CN111985719B - Power load prediction method based on improved long-term and short-term memory network - Google Patents
Power load prediction method based on improved long-term and short-term memory network Download PDFInfo
- Publication number
- CN111985719B CN111985719B CN202010878240.7A CN202010878240A CN111985719B CN 111985719 B CN111985719 B CN 111985719B CN 202010878240 A CN202010878240 A CN 202010878240A CN 111985719 B CN111985719 B CN 111985719B
- Authority
- CN
- China
- Prior art keywords
- load
- power load
- feature
- historical
- adopting
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 49
- 230000007787 long-term memory Effects 0.000 title claims description 6
- 230000006403 short-term memory Effects 0.000 title claims description 5
- 230000015654 memory Effects 0.000 claims abstract description 11
- 238000012216 screening Methods 0.000 claims abstract description 11
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 4
- 238000012549 training Methods 0.000 claims description 22
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 11
- 238000000605 extraction Methods 0.000 claims description 9
- 230000006870 function Effects 0.000 claims description 9
- 238000012360 testing method Methods 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 3
- 238000012937 correction Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 230000000644 propagated effect Effects 0.000 claims description 3
- 239000013598 vector Substances 0.000 claims description 3
- 238000010187 selection method Methods 0.000 claims description 2
- 238000013277 forecasting method Methods 0.000 abstract 1
- 230000000694 effects Effects 0.000 description 6
- 238000004458 analytical method Methods 0.000 description 5
- 230000003044 adaptive effect Effects 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 description 1
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
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
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
-
- 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
-
- 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)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- General Business, Economics & Management (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Marketing (AREA)
- Tourism & Hospitality (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Operations Research (AREA)
- Power Engineering (AREA)
- Primary Health Care (AREA)
- Water Supply & Treatment (AREA)
- Public Health (AREA)
- Quality & Reliability (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (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 an improved long-short-term memory network, which adopts the characteristic of primarily screening historical load by using a maximum information coefficient, adopts a maximum correlation minimum redundancy algorithm to further brush and select the historical load in combination with the influence caused by a load correlation factor, takes the screened characteristic set and the characteristic set as the input of a model, adopts the improved long-short-term memory network to conduct power load prediction, and verifies the obtained prediction result and the actual power grid load to prove the practicability of the model. The forecasting method (H-ILSTM) accurately considers the power load and the related factors affecting the power load, effectively improves the accuracy of power load forecasting, and improves the safety and the economical efficiency of power grid operation to a certain extent.
Description
Technical Field
The invention relates to the technical field of power load prediction, in particular to a power load prediction method based on an improved long-term and short-term memory network.
Background
The electric load prediction plays a key role in the operation of the power grid, and the safety and the economy of the operation of the power grid can be greatly improved by obtaining accurate short-term load. In addition, the method has important significance for the optimal combination, economic dispatch, optimal power flow and electric power market transaction of the unit. The higher the accuracy of the power load prediction, the better the utilization of the power plant and the better the effect of the economic dispatch will be. However, the electrical load is sensitive to external factors such as climate change, date type and social activity, etc. These uncertainties increase the randomness of the load sequence. Therefore, how to identify the strong relevant factors of the extracted load from a plurality of influencing factors and realize accurate prediction of the short-term load is a problem to be solved at present.
In the deep learning method, a long-short-term memory network (LSTM) consists of a cyclic neural network of complex units, and has good capability of representing sequence information. It achieves very good results in solving the problem of time series prediction. However, LSTM inputs are typically defined artificially, with poor physical meaning and interpretation for the model. The relationship among the related factors influencing the power load by adopting the characteristic engineering analysis has important significance for improving the prediction precision of the power load, and the physical significance and the interpretation of the model are enhanced. Therefore, how to accurately analyze the relationship between the power load and the related influencing factors and incorporate the relationship into the prediction model to improve the accuracy of the power load prediction is a theoretical and practical engineering problem that needs to be solved.
Feature engineering is the selection of a representative subset of features in a feature set. These features are highly correlated with output variables and are the most common method of extracting valid features. Common feature selection methods include Autocorrelation (AC), mutual Information (MI), reliefF (RF), correlation-based feature selection (CFS), and the like.
In the present invention, the maximum information coefficient is used to preliminarily screen the characteristic of the historical power load. And the maximum correlation minimum redundancy method is combined with a prediction model to carry out secondary screening on the characteristics.
