CN112508299A - Power load prediction method and device, terminal equipment and storage medium - Google Patents
Power load prediction method and device, terminal equipment and storage medium Download PDFInfo
- Publication number
- CN112508299A CN112508299A CN202011515128.3A CN202011515128A CN112508299A CN 112508299 A CN112508299 A CN 112508299A CN 202011515128 A CN202011515128 A CN 202011515128A CN 112508299 A CN112508299 A CN 112508299A
- Authority
- CN
- China
- Prior art keywords
- power load
- load data
- historical power
- historical
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 84
- 238000003860 storage Methods 0.000 title claims abstract description 20
- 238000012360 testing method Methods 0.000 claims abstract description 67
- 238000012549 training Methods 0.000 claims abstract description 45
- 238000012545 processing Methods 0.000 claims abstract description 19
- 238000004590 computer program Methods 0.000 claims description 24
- 230000015654 memory Effects 0.000 claims description 16
- 238000009499 grossing Methods 0.000 claims description 8
- 230000006870 function Effects 0.000 description 15
- 238000004364 calculation method Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 6
- 230000006872 improvement Effects 0.000 description 5
- 239000011159 matrix material Substances 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 238000012937 correction Methods 0.000 description 3
- 238000003066 decision tree Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000007774 longterm Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 230000001427 coherent effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 239000002994 raw material Substances 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/067—Enterprise or organisation modelling
-
- 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)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- General Business, Economics & Management (AREA)
- General Health & Medical Sciences (AREA)
- Entrepreneurship & Innovation (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Game Theory and Decision Science (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Development Economics (AREA)
- Mathematical Physics (AREA)
- Educational Administration (AREA)
- Primary Health Care (AREA)
- Water Supply & Treatment (AREA)
- Public Health (AREA)
- Supply And Distribution Of Alternating Current (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a power load prediction method, a power load prediction device, terminal equipment and a storage medium, wherein the method comprises the following steps: acquiring historical power load data of at least three time periods; processing the first historical power load data by using a weighted moving average method to obtain a plurality of groups of first historical power load data, and configuring different weights; arranging each group of first historical power load data according to a time sequence, and inputting the first historical power load data into an LSTM (least squares metric) model for training; inputting the second historical power load data serving as a test set into the LSTM model for testing to obtain a test result; taking the difference value between the test result and the second historical power load data as a training set of the GBDT model, and training the GBDT model; testing the GBDT model by using the third history power load data as a test set; and predicting the power load of the time interval to be measured according to a weighted moving average method, an LSTM model and a GBDT model. The invention can improve the universality and the precision of the power load model.
Description
Technical Field
The present invention relates to the field of power load prediction technologies, and in particular, to a power load prediction method, an apparatus, a terminal device, and a storage medium.
Background
With the gradual advance and deepening of electric power market reformation, more and more provinces develop electric power market settlement test operation and simulation analysis work, taking the test operation simulation result of the Guangdong electric power market in 2020 and 8 months as an example, the result of load prediction can be found to obviously influence settlement benefits to some extent, further influence market confidence and reporting strategies of market participants, further influence safe, stable, efficient and high-quality operation of a future power grid, and therefore the importance of load prediction can play an increasingly important role under a new electric power system.
With the hundred-year development of power systems, a great number of methods and systems have been derived in the field of load prediction, wherein a Long Short-Term Memory network (LSTM) is a commonly used prediction method at present, but the method has some disadvantages, such as time delay, low training efficiency, low prediction accuracy, effectiveness for data within a certain time length, and inability to isolate the influence of individual extreme historical data on a model, and therefore, the method cannot meet the requirements of the existing power market environment on high accuracy and universality of load prediction data.
Disclosure of Invention
The embodiment of the invention aims to provide a power load prediction method, a power load prediction device, terminal equipment and a storage medium.
To achieve the above object, an embodiment of the present invention provides a power load prediction method, including the following steps:
acquiring historical power load data of at least three time periods; wherein the historical power load data comprises first, second, and third historical power load data;
processing the first historical power load data by using a weighted moving average method to obtain multiple groups of first historical power load data with different influence degrees on the power load, and configuring different weights for each group of first historical power load data; wherein the greater the degree of influence, the greater the weight of the first historical power load data configuration;
arranging each set of configured first historical power load data according to a time sequence, and inputting a preset LSTM model for training;
inputting the second historical power load data serving as a test set into an LSTM model for testing to obtain a corresponding test result;
taking the difference value between the test result and the second historical power load data as a preset GBDT model training set, and training the GBDT model;
testing the GBDT model by using the third history power load data as a test set;
and predicting the power load of the time interval to be measured according to a prediction formula corresponding to the weighted moving average method, the trained LSTM model and the trained GBDT model.
Preferably, before the processing the first historical power load data by using the weighted moving average method, the method further includes:
and processing the first historical power load data by using a standard deviation method, and rejecting bad data or correcting the bad data.
