corresponding to the dates into [0021 ]]7 groups of different predicted values of 12 months and 30 days in 2009 are obtained by the obtained 7 prediction models; the 7 groups obtained on the day were differentThe predicted values are fully combined according to the method to obtain
127 combinations in total, wherein each combination is used as a group of independent variables, and all the obtained independent variables are put into an independent variable set; knowing the actual value of 30 days 12 months in 2009, and using the actual value of the day as a dependent variable;
127 independent variables in the independent variable set of 12-month-30-year 2009 are used, and each independent variable and dependent variable is used as a training set of a second sample set, so that 127 different training sets are used;
inputting each group of training sets into a ridge regression model for training in a cyclic traversal mode, and inputting a second sample test set of 12 months and 31 days in 2009 obtained by [0040] into the trained ridge regression model to obtain a final predicted value; taking the traversal of the training set as an inner loop, simultaneously taking the traversal of the ridge regression coefficient alpha as an outer loop, and determining the optimal ridge regression coefficient alpha by a grid search method; the range of the ridge regression coefficient alpha is 0 to 20, and the step length is 0.1; searching a training set and a ridge regression coefficient alpha corresponding to the optimal predicted value by constructing a double-layer loop;
carrying out accuracy evaluation on the final predicted value and the actual value obtained in 12, 31 and 12 months in 2009, wherein the process is carried out in an inner loop; degree of fitting R2And the root mean square error RMSE as an evaluation index; degree of fitting R2Determining the ridge regression coefficient alpha corresponding to the highest time as a final ridge regression coefficient; taking the training set corresponding to the highest fitting degree R2 as an optimal training set, and taking the input time scale combination corresponding to the independent variable in the optimal training set as an optimal input time scale combination;
the input time scale combination corresponding to the independent variable in the optimal training set obtained in the step is an hour combination of [144, 96 and 48], the optimal ridge regression coefficient alpha is 3.2, and the optimal input time scale combination and the optimal ridge regression coefficient alpha are used as parameters of a subsequent prediction model;
step S2: reconstructing historical data of a date to be predicted according to the optimal input time scale combination to obtain a first sample set, dividing the first sample set into a training set and a testing set, inputting the training set into a bidirectional neural network model, and taking an obtained predicted value as a testing set of a second sample set; reconstructing historical data of the day before the date to be predicted according to the optimal input time scale combination to obtain a first sample set, dividing the first sample set into a training set and a testing set, inputting the training set into a bidirectional long-time memory neural network model to obtain a predicted value of the date D-1, and taking the predicted value and an actual value of the day before the date to be predicted as a training set of a second sample set;
step S3: training the ridge regression model according to the training set of the second sample set obtained in the step S2 to obtain a mapping relation between a predicted value corresponding to the date to be predicted and an actual value, inputting the second sample set test set obtained in the step S2 into the trained ridge regression to predict to obtain a final predicted value of the date to be predicted, wherein the prediction accuracy is remarkably improved compared with a single input time scale;
the above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art should be considered to be within the technical scope of the present invention, and the technical solutions and the inventive concepts thereof according to the present invention are equivalent to or changed within the technical scope of the present invention.
Claims (3)
1. A power load prediction method based on a bidirectional long-short-term memory neural network is characterized in that a double-layer model architecture combining the bidirectional long-short-term memory neural network and a ridge regression algorithm is adopted to predict power loads for p hours in the future, firstly, an optimal input time scale combination and a ridge regression coefficient alpha are screened out from different input time scale combinations, then, the bidirectional long-short-term memory neural network is used as a prediction model, and the power loads for p hours in the future are predicted for a prediction date.