In summary, the problems of the prior art are:
(1) There are many factors that affect the electrical load, and these cannot be reasonably considered in the load prediction problem.
(2) The existing characteristic selection related to the power load prediction is mostly to singly consider the influence of a certain characteristic on the power load of the power load, and not consider the effect of the combination of the characteristic and the characteristic on the power load.
(3) Because of more factors in the input of the prediction model, the traditional machine learning method cannot predict the power load to achieve higher precision. The traditional LSTM can achieve better prediction effect, but the structure is complicated, the training time is long, and the practical application is inconvenient.
The difficulty of solving the technical problems is as follows: in order to take into account the influence of the correlation factors on the power load, it is one of the technical difficulties how to rationally handle these continuous and discrete features. The influence between factors is complicated, a single factor can not greatly improve the prediction, but the prediction precision can be greatly improved after the factors are combined, and how to screen out the combined factors and input the combined factors into a model for training is a second difficulty of the technology; furthermore, how to change the structure of LSTM so that the prediction accuracy can be improved and the required training time can be reduced in solving the load prediction problem is also a difficulty of the technology.
After solving the technical problems, the significance brought is as follows: in order to consider the factor-factor interaction, the invention provides a two-stage hybrid feature extraction method and a novel network (ILSTM) applied to power load prediction. The network is structurally different from the traditional LSTM, and the correct derivation of the forward and backward propagation formulas of the network is a difficulty in implementing the network.
The existing power load prediction by adopting the deep learning method is mostly to predict the power load of the next period. The method can accurately predict the power load of a plurality of future time periods and has practical application significance. Thus being very beneficial to the popularization of H-ILSTM.
Disclosure of Invention
The invention aims to solve the problems and provide a power load prediction method based on an improved long-short-term memory network, which can accurately analyze the influence among related features, screen out an optimal feature set, and predict by adopting a novel network so as to obtain a high-precision power load prediction result.
The invention realizes the above purpose through the following technical scheme:
the invention comprises the following steps:
(1) Collecting power load and related factor data thereof, and primarily screening future influence of the power load according to historical power load;
the historical power load sequences are collected, the relation between the load of 168 time periods (one time period is 1 hour) in seven days of the history and the predicted load required by the day is analyzed by adopting the Maximum Information Coefficient (MIC), the calculated historical load with the MIC larger than 0.6 is formed into a pre-selected set, and the rest load is formed into a candidate set.
(2) Performing preliminary treatment on the power load related factors;
factors affecting the power load correlation are collected, including continuous features and discrete features, which require some form of processing to be applied by the predictive model. The invention adopts the LabelEncoder coding mode to code the date and time types. Wherein the discrete features include a date category and a time category. The date categories are classified into workdays, rest days and legal holidays, and since the Chinese legal holidays have seven days at most, the legal holidays can be classified into 7 categories. In the time category, three periods are divided into one domain according to the power load characteristic.
(3) Performing secondary screening on the historical power load by combining the power load related factors and the prediction model;
the relationship between the load in the candidate subset and the load currently being predicted is analyzed. And analyzing the characteristics of the candidate subset by adopting a maximum correlation minimum redundancy method. The formula is as follows:
wherein S is m-1 Representing a preselect set, X-S m-1 Represents candidate set, c is category variable, m is the number of features, x j Is the j-th feature. I (x) j C) represents the number of mutual information between the j-th feature and the class variable.
The method finds features in the remaining feature space that maximize the value of the above formula based on the selected features. And (3) placing the features with the maximum calculated coefficient lambda into a preselected feature subset, combining other related factors influencing the power load, normalizing all the features, and then placing the features into a prediction model for prediction. Using the root mean square error RMSE as a discrimination coefficient, if the RMSE is reduced, the above process continues until the RMSE becomes high relative to before the feature is added.
(4) The screened final feature set is put into an improved long and short memory network ILSTM to be predicted, and firstly parameters of the NLSTM are set, including the number of nodes of an input layer, the number of nodes of a hidden layer, the number of nodes of an output layer, the learning efficiency, the batch size and the number of training rounds;
(5) And (3) training each weight parameter of the ILSTM on a training set by adopting an adaptive moment estimation (Adam) optimizer and combining a mini-batch mechanism. The method comprises the steps of carrying out a first treatment on the surface of the
(6) And inputting the test set into the trained ILSTM to predict, and obtaining a power load prediction result. Further, in the step (5), the step and the calculation formula of the forward propagation of the information of the improved long short memory network (ILSTM) in the t-th period are as follows:
u t ＝σ(net ut ) (3)
o t ＝σ(net ot ) (8)
h t ＝o t *tanh(c t ) (9)
z t ＝W y ·h t +b y (10)
y t ＝σ(zt) (11)
in the above formula, net ut ,net ot And z t Is the state of the current phase of the t period; w (W) u ,/>And W is o Respectively their weight matrices; b u ,/>And b o Representing their respective bias vectors; u (u) t ,/>C t ,o t ,h t And y t Updating the gate, information state, cell state, output gate, hidden layer output and predicted value of time period t respectively; tanh and sigma are tanh and sigmoid activation functions, respectively; symbol and represent matrix multiplication and multiplication between matrix elements, respectively.
Further, in the step (5), the step and the calculation formula of the information back propagation in the t-th period are as follows:
a. defining the most common square error function as the target to be optimized
b. Calculating errors of output layer
c. Calculating errors of hidden layers
W＝W-η·δW (28)
d. Adam optimization algorithm is adopted, and [ delta ] w is adopted h ,δw x ,δb]And [ δw ] y ,δb y ]To update [ w ] h ,w x ,b]And [ w ] y ,b y ]The method comprises the steps of carrying out a first treatment on the surface of the To better demonstrate the update process, the weights are represented by the symbol W, the gradient of the weights by δw, and the general formula for Adam update weights is:
m t ＝β 1 ·m t-1 +(1-β 1 )·δW t (29)
v t ＝β 2 ·v t-1 +(1-β 2 )·(δW t ) 2 (30)
wherein E is t As an error function, y t And Y t Respectively, a predicted value and an observed value. m is m t Andis the deviation of the first moment estimated and corrected, v t And->Is the bias of the bias moment estimation and correction moment. Beta 1 ,β 2 And epsilon is Adam parameters, 0.9,0.999 and 10 are respectively defaulted -8 The method comprises the steps of carrying out a first treatment on the surface of the η represents learning efficiency;
the predicted value is calculated by the previous forward propagation according to the above formula, and the updating weight is propagated back once, and each round of training is carried out by taking a batch with a certain size from the training set, and each batch is updated once.
(7) And inputting the test set into the trained ILSTM to predict, and obtaining a power load prediction result.
The invention has the beneficial effects that:
compared with the prior art, the method for predicting the electric load based on the improved long-short-term memory network has the advantages that the correlation factors influencing the electric load are analyzed through characteristic engineering, the characteristics are screened by adopting a two-stage characteristic extraction method, and the screening method comprises a filtering type and a wrapping type, so that the characteristic set of the adaptive model can be better obtained. Compared with the traditional LSTM, the power load prediction method based on the ILSTM saves time required by model training and improves prediction accuracy of the model. The prediction model provided by the invention can accurately predict the future power load, has higher practical application significance and is beneficial to popularization.
Drawings
FIG. 1 is a flow chart of a method for predicting power load of an improved long-term memory network provided by an embodiment of the present invention;
FIG. 2 is a block diagram of an ILSTM network provided by an embodiment of the present invention;
FIG. 3 is a view of a cause and effect relationship structure of the wind speed of Xinjiang Fuzheng station case and an equivalent tree structure thereof according to the embodiment of the present invention;
FIG. 4 is a graph showing comparison of the wind speed prediction results of the Xinjiang Fu station case according to the embodiment of the invention.
Detailed Description
The invention is further described below with reference to the accompanying drawings:
FIG. 1 is a general flow chart of a method for predicting the power load of an improved long-term and short-term memory network according to the present invention, which specifically comprises the following steps:
(1) Collecting power load and related factor data thereof, and primarily screening future influence of the power load according to historical power load;
the historical power load sequences are collected, the relation between the load of 168 time periods (one time period is 1 hour) in seven days of the history and the predicted load required by the day is analyzed by adopting the Maximum Information Coefficient (MIC), the calculated historical load with the MIC larger than 0.6 is formed into a pre-selected set, and the rest load is formed into a candidate set.
(2) Performing preliminary treatment on the power load related factors;
factors affecting the power load correlation are collected, including continuous features and discrete features, which require some form of processing to be applied by the predictive model. The invention adopts the LabelEncoder coding mode to code the date and time types. Wherein the discrete features include a date category and a time category. The date categories are classified into workdays, rest days and legal holidays, and the specific coding modes are shown in table 1 because the Chinese legal holidays comprise seven days at most:
table 1 date type encoding table
| Workday day | Saturday (Saturday) | (Sunday) | Holidays of the day | Holidays of two days |
| 1 | 2 | 3 | 4 | 5 |
| Three-day holiday | Holidays of four days | Five-day holiday | Holidays of six days | Seven day holiday |
| 6 | 7 | 8 | 9 | 10 |
In the time category, three periods are divided into one domain according to the power load characteristics, and the specific coding modes are as shown in table 2:
table 2 time type encoding table
| 00:00-03:00 | 03:00-06:00 | 06:00-09:00 | 09:00-12:00 |
| 1 | 2 | 3 | 4 |
| 12:00-15:00 | 15:00-18:00 | 18:00-21:00 | 21:00-24:00 |
| 5 | 6 | 7 | 8 |
(3) Performing secondary screening on the historical power load by combining the power load related factors and the prediction model;
the relationship between the load in the candidate subset and the load currently being predicted is analyzed. And analyzing the characteristics of the candidate subset by adopting a maximum correlation minimum redundancy method. The formula is as follows:
wherein S is m-1 Representing a preselect set, X-S m-1 Represents candidate set, c is category variable, m is the number of features, x j Is the j-th feature. I (x) j C) represents the number of mutual information between the j-th feature and the class variable.
The method finds features in the remaining feature space that maximize the value of the above formula based on the selected features. And (3) placing the features with the maximum calculated coefficient lambda into a preselected feature subset, combining other related factors influencing the power load, normalizing all the features, and then placing the features into a prediction model for prediction. Using the root mean square error RMSE as a discrimination coefficient, if the RMSE is reduced, the above process continues until the RMSE becomes high relative to before the feature is added.
(4) The screened final feature set is put into an improved long and short memory network ILSTM to be predicted, firstly, parameters of the NLSTM are set, including the number of nodes at an input layer, the number of nodes at a hidden layer, the number of nodes at an output layer, the learning efficiency, the batch size, the number of training rounds and the initial weight are set;
(5) And (3) training each weight parameter of the ILSTM on a training set by adopting an adaptive moment estimation (Adam) optimizer and combining a mini-batch mechanism.
The forward propagation step and the calculation formula of the information of the t period are as follows:
u t ＝σ(net ut ) (3)
o t ＝σ(net ot ) (8)
h t ＝o t *tanh(c t ) (9)
z t ＝W y ·h t +b y (10)
y t ＝σ(zt) (11)
in the above formula, net ut ,net ot And z t Is the state of the current phase of the t period; w (W) u ,/>And W is o Respectively their weight matrices; b u ,/>And b o Representing their respective bias vectors; u (u) t ,/>C t ,o t ,h t And y t Updating the gate, information state, cell state, output gate, hidden layer output and predicted value of time period t respectively; tanh and sigma are tanh and sigmoid activation functions, respectively; symbol and represent matrix multiplication and multiplication between matrix elements, respectively.
The step and the calculation formula of the information back propagation in the t-th period are as follows:
a. defining the most common square error function as the target to be optimized
b. Calculating errors of output layer
c. Calculating errors of hidden layers
W＝W-η·δW (28)
d. Adam optimization algorithm is adopted, and [ delta ] w is adopted h ,δw x ,δb]And [ δw ] y ,δb y ]To update [ w ] h ,w x ,b]And [ w ] y ,b y ]The method comprises the steps of carrying out a first treatment on the surface of the To better demonstrate the update process, the weights are represented by the symbol W, the gradient of the weights by δw, and the general formula for Adam update weights is:
m t ＝β 1 ·m t-1 +(1-β 1 )·δW t (29)
v t ＝β 2 ·v t-1 +(1-β 2 )·(δW t ) 2 (30)
wherein E is t As an error function, y t And Y t Respectively, a predicted value and an observed value. m is m t Andis the deviation of the first moment estimated and corrected, v t And->Is the bias of the bias moment estimation and correction moment. Beta 1 ,β 2 And epsilon is Adam parameters, 0.9,0.999 and 10 are respectively defaulted -8 The method comprises the steps of carrying out a first treatment on the surface of the η represents learning efficiency;
the predicted value is calculated by the previous forward propagation according to the above formula, and the updating weight is propagated back once, and each round of training is carried out by taking a batch with a certain size from the training set, and each batch is updated once.
(6) And inputting the test set into the trained ILSTM to predict, and obtaining a power load prediction result.
Fig. 2 shows a block diagram of an ILSTM network.
Fig. 3 shows a radar profile of the power load maximum information coefficient for the first 7 days (168 periods).
The application of the invention is further described below in connection with specific experiments.
The invention predicts the electric load of the Wuhan in China by taking the electric load of the Wuhan in China as a target and adopts meteorological data of two months from the day of 22 days of 3 months in 2015 to the day of 23 months in 5 years. The data time step is 1 hour for 1488 periods, dividing the first 1191 periods into training sets and the last 297 periods into test sets. Relevant factors that affect the electrical load include temperature, humidity, dew point, date category and time category. The Maximum Information Coefficient (MIC) for the load of the previous 7 days was calculated as shown in fig. 3. And the load is divided into a pre-selection set and a candidate set according to MIC. And further screening the load by adopting a maximum correlation minimum redundancy method, and combining the model to obtain a final feature set. And combining the feature set and the forecasting factors into a training set to train a model, and finally inputting the testing set into the trained ILSTM to predict, so as to obtain a power load prediction result.
To verify the predictive performance of the ILSTM, the following eight models were constructed to predict and compare the average wind speed:
(1) H-ILSTM: the method adopts ILSTM, and the feature selection adopts the feature extraction method of the invention;
(2) ILSTM: the method adopts ILSTM, and the characteristic selection only adopts a single maximum information coefficient method for analysis;
(3) H-LSTM: the method adopts the traditional LSTM, and the characteristic selection adopts the characteristic extraction method of the invention;
(4) LSTM: the method adopts the traditional LSTM, and the characteristic selection only adopts a single maximum information coefficient method for analysis;
(5) H-GRU: the method adopts GRU, and the characteristic selection adopts the characteristic extraction method of the invention;
(6) GRU: the method adopts GRU, and the characteristic selection only adopts a single maximum information coefficient method for analysis;
(7) H-SVR: SVR is adopted in the method, and the feature selection adopts the feature extraction method of the invention;
(8) SVR: the method adopts SVR, and the feature selection only adopts a single maximum information coefficient method for analysis;
to avoid the effect of randomness, each model was run 10 times to average. Table 3 lists the evaluation indexes of the eight model predictions. The evaluation index adopts Root Mean Square Error (RMSE) and average absolute error percent (MAPE) and average absolute error (MAE), and the smaller the value, the higher the prediction precision. As can be seen from Table 3, the prediction accuracy of H-ILSTM is higher than that of ILSTM, which demonstrates that the feature extraction method of the present invention improves the prediction accuracy. The prediction accuracy of H-ILSTM is higher than that of H-LSTM and the time required for training (TT) is short, which indicates that the method of the invention ILSTM is better than standard LSTM. The difference in the prediction accuracy of the 8 models can be seen more clearly in fig. 4.
Table 3 eight model predictive index tables
The foregoing has shown and described the basic principles and main features of the present invention and the advantages of the present invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, and that the above embodiments and descriptions are merely illustrative of the principles of the present invention, and various changes and modifications may be made without departing from the spirit and scope of the invention, which is defined in the appended claims. The scope of the invention is defined by the appended claims and equivalents thereof.
Claims (1)
1. An improved long-term and short-term memory network-based power load prediction method is characterized by comprising the following steps of:
firstly, primarily screening a characteristic of a historical load according to the historical power load as an input of a model; analyzing the historical power load by adopting the maximum information coefficient MIC, and taking the historical load with the maximum information coefficient larger than 0.6 as a pre-selection set [ x ] 1 ,x 2 ,,x k ]The remaining unselected historical loads are used as an alternative set [ c ] 1, c 2 ,,c k ]；
Secondly, considering a correlation factor influencing the power load, and adopting a maximum correlation minimum redundancy method to carry out secondary screening on the historical load; collecting data of a correlation factor affecting the electrical load including temperature TI 1 ,TI 2 ,,TI k ]Humidity [ HI ] 1 ,HI 2 ,,HI k ]Dew point [ DI ] 1 ,DI 2 ,,DI k ]Date, time period; the date type and the time period type are discrete characteristics, and the date type and the time period type are coded; namely, discrete data are converted into numbers between 1 and n, wherein n is the number of different values of a list and can be considered as the number of all different values of a certain feature; further analyzing the relevant factors influencing the power load by adopting a feature extraction method to obtain a feature set; the feature extraction method adopts a wrapped feature selection method to carry out secondary screening on the historical load; analyzing the historical load under the pre-selection set and the historical load under the alternative set by adopting a maximum correlation minimum redundancy method to obtain the characteristic of the maximum coefficient, putting the characteristic under the pre-selection set, and combining with other related factors which influence the load to form a preliminary training set TR= [ x ] tr ,Y]And a test set te= [ x ] consisting of only predictors te ]And normalize the dataIs processed by (1); training the preliminary training set by adopting an improved long and short memory network, comparing the obtained prediction result with a test set, and judging whether to continue to add features to the preselect set or not by adopting root mean square error as a threshold value; when the RMSE is adopted for discrimination, the RMSE obtained before the feature is added is used as a record, the RMSE obtained after the feature is added is compared with the record, and if the RMSE is larger after the feature is added, the feature is stopped being added;
thirdly, taking the screened final feature set as the input of a model, and predicting by adopting an improved long and short memory network to obtain a final prediction result; the improved long and short memory network ILSTM forward propagates the information in the t period and has the following calculation formula:
u t ＝σ(net ut ) (2)
o t ＝σ(net ot ) (7)
h t ＝o t *tanh(c t ) (8)
z t ＝W y ·h t +b y (9)
y t ＝σ(zt) (10)
in the above formula, net ut ,net ot And z t Is the state of the current phase of the t period; w (W) u ,/>And W is o Respectively their weight matrices; b u ,/>And b o Representing their respective bias vectors; u (u) t ,/>C t ,o t ,h t And y t Updating the gate, information state, cell state, output gate, hidden layer output and predicted value of time period t respectively; tanh and sigma are tanh and sigmoid activation functions, respectively; symbol and represent matrix multiplication and multiplication between matrix elements, respectively;
in the ILSTM, the step and the calculation formula of back propagation of the information in the t-th period are as follows:
a. defining the most common square error function as the target to be optimized
b. Calculating errors of output layer
c. Calculating errors of hidden layers
W＝W-η·δW (27)。
In the ILSTM, the step of updating the weight of the t-th period includes:
adam optimization algorithm is adopted, and [ delta ] w is adopted h ,δw x ,δb]And [ δw ] y ,δb y ]To update [ w ] h ,w x ,b]And [ w ] y ,b y ]The method comprises the steps of carrying out a first treatment on the surface of the To better demonstrate the update process, the weights are represented by the symbol W, the gradient of the weights by δw, and the general formula for Adam update weights is:
m t ＝β 1 ·m t-1 +(1-β 1 )·δW t (28)
v t ＝β 2 ·v t-1 +(1-β 2 )·(δW t ) 2 (29)
wherein E is t As an error function, y t And Y t Respectively a predicted value and an observed value; m is m t Andis the deviation of the first moment estimated and corrected, v t And->The bias moment estimation and correction are the bias of the second moment; beta 1 ,β 2 And epsilon is Adam parameters, 0.9,0.999 and 10 are respectively defaulted -8 The method comprises the steps of carrying out a first treatment on the surface of the η represents learning efficiency;
the predicted value is calculated by the previous forward propagation according to the above formula, and the updating weight is propagated back once, and each round of training is carried out by taking a batch with a certain size from the training set, and each batch is updated once.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010878240.7A CN111985719B (en) | 2020-08-27 | 2020-08-27 | Power load prediction method based on improved long-term and short-term memory network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010878240.7A CN111985719B (en) | 2020-08-27 | 2020-08-27 | Power load prediction method based on improved long-term and short-term memory network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111985719A CN111985719A (en) | 2020-11-24 |