Preferably, the LSTM model comprises transverse layers and longitudinal layers; the transverse layers are spliced in sequence according to a time sequence, the longitudinal layers comprise five layers, and all the layers are spliced in a full-connection mode.
Preferably, the method further comprises:
when the LSTM model is initialized, adjusting a forced forgetting factor according to the weight configured by each group of first historical power load data; and the forced forgetting factor corresponding to the power load data with the weight smaller than the preset weight threshold is 0.
Preferably, the historical power load data includes power network, time node, date, temperature, humidity, whether it is raining, wind level, whether it is a holiday and load.
Preferably, the variable input by the GBDT model at initialization includes at least one of whether it is rainfall, whether it is holiday, and wind level.
Another embodiment of the present invention provides a power load prediction apparatus, including:
the data acquisition module is used for acquiring historical power load data of at least three time periods; wherein the historical power load data comprises first, second, and third historical power load data;
the smoothing processing module is used for processing the first historical power load data by using a weighted moving average method to obtain multiple groups of first historical power load data with different influence degrees on the power load, and different weights are configured for each group of first historical power load data; wherein the greater the degree of influence, the greater the weight of the first historical power load data configuration;
the first training module is used for arranging each set of configured first historical power load data according to a time sequence and inputting a preset LSTM model for training;
the first test module is used for inputting the second historical power load data serving as a test set into an LSTM model for testing to obtain a corresponding test result;
the second training module is used for taking the difference value between the test result and the second historical power load data as a preset GBDT model training set to train the GBDT model;
the second testing module is used for testing the GBDT model by using the third history power load data as a testing set;
and the prediction module is used for predicting the power load of the time interval to be measured according to the prediction formula corresponding to the weighted moving average method, the trained LSTM model and the trained GBDT model.
Another embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor implements the power load prediction method according to any one of the above items when executing the computer program.
An embodiment of the present invention provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program runs, a device in which the computer-readable storage medium is located is controlled to execute the power load prediction method according to any one of the above items.
Compared with the prior art, the power load prediction method, the power load prediction device, the terminal equipment and the storage medium provided by the embodiment of the invention have the advantages that the influence of historical data on the load is visualized by using a weighted moving average method, and small-probability high-influence data are eliminated through threshold discrimination, so that the calculation speed of the LSTM is increased, the training difficulty is simplified, the model universality and the precision level are improved, then the GBDT is designed to inherit the calculation result of the LSTM, the small-probability influence factor is restored, and the model universality and the precision are further improved.
Drawings
Fig. 1 is a schematic flow chart illustrating a power load prediction method according to an embodiment of the present invention;
FIG. 2 is a schematic structural diagram of an LSTM model according to an embodiment of the present invention;
FIG. 3 is a flow chart illustrating a method for predicting a power load according to another embodiment of the present invention;
fig. 4 is a schematic structural diagram of an electrical load prediction apparatus according to an embodiment of the present invention;
fig. 5 is a schematic diagram of a terminal device according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Referring to fig. 1, a flow chart of a power load prediction method according to the embodiment of the present invention is shown, where the method includes steps S1 to S7:
s1, acquiring historical power load data of at least three time periods; wherein the historical power load data comprises first, second, and third historical power load data;
s2, processing the first historical power load data by using a weighted moving average method to obtain multiple groups of first historical power load data with different influence degrees on the power load, and configuring different weights for each group of first historical power load data; wherein the greater the degree of influence, the greater the weight of the first historical power load data configuration;
s3, arranging each set of configured first historical power load data according to a time sequence, and inputting a preset LSTM model for training;
s4, inputting the second historical power load data serving as a test set into an LSTM model for testing to obtain a corresponding test result;
s5, taking the difference value between the test result and the second historical power load data as a preset GBDT model training set, and training the GBDT model;
s6, testing the GBDT model by using the third history power load data as a test set;
and S7, predicting the power load of the time interval to be measured according to the prediction formula corresponding to the weighted moving average method, the trained LSTM model and the trained GBDT model.
Specifically, historical power load data for at least three time periods is acquired, wherein the historical power load data comprises attribute values related to the power load values, such as date, temperature, humidity, wind level and the like, besides the power load values; wherein the historical power load data comprises first historical power load data, second historical power load data, and third historical power load data; the first historical power load data is prior in time to the second historical power load data, which is prior in time to the third historical power load data. For example, if the power load of a month in a certain year needs to be predicted, historical data of at least one month, similar time periods of the previous three years, and load data of a week before and after the similar time periods are needed, for example, if the power load of 5 months in 2021 is predicted, load data of 5 months in 2016 (or 2019 or 2004), + load of 5 months in 2020 + load of a week before and after 5 months in three years are needed, in order to find a complete correspondence relationship between day types. If the load of the first week of 5 months in 2021 is predicted, data of at least one year is needed on the basis of the historical data, for example, year 2020 annual load data is needed, and if the year 2021 annual load is predicted, 2018 and 2020 full load data are needed.