2. The method for predicting the power load based on the bidirectional long-and-short-term memory neural network as claimed in claim 1, wherein the method comprises the following steps:
step S1: firstly, determining the optimal input time scale combination and ridge regression coefficient alpha of a first layer of bidirectional long-short time memory neural network model by adopting a double-layer model architecture combining a bidirectional long-short time memory neural network and a ridge regression algorithm;
step S2: reconstructing historical data of a date D to be predicted according to the optimal input time scale combination to obtain a first sample set, dividing the first sample set into a training set and a testing set, inputting the training set into a bidirectional long-time memory neural network model, and combining the obtained predicted values to serve as a testing set of a second sample set; reconstructing historical data of the date D-1 according to the optimal input time scale combination to obtain a first sample set, dividing the first sample set into a training set and a testing set, inputting the training set into a bidirectional neural network model to obtain a predicted value of the date D-1, and taking the predicted value and an actual value of the date D-1 as a training set of a second sample set;
step S3: training the ridge regression model according to the training set of the second sample set obtained in the step S2 to obtain the mapping relation between the predicted value corresponding to the date D-1 and the actual value, and inputting the second sample set test set obtained in the step S2 into the trained ridge regression for prediction to obtain the final predicted value of the date D;
3. the method for predicting the power load based on the bidirectional long-and-short term memory neural network as claimed in claim 2, wherein the specific method of step S1 is as follows:
step S1.1: firstly, taking collected historical power load data as a sample data set, determining m different input time scales, determining an output step number as p, reconstructing the sample data set according to the output step number p and the m different input time scales, and generating m different sample data sets; taking m different reconstruction sample data sets as a first sample set, dividing a training set and a testing set, and respectively training a bidirectional long-time memory neural network by using the m training sets to obtain m prediction models. Taking the date T as a prediction object, dividing a training set and a test set of a first sample set before the date T, and dividing the test set corresponding to the date TRespectively inputting the prediction models, and obtaining m groups of different predicted values in the date T, wherein each group of predicted values represents information characteristics extracted from the corresponding input time scale; fully combining m groups of different predicted values to obtain
Combining the seeds and using the combinations as a test set of a second sample set;
step S1.2: dividing a first sample set before the date T-1 into a training set and a test set, and respectively inputting the test set corresponding to the date T-1 into the m prediction models obtained in the step S1.1 to obtain m different prediction values of the date T-1;
step S1.3: knowing the actual value of the date T-1, taking the actual value of the date T-1 as a dependent variable, and fully combining the m different predicted values of the date T-1 obtained in step S1.2 to obtain
The independent variables of each group, namely the combination types of different input time scales, correspond to the same dependent variable after combination, each independent variable and dependent variable of each group are used as a training set of a second sample set to train ridge regression, the test set of the second sample set obtained in the step S1.1 is input into the trained ridge regression model, and the output value is used as a final predicted value corresponding to the date T;
step S1.4: carrying out accuracy evaluation on the final predicted value and the actual value of the date T obtained in the step S1.3, and carrying out fitting degree R2And the root mean square error RMSE is used as an evaluation index, and the optimal ridge regression coefficient alpha and the fitting degree R are determined by a grid search method2Determining the ridge regression coefficient alpha corresponding to the highest time as a final ridge regression coefficient; the independent variable combination corresponding to the highest fitting degree R2 is used as the optimal input time scale combination, and the optimal input time scale combination and the optimal ridge regression coefficient alpha are applied to the prediction model, so that the optimal input time scale combination and the ridge regression coefficient alpha of the first-layer bidirectional long-short time memory neural network model are determined.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111561942.3A CN114243695B (en) | 2021-12-17 | 2021-12-17 | Power load prediction method based on bidirectional long-short-term memory neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111561942.3A CN114243695B (en) | 2021-12-17 | 2021-12-17 | Power load prediction method based on bidirectional long-short-term memory neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114243695A true CN114243695A (en) | 2022-03-25 |
| CN114243695B CN114243695B (en) | 2023-09-15 |
Family
ID=80759116
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111561942.3A Active CN114243695B (en) | 2021-12-17 | 2021-12-17 | Power load prediction method based on bidirectional long-short-term memory neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114243695B (en) |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070185823A1 (en) * | 2005-07-28 | 2007-08-09 | Dingguo Chen | Load prediction based on-line and off-line training of neural networks |