| CN111985719B true CN111985719B (en) | 2023-07-25 |
Family
ID=73441162
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010878240.7A Active CN111985719B (en) | 2020-08-27 | 2020-08-27 | Power load prediction method based on improved long-term and short-term memory network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111985719B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113505923B (en) * | 2021-07-08 | 2022-06-17 | 广西大学 | Regional power grid short-term load prediction method and system |
| CN113837480B (en) * | 2021-09-29 | 2023-11-07 | 河北工业大学 | Impact load prediction method based on improved GRU and differential error compensation |
| CN114187059A (en) * | 2021-12-27 | 2022-03-15 | 广东工业大学 | Short-term electricity price prediction method |
| CN114881343B (en) * | 2022-05-18 | 2023-11-14 | 清华大学 | Short-term load prediction method and device for power system based on feature selection |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106779505A (en) * | 2017-02-28 | 2017-05-31 | 中国南方电网有限责任公司 | A kind of transmission line malfunction method for early warning driven based on big data and system |
| CN108830487A (en) * | 2018-06-21 | 2018-11-16 | 王芊霖 | Methods of electric load forecasting based on long neural network in short-term |
| CN109948845A (en) * | 2019-03-15 | 2019-06-28 | 国网江苏省电力有限公司经济技术研究院 | A kind of distribution network load shot and long term Memory Neural Networks prediction technique |
| CN110298501A (en) * | 2019-06-21 | 2019-10-01 | 河海大学常州校区 | Electric load prediction technique based on long Memory Neural Networks in short-term |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109214948A (en) * | 2018-09-25 | 2019-01-15 | 新智数字科技有限公司 | A kind of method and apparatus of electric system heat load prediction |
-
2020
- 2020-08-27 CN CN202010878240.7A patent/CN111985719B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106779505A (en) * | 2017-02-28 | 2017-05-31 | 中国南方电网有限责任公司 | A kind of transmission line malfunction method for early warning driven based on big data and system |
| CN108830487A (en) * | 2018-06-21 | 2018-11-16 | 王芊霖 | Methods of electric load forecasting based on long neural network in short-term |
| CN109948845A (en) * | 2019-03-15 | 2019-06-28 | 国网江苏省电力有限公司经济技术研究院 | A kind of distribution network load shot and long term Memory Neural Networks prediction technique |
| CN110298501A (en) * | 2019-06-21 | 2019-10-01 | 河海大学常州校区 | Electric load prediction technique based on long Memory Neural Networks in short-term |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111985719A (en) | 2020-11-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111985719B (en) | Power load prediction method based on improved long-term and short-term memory network | |
| CN109754113B (en) | Load prediction method based on dynamic time warping and long-and-short time memory | |
| Wang et al. | Adaptive learning hybrid model for solar intensity forecasting | |
| CN111260136A (en) | Building short-term load prediction method based on ARIMA-LSTM combined model | |
| CN111563706A (en) | Multivariable logistics freight volume prediction method based on LSTM network | |
| CN109284866B (en) | Commodity order prediction method and device, storage medium and terminal | |
| CN110619389A (en) | Load prediction method and system of combined cooling heating and power system based on LSTM-RNN | |
| CN111027772A (en) | Multi-factor short-term load prediction method based on PCA-DBILSTM | |
| CN111553543A (en) | Power load prediction method based on TPA-Seq2Seq and related assembly | |
| CN113554466A (en) | Short-term power consumption prediction model construction method, prediction method and device | |
| CN113052214B (en) | Heat exchange station ultra-short-term heat load prediction method based on long-short-term time sequence network | |
| CN113537469B (en) | Urban water demand prediction method based on LSTM network and Attention mechanism | |
| CN111160659A (en) | Power load prediction method considering temperature fuzzification | |
| CN116542382A (en) | Sewage treatment dissolved oxygen concentration prediction method based on mixed optimization algorithm | |
| CN115409292A (en) | Short-term load prediction method for power system and related device | |
| CN112508286A (en) | Short-term load prediction method based on Kmeans-BilSTM-DMD model | |
| CN115456306A (en) | Bus load prediction method, system, equipment and storage medium | |
| CN115766125A (en) | Network flow prediction method based on LSTM and generation countermeasure network | |
| CN114091776A (en) | K-means-based multi-branch AGCNN short-term power load prediction method | |
| CN117291069A (en) | LSTM sewage water quality prediction method based on improved DE and attention mechanism | |
| CN116703644A (en) | Attention-RNN-based short-term power load prediction method | |
| CN116885705A (en) | Regional load prediction method and device | |
| CN111539558A (en) | Power load prediction method adopting optimized extreme learning machine | |
| CN116542701A (en) | Carbon price prediction method and system based on CNN-LSTM combination model | |
| CN115860277A (en) | Data center energy consumption prediction method and system |
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 |