Processing the first historical power load data by using a weighted moving average method to obtain multiple groups of first historical power load data with different influence degrees on the power load, and configuring different weights for each group of first historical power load data; wherein the greater the degree of influence the greater the weight of the first historical power load data configuration. Preferably, the plurality of sets of first historical power load data are three sets of first historical power load data, which are respectively long-term data, medium-term data and short-term data, and the three sets of first historical power load data do not refer to the time scale of the distance but the distance from the expected result. Preferably, the weighted moving average method is an exponential smoothing method.
To further understand the above, an example is described below, for example, if the load on a certain day of year 5 in 2019 is much lower than the average value on the same day of years 2018 and 2020, the day is found to be extremely storm by comparing the date type and the weather condition data, so that the day data is configured as long-term data and is given little weight when the weight is configured. Because the influence of the extreme storm is ignored, and the influence of the extreme storm can be introduced into a later-designed Gradient Boosting Decision Tree (GBDT), the configuration of the weighting after the weighted moving average method is used for processing and the selection of the input variables of the Decision Tree in the later period are combined skillfully, the calculation rate in the earlier period is improved, the actual situation of influencing the load is fitted, and the prediction of the power load has higher prediction precision and universality.
And arranging each set of the configured first historical power load data according to a time sequence, and inputting a preset LSTM model for training. It should be noted that although the input training data are arranged in chronological order, each data is provided with a corresponding weight.
Inputting the second historical power load data serving as a test set into the LSTM model for testing to obtain a corresponding test result;
and taking the difference value between the test result and the second historical power load data as a preset GBDT model training set, and training the GBDT model. Namely, the difference value between the test result and the true value of the LSTM model is used as the training set of the GBDT model, so that the GBDT model inherits the optimization result of the LSTM model, on one hand, the efficiency is improved, the number of layers of the LSTM model is reduced, on the other hand, the precision is improved, and the universality is improved.
The GBDT model is tested using the third history power load data as a test set. And when the test result is within the preset error range, the LSTM model and the GBDT model are trained and can be put into use.
And predicting the power load of the time interval to be measured according to a prediction formula corresponding to the weighted moving average method, the trained LSTM model and the trained GBDT model. Generally, the prediction formula corresponding to the weighted moving average method affects the packet classification of the first historical power load data, and is mainly related to the smoothing coefficient α and the threshold T of the prediction formula.
Preferably, the prediction formula of the weighted moving average method is St＝αxt+(1-α)St-1Wherein S istIs the smooth value of the t period; x is the number oftActual observed value of the t period; st-1Is the smooth value of the t-1 period; alpha is a smooth coefficient and the value range is (0, 1); will St-1＝αxt-1+(1-α)St-2，St-2＝αxt-1+(1-α)St-3，St-3＝αxt-1+(1-α)St-4… … into St＝αxt+(1-α)St-1Obtaining:
St＝αxt+α(1-α)xt-1+α(1-α)2xt-2+α(1-α)3xt-3+…+α(1-α)t-1x1+(1-α)tS0
the sum of the coefficients is:
when t → ∞, (1-. alpha.)t→ 0, coefficient sum → 1, so it can be said that StThe exponential weighted average of the observed values in the t period and before, the farther away from the t period, the smaller the coefficient of the data in each period, and the smaller the influence on the predicted value.
According to the power load prediction method provided by the embodiment of the invention, the influence of historical data on the load is visualized by using a weighted moving average method, and small-probability high-influence data are eliminated through threshold discrimination, so that the calculation speed of the LSTM is increased, the training difficulty is simplified, the universality and the precision level of the model are improved, then the GBDT is designed to inherit the calculation result of the LSTM, the small-probability influence factor is restored, and the universality and the precision of the model are further improved.
As an improvement of the above, before the processing the first historical power load data by using the weighted moving average method, the method further includes:
and processing the first historical power load data by using a standard deviation method, and rejecting bad data or correcting the bad data.
Specifically, in order to further improve the prediction accuracy, data cleaning is required, and the first historical power load data is processed by a standard deviation method to remove or correct the defective data. Calculating to obtain a standard deviation of the first historical power load data, calculating a difference value between each first historical power load data and the standard deviation, if the difference value is larger than 3 times of the standard deviation, determining that the data quality is poor, removing the data, and after removing the data, multiplying an average value of a plurality of historical data by a correction coefficient to serve as a correction value, wherein preferably, the value range of the correction coefficient is 1.03-1.09.
As an improvement of the above scheme, the LSTM model includes a transverse layer and a longitudinal layer; the transverse layers are spliced in sequence according to a time sequence, the longitudinal layers comprise five layers, and all the layers are spliced in a full-connection mode.
Specifically, the LSTM model includes transverse layers and longitudinal layers; the transverse layers are spliced in sequence according to a time sequence, the longitudinal layers comprise five layers, and all the layers are spliced in a full-connection mode. Fig. 2 is a schematic structural diagram of an LSTM model according to the embodiment of the present invention. As can be seen from FIG. 2, n transverse layers are arranged in the time direction, n longitudinal layers are arranged on each transverse layer, and a parameter X is inputiCorresponding to an output parameter of Yi,. Generally, before entering the horizontal layer and the vertical layer for training, an initial feature vector of training data is obtained, and preferably, the initial feature vector is obtained by using a Support Vector Machine (SVM) or a Convolutional Neural Network (CNN) method.
The following formula is used in the LSTM model training:
ft＝σ(Wfxxt+Wfhht-1+bf)
it＝σ(Wixxt+Wihht-1+bi)
ot＝σ(Woxxi+Wohht-1+bo)
St＝gt⊙it+St-1⊙ft
wherein f ist、it、gt、ot、St、htStates of a forgetting gate, an input node, an output gate, an intermediate output and a state unit are respectively set; wfxAnd WfhInput parameters x for forgetting gates respectivelytAnd an intermediate output parameter ht-1The matrix weight of (2); wixAnd WihInput parameters x for the input gates respectivelytAnd an intermediate output parameter ht-1The matrix weight of (2); wgxAnd WghInput parameters x of the input nodes respectivelytAnd an intermediate output parameter ht-1The matrix weight of (2); woxAnd WohInput parameters x for the output gates respectivelytAnd an intermediate output parameter ht-1The matrix weight of (2); bf、bi、bg、boA bias items of a forgetting gate, an input node, an output gate, an intermediate output and a state unit are respectively; an element in a vector is multiplied by a bit; sigma represents sigmoid function variation;
denotes tan hThe function changes.
As an improvement of the above, the method further comprises:
when the LSTM model is initialized, adjusting a forced forgetting factor according to the weight configured by each group of power load data; and the forced forgetting factor corresponding to the power load data with the weight smaller than the preset weight threshold is 0.
Specifically, when the LSTM model is initialized, a forced forgetting factor is adjusted according to the weight configured by each group of first historical power load data; the forced forgetting factor corresponding to the power load data with the weight smaller than the preset weight threshold is 0, and the LSTM model omits the data with the forced forgetting factor of 0 during calculation, so that some influence factors with low occurrence probability are eliminated, and the operation rate is further improved. For the forced forgetting factor of 1 corresponding to the power load data with the weight equal to or greater than the preset weight threshold, the LSTM model is mainly considered when calculating the data relationship before and after the time series.
As an improvement of the scheme, the historical power load data comprises a power network, a time node, a date, a temperature, a humidity, whether rainfall occurs, a wind power level, whether holidays occur and a load.
Specifically, historical power load data includes power network, time node, date, temperature, humidity, whether it is raining, wind level, whether it is a holiday, and load. Whether the holiday is a holiday or not includes the conditions of working days, weekends, important holidays and the like. Whether rainfall is rainfall or not can be subdivided according to the difference of rainfall. That is, the factors that affect the load can be enriched and collected as needed to be the independent variables for model training. The data acquisition method includes, for example: the electrical load may be derived from a SCADA system of the electrical power system; weather data can be obtained from China weather bureau websites; economic data and holiday data are sourced from government related websites.
As an improvement of the scheme, the variable input by the GBDT model during initialization comprises at least one of whether rainfall is present, whether holidays are present and wind power level.
Specifically, the variables input by the GBDT model at initialization include at least one of whether it is raining, whether it is a holiday, and a wind level.
In order to enhance understanding of the present embodiment, the following description is given as an example. Taking the case of a storm of 5 months and 1 day, the GBDT model inputs the variables: a is 3(1, 2,3 respectively indicate working day, weekend, holiday), b is 1(0, 1 respectively indicate no rain, rain), c is 12(12 indicates 12 wind power), and d is 50 (indicating rainfall 50mm at the current analysis time).
If the condition to be measured at a certain time of a certain day meets a 3, b 1, c 11, 50< d ≦ 100, and e <20 ℃, then it is classified into this class by GBDT with a high probability, and although it does not completely correspond to the variables input by the GBDT model, GBDT calculates a load value for the day based on weighted average as a prediction result.
In addition, for deepening understanding of the present invention, referring to fig. 3, a flowchart of a power load prediction method according to another embodiment of the present invention is shown. In fig. 3, the target to be predicted includes the power network to be subjected to load prediction, the time length, the start date, the type of load prediction, the influence factor, whether to consider the network topology change, and the like. The time length of a power network such as a Guangdong power grid whole network is 3 months, 10 years and the like, the starting date is 2099 years, 1 month and 2 days and the like, the load prediction type is such as bus 96-point load prediction, whole network daily maximum load prediction and the like, and influence factors such as temperature, humidity, national economic growth and the like. This step is to determine whether there is a model, such as may be set by a user, that three of the above criteria are met are considered the same, or that only one condition is allowed to be inconsistent, such as a predicted duration.
The GBDT input variables refer to actual attributes that affect the change of load data, that is, various judgment conditions used when the cart tree branches in the GBDT method, and the present invention includes: year, month, date, time, same time of day before minus a period load (for example, each point of 15min is predicted, that is 15min), same time of day before plus a period load, same time of day after plus or minus a period load, same time of day before plus or minus a period load of two days after, t-1p (representing the current time minus 1 period) load of the present day, t-2 p-5 p load of the present day, t +2 p-5 p load of the present day, temperature, humidity, rainfall probability, solar radiation index, snowfall probability (here, all weather indexes desired for classification can be included and quantization standards are given), date type (holiday, weekend, etc.), GDP increase value predicted by the degree of year, and load at the current time.
GBDT is a configuration value obtained by constructing a decision tree m times to minimize a loss function, preferably a squared loss function of
Negative gradient of the loss function is yi-f(xi) Wherein x isiAs independent variables, e.g. temperature, humidity, etc., yiFor the fitted target, e.g., predicted load, fm is the mth weak learning period, F m is the combination of the previous m weak learning periods, given the sample set T { (X)1，y1),(X2，y2),……，(XN，yN)}，XNFor the Nth attribute vector, each attribute vector contains a z-dimensional variable, XN＝{xi1,xi2,……，xiz}，yiFor the target variable, N is the number of samples in the set. And (3) setting the data of the weak regression model as M, selecting a square error loss function as the loss function, and executing the Gradient Boosting algorithm according to the following steps:
1. initialization
First, a first regression tree, f, is created1(x) In the regression problem, it is the result of using the regression tree directly to find the target value, so
Wherein c is the forced forgetting factor of the LSTM model.
2. Iteration
a. For the 2 nd to m th regression trees, we calculate the training target for each tree, i.e. the residual of the previous results:
b. for the current mth subtree, a feasible segmentation point and a threshold value of the mth subtree need to be traversed, and a parameter corresponding to the optimal predicted value c is found, so that a residual error is approximated as much as possible, and the following results are obtained:
wherein R ism,jThe method refers to a set of predicted values of leaf nodes in all the partitioning methods of the mth subtree, namely, the predicted values which can be reached by the mth regression tree. Where j ranges from 1,2,3 … … j.
Then, update
I is a function if the sample falls on Rm,jOn a node, then I equals 1, otherwise I equals 0.
3. Obtaining a regression tree:
referring to fig. 4, a schematic structural diagram of an electrical load prediction apparatus according to an embodiment of the present invention is shown, where the apparatus includes:
the data acquisition module 11 is configured to acquire historical power load data of at least three time periods; wherein the historical power load data comprises first, second, and third historical power load data;
the smoothing processing module 12 is configured to process the first historical power load data by using a weighted moving average method to obtain multiple groups of first historical power load data with different degrees of influence on the power load, and configure different weights for each group of the first historical power load data; wherein the greater the degree of influence, the greater the weight of the first historical power load data configuration;
the first training module 13 is configured to arrange each set of configured first historical power load data according to a time sequence, and input a preset LSTM model for training;
the first testing module 14 is configured to input the second historical power load data as a testing set into an LSTM model for testing, so as to obtain a corresponding testing result;
the second training module 15 is configured to train the GBDT model by using a difference between the test result and the second historical power load data as a preset training set of the GBDT model;
a second testing module 16, configured to test the GBDT model by using the third historical power load data as a test set;
and the prediction module 17 is configured to predict the power load of the time interval to be measured according to the prediction formula corresponding to the weighted moving average method, the trained LSTM model, and the trained GBDT model.
The power load prediction apparatus provided in the embodiment of the present invention can implement all the processes of the power load prediction method described in any one of the above embodiments, and the functions and implemented technical effects of each module and unit in the apparatus are respectively the same as those of the power load prediction method described in the above embodiment, and are not described again here.
Referring to fig. 5, the terminal device provided in the embodiment of the present invention includes a processor 10, a memory 20, and a computer program stored in the memory 20 and configured to be executed by the processor 10, and when the processor 10 executes the computer program, the method for predicting the power load according to any of the above embodiments is implemented.
Illustratively, the computer program may be divided into one or more modules/units, which are stored in the memory 20 and executed by the processor 10 to implement the present invention. One or more of the modules/units may be a series of computer program instruction segments capable of performing specific functions, the instruction segments being used to describe the execution of the computer program in a method for power load prediction. For example, the computer program may be divided into a data acquisition module, a smoothing module, a first training module, a first testing module, a second training module, a second testing module, and a prediction module, and the specific functions of each module are as follows:
the data acquisition module 11 is configured to acquire historical power load data of at least three time periods; wherein the historical power load data comprises first, second, and third historical power load data;
the smoothing processing module 12 is configured to process the first historical power load data by using a weighted moving average method to obtain multiple groups of first historical power load data with different degrees of influence on the power load, and configure different weights for each group of the first historical power load data; wherein the greater the degree of influence, the greater the weight of the first historical power load data configuration;
the first training module 13 is configured to arrange each set of configured first historical power load data according to a time sequence, and input a preset LSTM model for training;
the first testing module 14 is configured to input the second historical power load data as a testing set into an LSTM model for testing, so as to obtain a corresponding testing result;
the second training module 15 is configured to train the GBDT model by using a difference between the test result and the second historical power load data as a preset training set of the GBDT model;
a second testing module 16, configured to test the GBDT model by using the third historical power load data as a test set;
and the prediction module 17 is configured to predict the power load of the time interval to be measured according to the prediction formula corresponding to the weighted moving average method, the trained LSTM model, and the trained GBDT model.
The terminal device can be a desktop computer, a notebook, a palm computer, a cloud server and other computing devices. The terminal device may include, but is not limited to, a processor, a memory. It will be understood by those skilled in the art that the schematic diagram 5 is merely an example of a terminal device, and is not intended to limit the terminal device, and may include more or less components than those shown, or some components may be combined, or different components, for example, the terminal device may further include an input-output device, a network access device, a bus, etc.
The Processor 10 may be a Central Processing Unit (CPU), other general purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf Programmable Gate Array (FPGA) or other Programmable logic device, discrete Gate or transistor logic, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor 10 may be any conventional processor or the like, the processor 10 being the control center of the terminal device and connecting the various parts of the whole terminal device with various interfaces and lines.
The memory 20 may be used to store the computer programs and/or modules, and the processor 10 implements various functions of the terminal device by running or executing the computer programs and/or modules stored in the memory 20 and calling data stored in the memory 20. The memory 20 may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required by at least one function (such as a sound playing function, an image playing function, etc.), and the like; the storage data area may store data (such as audio data, a phonebook, etc.) created according to the use of the cellular phone, and the like. In addition, the memory 20 may include high speed random access memory, and may also include non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), at least one magnetic disk storage device, a Flash memory device, or other volatile solid state storage device.
Wherein, the module integrated with the terminal device can be stored in a computer readable storage medium if it is implemented in the form of a software functional unit and sold or used as a stand-alone product. Based on such understanding, all or part of the flow of the method according to the embodiments of the present invention may also be implemented by a computer program, which may be stored in a computer-readable storage medium and can implement the steps of the embodiments of the method when the computer program is executed by a processor. The computer program includes computer program code, and the computer program code may be in a source code form, an object code form, an executable file or some intermediate form. The computer readable medium may include: any entity or device capable of carrying computer program code, recording medium, U.S. disk, removable hard disk, magnetic disk, optical disk, computer Memory, Read-Only Memory (ROM), Random Access Memory (RAM), electrical carrier wave signals, telecommunications signals, software distribution media, and the like. It should be noted that the computer readable medium may contain other components which may be suitably increased or decreased as required by legislation and patent practice in jurisdictions, for example, in some jurisdictions, in accordance with legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunications signals.
The embodiment of the present invention further provides a computer-readable storage medium, where the computer-readable storage medium includes a stored computer program, and when the computer program runs, a device in which the computer-readable storage medium is located is controlled to execute the method for predicting a power load according to any of the above embodiments.
In summary, the power load prediction method, the apparatus, the terminal device and the storage medium provided by the embodiments of the present invention adopt a combination method of exponential smoothing + LSTM + GBDT, so that the disadvantage of a single method is effectively avoided, the effect of 1+1+1>3 is obtained, and the method is a creative solution for large-capacity coherent load prediction. According to the invention, the influence of random small-probability high-influence events on the load prediction value is creatively solved by smooth array weight configuration and introduction of event influence comprehensive factors, firstly, weight configuration is adopted to ensure that small-probability events do not influence a large number of results in the LSTM operation process, namely pain notes are selectively deleted, and then the specific influence is restored through event comprehensive influence attributes in the GBDT stage, so that the authenticity of the load under the condition of small-probability events is kept. When the LSTM and GBDT methods are combined, the result of the LSTM training set 1 is used as the residual value construction raw material of the GBDT training set 2, the GBDT inherits the LSTM optimization result creatively, and therefore efficiency is improved, the number of LSTM layers is reduced, accuracy is improved, and universality is improved.
While the foregoing is directed to the preferred embodiment of the present invention, it will be understood by those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the invention.
Claims (9)
1. A power load prediction method is characterized by comprising the following steps:
acquiring historical power load data of at least three time periods; wherein the historical power load data comprises first, second, and third historical power load data;
processing the first historical power load data by using a weighted moving average method to obtain multiple groups of first historical power load data with different influence degrees on the power load, and configuring different weights for each group of first historical power load data; wherein the greater the degree of influence, the greater the weight of the first historical power load data configuration;
arranging each set of configured first historical power load data according to a time sequence, and inputting a preset LSTM model for training;
inputting the second historical power load data serving as a test set into an LSTM model for testing to obtain a corresponding test result;
taking the difference value between the test result and the second historical power load data as a preset GBDT model training set, and training the GBDT model;
testing the GBDT model by using the third history power load data as a test set;
and predicting the power load of the time interval to be measured according to a prediction formula corresponding to the weighted moving average method, the trained LSTM model and the trained GBDT model.
2. The power load prediction method of claim 1, wherein prior to the processing the first historical power load data using a weighted moving average, further comprising:
and processing the first historical power load data by using a standard deviation method, and rejecting bad data or correcting the bad data.
3. The power load prediction method of claim 1 wherein the LSTM model comprises a transverse layer and a longitudinal layer; the transverse layers are spliced in sequence according to a time sequence, the longitudinal layers comprise five layers, and all the layers are spliced in a full-connection mode.
4. The power load prediction method of claim 1, further comprising:
when the LSTM model is initialized, adjusting a forced forgetting factor according to the weight configured by each group of first historical power load data; and the forced forgetting factor corresponding to the power load data with the weight smaller than the preset weight threshold is 0.
5. A power load prediction method according to any of claims 1-4, characterized in that the historical power load data comprises power network, time node, date, temperature, humidity, whether it is raining, wind level, whether it is a holiday and load.
6. A power load prediction method according to claim 3 wherein the variables input by the GBDT model at initialization include at least one of whether it is raining, whether it is a holiday, and a wind level.
7. An electric load prediction apparatus, comprising:
the data acquisition module is used for acquiring historical power load data of at least three time periods; wherein the historical power load data comprises first, second, and third historical power load data;
the smoothing processing module is used for processing the first historical power load data by using a weighted moving average method to obtain multiple groups of first historical power load data with different influence degrees on the power load, and different weights are configured for each group of first historical power load data; wherein the greater the degree of influence, the greater the weight of the first historical power load data configuration;
the first training module is used for arranging each set of configured first historical power load data according to a time sequence and inputting a preset LSTM model for training;
the first test module is used for inputting the second historical power load data serving as a test set into an LSTM model for testing to obtain a corresponding test result;
the second training module is used for taking the difference value between the test result and the second historical power load data as a preset GBDT model training set to train the GBDT model;
the second testing module is used for testing the GBDT model by using the third history power load data as a testing set;
and the prediction module is used for predicting the power load of the time interval to be measured according to the prediction formula corresponding to the weighted moving average method, the trained LSTM model and the trained GBDT model.
8. A terminal device comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, the processor implementing the power load prediction method according to any one of claims 1 to 6 when executing the computer program.
9. A computer-readable storage medium, comprising a stored computer program, wherein the computer program, when executed, controls an apparatus in which the computer-readable storage medium is located to perform the power load prediction method according to any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011515128.3A CN112508299A (en) | 2020-12-18 | 2020-12-18 | Power load prediction method and device, terminal equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011515128.3A CN112508299A (en) | 2020-12-18 | 2020-12-18 | Power load prediction method and device, terminal equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112508299A true CN112508299A (en) | 2021-03-16 |
Family
ID=74922663
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011515128.3A Pending CN112508299A (en) | 2020-12-18 | 2020-12-18 | Power load prediction method and device, terminal equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112508299A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113131476A (en) * | 2021-04-28 | 2021-07-16 | 南方电网科学研究院有限责任公司 | Power load prediction method |
| CN113139698A (en) * | 2021-05-13 | 2021-07-20 | 国网山东省电力公司枣庄供电公司 | Load prediction method, device and equipment |
| CN114118641A (en) * | 2022-01-29 | 2022-03-01 | 华控清交信息科技(北京)有限公司 | Wind power plant power prediction method, GBDT model longitudinal training method and device |
| CN114841457A (en) * | 2022-05-18 | 2022-08-02 | 上海玫克生储能科技有限公司 | Power load estimation method and system, electronic device, and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180150783A1 (en) * | 2016-08-24 | 2018-05-31 | Clari Inc. | Method and system for predicting task completion of a time period based on task completion rates and data trend of prior time periods in view of attributes of tasks using machine learning models |
| CN110084424A (en) * | 2019-04-25 | 2019-08-02 | 国网浙江省电力有限公司 | A kind of Methods of electric load forecasting based on LSTM and LGBM |
| CN110245801A (en) * | 2019-06-19 | 2019-09-17 | 中国电力科学研究院有限公司 | A kind of Methods of electric load forecasting and system based on combination mining model |
| US20200159207A1 (en) * | 2018-05-07 | 2020-05-21 | Strong Force Iot Portfolio 2016, Llc | Methods and systems for data collection and analysis of machine signals for analytics and maintenance using the industrial internet of things and a mobile data collector |
| CN111950805A (en) * | 2020-08-25 | 2020-11-17 | 润联软件系统(深圳)有限公司 | Medium-and-long-term power load prediction method and device, computer equipment and storage medium |
-
2020
- 2020-12-18 CN CN202011515128.3A patent/CN112508299A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180150783A1 (en) * | 2016-08-24 | 2018-05-31 | Clari Inc. | Method and system for predicting task completion of a time period based on task completion rates and data trend of prior time periods in view of attributes of tasks using machine learning models |
| US20200159207A1 (en) * | 2018-05-07 | 2020-05-21 | Strong Force Iot Portfolio 2016, Llc | Methods and systems for data collection and analysis of machine signals for analytics and maintenance using the industrial internet of things and a mobile data collector |
| CN110084424A (en) * | 2019-04-25 | 2019-08-02 | 国网浙江省电力有限公司 | A kind of Methods of electric load forecasting based on LSTM and LGBM |
| CN110245801A (en) * | 2019-06-19 | 2019-09-17 | 中国电力科学研究院有限公司 | A kind of Methods of electric load forecasting and system based on combination mining model |
| CN111950805A (en) * | 2020-08-25 | 2020-11-17 | 润联软件系统(深圳)有限公司 | Medium-and-long-term power load prediction method and device, computer equipment and storage medium |
Non-Patent Citations (3)
| Title |
|---|
| ZHENG J ETAL.: "Electric load forecasting in smart grids using Long-Short-Term-Memory based Recurrent Neural Network", 《2017 51ST ANNUAL CONFERENCE ON INFORMATION SCIENCES AND SYSTEMS (CISS)》 * |
| 杨帆 等: "时间序列趋势分析方法的一种改进", 《计算机技术与发展》 * |
| 罗澍忻 等: "考虑相关因素的长短时记忆网络短期负荷预测方法", 《机电工程技术》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113131476A (en) * | 2021-04-28 | 2021-07-16 | 南方电网科学研究院有限责任公司 | Power load prediction method |
| CN113139698A (en) * | 2021-05-13 | 2021-07-20 | 国网山东省电力公司枣庄供电公司 | Load prediction method, device and equipment |
| CN114118641A (en) * | 2022-01-29 | 2022-03-01 | 华控清交信息科技(北京)有限公司 | Wind power plant power prediction method, GBDT model longitudinal training method and device |
| CN114118641B (en) * | 2022-01-29 | 2022-04-19 | 华控清交信息科技(北京)有限公司 | Wind power plant power prediction method, GBDT model longitudinal training method and device |
| CN114841457A (en) * | 2022-05-18 | 2022-08-02 | 上海玫克生储能科技有限公司 | Power load estimation method and system, electronic device, and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112508299A (en) | Power load prediction method and device, terminal equipment and storage medium | |
| CN106022521B (en) | Short-term load prediction method of distributed BP neural network based on Hadoop architecture | |
| US20210133536A1 (en) | Load prediction method and apparatus based on neural network | |
| CN108304623B (en) | Probability load flow online calculation method based on stack noise reduction automatic encoder | |
| CN111860982A (en) | Wind power plant short-term wind power prediction method based on VMD-FCM-GRU | |
| CN114330935B (en) | New energy power prediction method and system based on multiple combination strategies integrated learning | |
| CN113887801A (en) | Building cold load prediction method, system, equipment and readable storage medium | |
| CN113344288A (en) | Method and device for predicting water level of cascade hydropower station group and computer readable storage medium | |
| CN115660233A (en) | Photovoltaic power prediction method and device, electronic equipment and storage medium | |
| CN116799796A (en) | Photovoltaic power generation power prediction method, device, equipment and medium | |
| CN111950752A (en) | Photovoltaic power station generating capacity prediction method, device and system and storage medium thereof | |
| CN114154753A (en) | Load prediction method and system | |
| Sun et al. | Short-term power load prediction based on VMD-SG-LSTM | |
| Sang et al. | Ensembles of gradient boosting recurrent neural network for time series data prediction | |
| CN111612648B (en) | Training method and device for photovoltaic power generation prediction model and computer equipment | |
| CN117543544A (en) | Load prediction method, device, equipment and storage medium | |
| CN113850438A (en) | Public building energy consumption prediction method, system, equipment and medium | |
| CN113095547A (en) | Short-term wind power prediction method based on GRA-LSTM-ICE model | |
| CN112287605A (en) | Flow check method based on graph convolution network acceleration | |
| CN111709585A (en) | Air conditioner load prediction method and device and storage medium | |
| CN115907131A (en) | Method and system for building electric heating load prediction model in northern area | |
| CN115689067A (en) | Solar irradiance prediction method, device and storage medium | |
| CN115545164A (en) | Photovoltaic power generation power prediction method, system, equipment and medium | |
| CN115619449A (en) | Deep learning model-based fertilizer price prediction method, device and storage medium | |
| CN115616333A (en) | Power distribution network line loss 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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210316 |
|
| RJ01 | Rejection of invention patent application after publication |