| CN106936127A (en) * | 2017-02-28 | 2017-07-07 | 国家电网公司 | A kind of line load regression analysis and Forecasting Methodology and system |
| CN108876034A (en) * | 2018-06-13 | 2018-11-23 | 重庆邮电大学 | A kind of improved Lasso+RBF neural network ensemble prediction model |
| CN109389238A (en) * | 2017-08-14 | 2019-02-26 | 中国电力科学研究院 | A kind of short-term load forecasting method and device based on ridge regression |
| CN110309979A (en) * | 2019-07-09 | 2019-10-08 | 南方电网科学研究院有限责任公司 | Power load prediction method, device and equipment based on echo state network |
| US20200302524A1 (en) * | 2019-03-18 | 2020-09-24 | Zestfinance, Inc. | Systems and methods for model fairness |
| CN111932024A (en) * | 2020-08-21 | 2020-11-13 | 润联软件系统(深圳)有限公司 | Energy load prediction method and device, computer equipment and storage medium |
| CN112001740A (en) * | 2020-06-19 | 2020-11-27 | 南京理工大学 | Combined prediction method based on adaptive neural network |
| CN112116144A (en) * | 2020-09-15 | 2020-12-22 | 山东科技大学 | Regional power distribution network short-term load prediction method |
| WO2021105246A1 (en) * | 2019-11-26 | 2021-06-03 | Basf Se | Forecasting industrial aging processes with machine learning methods |
-
2021
- 2021-12-17 CN CN202111561942.3A patent/CN114243695B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070185823A1 (en) * | 2005-07-28 | 2007-08-09 | Dingguo Chen | Load prediction based on-line and off-line training of neural networks |
| CN106936127A (en) * | 2017-02-28 | 2017-07-07 | 国家电网公司 | A kind of line load regression analysis and Forecasting Methodology and system |
| CN109389238A (en) * | 2017-08-14 | 2019-02-26 | 中国电力科学研究院 | A kind of short-term load forecasting method and device based on ridge regression |
| CN108876034A (en) * | 2018-06-13 | 2018-11-23 | 重庆邮电大学 | A kind of improved Lasso+RBF neural network ensemble prediction model |
| US20200302524A1 (en) * | 2019-03-18 | 2020-09-24 | Zestfinance, Inc. | Systems and methods for model fairness |
| CN110309979A (en) * | 2019-07-09 | 2019-10-08 | 南方电网科学研究院有限责任公司 | Power load prediction method, device and equipment based on echo state network |
| WO2021105246A1 (en) * | 2019-11-26 | 2021-06-03 | Basf Se | Forecasting industrial aging processes with machine learning methods |
| CN112001740A (en) * | 2020-06-19 | 2020-11-27 | 南京理工大学 | Combined prediction method based on adaptive neural network |
| CN111932024A (en) * | 2020-08-21 | 2020-11-13 | 润联软件系统(深圳)有限公司 | Energy load prediction method and device, computer equipment and storage medium |
| CN112116144A (en) * | 2020-09-15 | 2020-12-22 | 山东科技大学 | Regional power distribution network short-term load prediction method |
Non-Patent Citations (4)
| Title |
|---|
| DOHYUN KIM 等: ""Multiscale LSTM-Based Deep Learning for Very-Short-Term Photovoltaic Power Generation Forecasting in Smart City Energy Management"", vol. 15, no. 1, pages 346 - 354, XP011841727, DOI: 10.1109/JSYST.2020.3007184 * |
| 徐晴 等: ""基于机器学习的短期电力负荷预测方法研究"", vol. 56, no. 23, pages 70 - 75 * |
| 蒋惠凤;何有世;张兵;: "基于回归神经网络的长期电力负荷组合预测模型研究", no. 24, pages 62 - 64 * |
| 蓝新斌;李富盛;余涛;: "基于改进长短时记忆网络的短期负荷预测方法", no. 11, pages 10 - 13 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114243695B (en) | 2023-09-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Hong et al. | Energy forecasting: A review and outlook | |
| CN113962364B (en) | Multi-factor power load prediction method based on deep learning | |
| CN110110912B (en) | Photovoltaic power multi-model interval prediction method | |
| CN111178616B (en) | Wind speed prediction method based on negative correlation learning and regularization extreme learning machine integration | |
| CN110443417A (en) | Multi-model integrated load prediction method based on wavelet transformation | |
| CN111680841B (en) | Short-term load prediction method, system and terminal equipment based on principal component analysis | |
| Rizwan et al. | Artificial intelligence based approach for short term load forecasting for selected feeders at madina saudi arabia | |
| CN113591368A (en) | Comprehensive energy system multi-energy load prediction method and system | |
| CN116258269A (en) | Ultra-short-term load dynamic prediction method based on load characteristic decomposition | |
| CN114862032B (en) | XGBoost-LSTM-based power grid load prediction method and device | |
| CN115860797A (en) | Electric quantity demand prediction method suitable for new electricity price reform situation | |
| CN116205359A (en) | Method, device, terminal and storage medium for predicting daily power load | |
| He et al. | Forecasting power grid states for regional energy markets with deep neural networks | |
| CN116611702A (en) | Integrated learning photovoltaic power generation prediction method for building integrated energy management | |
| CN115965150A (en) | Power load prediction method based on weight distribution | |
| Tian et al. | A new information priority accumulated grey model with hyperbolic sinusoidal term and its application | |
| Qin et al. | Forecasting of China consumer price index based on EEMD and SVR method | |
| Zou et al. | Predicting the electric power consumption of office buildings based on dynamic and static hybrid data analysis | |
| CN114243695B (en) | Power load prediction method based on bidirectional long-short-term memory neural network | |
| Chen et al. | Short-term photovoltaic power prediction based on LGBM-XGBoost | |
| Cheng et al. | Temperature & humidity-weighted-modified GRNN based prediction model of photovoltaic power generation | |
| CN114330818A (en) | Dynamic water demand prediction method based on main driving factor screening and deep learning | |
| Balraj et al. | A DNN based LSTM model for predicting future energy consumption | |
| CN117526316B (en) | Load prediction method based on GCN-CBAM-BiGRU combined model | |
| Mbasso et al. | Demand-supply forecasting based on deep learning for electricity balance in Cameroon |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |