CN115545319A - Power grid short-term load prediction method based on meteorological similar day set - Google Patents

Power grid short-term load prediction method based on meteorological similar day set Download PDF

Info

Publication number
CN115545319A
CN115545319A CN202211257874.6A CN202211257874A CN115545319A CN 115545319 A CN115545319 A CN 115545319A CN 202211257874 A CN202211257874 A CN 202211257874A CN 115545319 A CN115545319 A CN 115545319A
Authority
CN
China
Prior art keywords
load
meteorological
data
similar
day
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202211257874.6A
Other languages
Chinese (zh)
Inventor
刘辉
凌宁青
谢海敏
汪旎
马斯宇
黄立冬
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangxi University
Original Assignee
Guangxi University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Guangxi University filed Critical Guangxi University
Priority to CN202211257874.6A priority Critical patent/CN115545319A/en
Publication of CN115545319A publication Critical patent/CN115545319A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

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

Abstract

The invention discloses a power grid short-term load prediction method based on a meteorological similar day set, which comprises the following steps of: 1. collecting historical load and meteorological data, carrying out interpolation processing on the meteorological data to obtain data with 15min resolution, and carrying out minimum and maximum normalization processing on the load and the meteorological data; 2. selecting meteorological factors strongly related to the load according to the Pearson coefficient and the maximum information coefficient; 3. selecting the optimal similar days to form a meteorological similar day set according to the meteorological factors; 4. taking weather similar day set load, historical load, weather factors and time factors as input data of a prediction model; 5. setting hyper-parameters of a prediction model, training the prediction model, and storing the optimal prediction model; 6. and inputting the test set to the optimal prediction model to obtain a prediction result, and performing inverse normalization on the prediction result to obtain a load prediction value. The method is beneficial to improving the load prediction precision by exploring the information contained in similar days of different meteorology.

Description

Power grid short-term load prediction method based on meteorological similar day set
Technical Field
The invention relates to a power grid short-term load forecasting method, in particular to a power grid short-term load forecasting method based on a meteorological similar daily set.
Background
With the scale access of uncertain elements such as distributed power sources and electric vehicles to the power grid, the load uncertainty is increased rapidly, and the safe and economic operation of the power grid faces more severe examination. The load prediction is an important component of an energy management system of a power grid dispatching center, and can provide key information for dispatching decision-making personnel, so that the safe and economic operation of a power grid is effectively guaranteed.
The power grid short-term load prediction is a key link of power grid safe and economic operation, and has important guiding significance for scheduling decision-makers to make reasonable and efficient scheduling plans. The similar day selection is a key step of short-term load prediction, the similar day selection actually selects an input feature set of a prediction model, and sufficiently excavates key information contained in the similar day, so that the method is an effective way for improving the short-term load prediction precision of the power grid.
The meteorological factors are important influence factors of load change, and the similarity of the loads can be better reflected by selecting similar days through the meteorological factors. In order to fully mine similar day information of different meteorological factors, a power grid short-term load prediction method based on a meteorological similar day set is provided.
Disclosure of Invention
The invention aims to further improve the accuracy of the short-term load prediction of the power grid.
In order to achieve the above purpose, the technical scheme provided by the invention is as follows: the invention discloses a power grid short-term load prediction method based on a meteorological similar day set, which comprises the following steps of:
step A: collecting power grid load and related data thereof, and preprocessing the data;
and B: analyzing the correlation between the meteorological factors and the load by adopting a Pearson coefficient and a maximum information coefficient, and selecting the meteorological factors which are strongly correlated with the load;
and C: selecting corresponding optimal similar days to form a meteorological similar day set according to meteorological factors strongly related to the load, and taking the meteorological similar day set load, the historical load, the meteorological factors and the time factors as input data of a prediction model;
step D: and constructing a TCN-LSTM prediction model for prediction, and outputting a load prediction result.
The calculation formula of the Pearson coefficient is as follows:
Figure BDA0003888181350000021
in the formula: r is the Pearson correlation coefficient between variables, n is the total number of samples,
Figure BDA0003888181350000022
is the average value of the variable X and,
Figure BDA0003888181350000023
is the average of the variable Y.
The maximum information coefficient is calculated according to the following formula:
Figure BDA0003888181350000024
Figure BDA0003888181350000025
in the formula: n is the number of data sets, the product of x and y needs to be less than B (n), and generally B (n) = n 0.6 。
The meteorological factors comprise temperature, humidity, rainfall and wind speed.
In the step A, collecting historical load and meteorological data, and preprocessing the data, wherein the preprocessing comprises the following steps:
step A1: collecting power grid load and relevant data thereof, wherein the relevant data comprises meteorological data and day type data;
step A2: processing abnormal values and missing values in the load data by adopting a filling method;
step A3: processing the meteorological data by adopting a linear interpolation mode, converting the meteorological data into data with the resolution of 15min, and matching the resolution of the load data;
step A4: and carrying out minimum and maximum normalization processing on the load and the meteorological data.
The filling method is calculated by the following formula:
Figure BDA0003888181350000031
in the formula: y' i Load value, y, restored at time i i-1 Is the load value at time i-1, y i+1 The load value at the time i + 1.
The minimum and maximum normalization processing of the load and meteorological data is obtained by the following formula:
Figure BDA0003888181350000032
in the formula: x is the number of n Is the normalized value, x is the value to be normalized, x max Is the maximum value, x, in the data min Is the minimum value in the data.
And in the step C, selecting the corresponding optimal similar days to form a meteorological similar day set according to meteorological factors strongly related to the load, and comprising the following steps of:
step C1: the daily load curve has a large difference between two types of days, namely a working day and a resting day; for dates with similar weather conditions in the same type of day, the load curve and the load size of the dates also have certain similarity; according to the correlation analysis result of the load and the meteorological factors, respectively selecting corresponding optimal similar days by the two meteorological factors of temperature and humidity to form a meteorological similar day set; when the temperature and the humidity select the corresponding optimal similar day as the same day, a second similar day of the temperature and the optimal similar day are taken to form a meteorological similar day set;
and step C2: selecting a weather similar day of the day to be predicted by adopting the form similar distance, wherein the form similar distance is defined as follows:
Figure BDA0003888181350000033
in the formula: l is i Real-time weather sequences, L, representing days to be predicted j Real-time weather sequences, representing historical days,/ ik Represents the sequence L i The kth element of (1), l jk Represents the sequence L j The kth element of (1); d Euclid Representing the Euclidean distance of the two sequences, ASD representing the absolute value of the sum of the numerical differences of the two sequences, and SAD representing the Manhattan distance of the two sequences. D (L) i ,L j ) Represents the morphologically similar distance of the two sequences;
and C3: selecting a weather similar day of the day to be predicted, considering the principle of 'big-end-up and small-end-up', namely that the predicted load point is more related to recent information in the historical time period, and setting the search range of the similar day as the first 10 days of the same type of the day to be predicted.
In the step D, a TCN-LSTM prediction model is constructed for prediction, and a load prediction result is output, wherein the method comprises the following steps:
step D1: the parameters of the TCN-LSTM prediction model are: the number of TCN layers is 1, the number of convolution kernels is 32, the size of the convolution kernels is 2, the expansion coefficient is 1, 2 and 4, the number of LSTM layers is 1, the number of neurons is 32, and the activation functions are all relu; the model optimizer selects Adam, the learning rate is set to be 0.001, the loss function is set to be mae, the iteration times are set to be 40 times, and the batch size is set to be 128; the rear of the TCN-LSTM is connected with 1 layer of full-connection layer, the number of neurons is 1, and the neuron represents the load predicted value at 1 moment;
step D2: and performing inverse normalization processing on the output result of the TCN-LSTM prediction model to obtain a load prediction value.
The denormalization is calculated by:
Figure BDA0003888181350000041
in the formula: x is the number of l Is the load predicted value after the reverse normalization,
Figure BDA0003888181350000042
in order to predict the result for the model,
Figure BDA0003888181350000043
is the maximum value of the original data and,
Figure BDA0003888181350000044
is the minimum of the raw data.
The beneficial effects obtained by the invention are as follows:
1. according to the power grid short-term load forecasting method based on the meteorological similarity day set, linear and nonlinear correlations of loads and meteorological factors are analyzed through Pearson coefficients and maximum information coefficients, and the meteorological factors strongly correlated with the loads are selected. And then selecting the corresponding optimal similar day to form a meteorological similar day set by adopting the morphological similar distance according to the meteorological factors, and taking the meteorological similar day set load, the historical load, the meteorological factors and the time factors as the input characteristics of the prediction model. Compared with other methods, the method considers the weather similar day set as the input of the prediction model, can fully mine the similar day information of different weather factors, and has better prediction effect.
2. According to the power grid short-term load prediction method based on the meteorological similarity day set, the TCN-LSTM short-term load prediction model is constructed by utilizing the feature extraction capability of TCN and the time sequence prediction capability of LSTM, compared with other models, the prediction model can fully mine feature information of input data, and has a better prediction effect.
Drawings
FIG. 1 is a flowchart of a power grid short-term load forecasting method based on a meteorological similar day set according to an embodiment of the invention;
FIG. 2 is a graph of the load profile for each best similar day;
FIG. 3 is a diagram of predicted results of different input feature methods.
Detailed Description
The following description will explain embodiments of the present invention in further detail with reference to the accompanying drawings.
The flow chart of the power grid short-term load prediction method based on the meteorological similar day set in the embodiment of the invention is shown in fig. 1, and the specific implementation steps are as follows:
step A: collecting power grid load and relevant data thereof, and preprocessing the data, wherein the method comprises the following specific steps:
step A1: collecting power grid load and relevant data thereof, wherein the relevant data comprises meteorological data and day type data;
step A2: processing abnormal values and missing values in the load data by a filling method, wherein the filling method has the following formula:
Figure BDA0003888181350000061
in the formula: y' i Load value, y, restored at time i i-1 Is the load value at time i-1, y i+1 The load value at the time i + 1. Particularly, if the data at a plurality of continuous time points are missing and abnormal, the data are filled in by adopting a linear interpolation mode.
Step A3: processing the meteorological data by adopting a linear interpolation mode, converting the meteorological data into data with the resolution of 15min, and matching the resolution of the load data;
step A4: carrying out minimum and maximum normalization processing on the load and meteorological data, wherein the formula is as follows:
Figure BDA0003888181350000062
in the formula: x is the number of n Is the normalized value, x is the value to be normalized, x max Is the maximum value, x, in the data min Is the minimum value in the data.
And B: analyzing the correlation between the meteorological factors and the load by adopting a Pearson coefficient and a maximum information coefficient, and selecting the meteorological factors which are strongly correlated with the load; the meteorological factors comprise temperature (centigrade), humidity (%), rainfall (mm/15 min) and wind speed (m/s); the specific process is as follows:
the Pearson coefficient is calculated as follows:
Figure BDA0003888181350000063
in the formula: r is the Pearson correlation coefficient between variables, n is the total number of samples,
Figure BDA0003888181350000064
is the average value of the variable X and,
Figure BDA0003888181350000065
is the average of the variable Y. When | r | is larger, the correlation is stronger. Generally speaking, when | r | > 0.6 represents strong correlation, | r |. Belongs to [0.4,0.6 ]]Indicating moderate correlation, | r | ∈ [0.2,0.4]Indicating a weak correlation, in | r | ∈ [0,0.2]Indicating very weak or no correlation.
And step B3: the maximum information coefficient is calculated as follows:
Figure BDA0003888181350000071
Figure BDA0003888181350000072
in the formula: n is the number of data sets, the product of x and y needs to be less than B (n), and generally B (n) = n 0.6 . The MIC metric generally reflects the correlation strength of variables by comparing the magnitude of MIC values, and if the MIC value is larger, the correlation strength of two variables is higher.
And step B4: linear and nonlinear correlations of each seasonal load with meteorological factors were analyzed using Pearson correlation coefficients and MICs, and the calculation results are shown in tables 1 and 2:
TABLE 1 Linear correlation analysis of load and meteorological factors
Figure BDA0003888181350000073
TABLE 2 nonlinear correlation analysis of load and meteorological factors
Figure BDA0003888181350000074
As can be seen from Table 1, in spring, the temperature and humidity | r | is between [0.4-0.6], indicating that there is a moderate correlation of load with temperature. In summer, the temperature and humidity | r | is greater than 0.6, indicating a strong correlation with load. In autumn, | r | of temperature is greater than 0.6, indicating that there is a strong correlation between load and temperature; the | r | of humidity is between [0.2-0.4], weakly related to load, but significantly greater than the | r | of rain and wind speed. In winter, the | r | of temperature and humidity is between [0.2-0.4], indicating a weak correlation with load, but significantly greater than the | r | of rainfall and wind speed, indicating a stronger correlation of load with temperature and humidity relative to rainfall and wind speed. In all seasons, the | r | of rainfall and wind speed are both between [0-0.2], indicating that there is little or no correlation with the load. The analysis results of all seasons are combined, and the load has stronger linear correlation with temperature and humidity relative to rainfall and wind speed.
As can be seen from table 2, the MICs of temperature and humidity are much greater than those of rainfall and wind speed in spring, summer and fall, indicating a stronger non-linear dependence of load on temperature and humidity relative to rainfall and wind speed. In winter, the MIC of temperature and humidity is smaller than the values in spring, summer and autumn, but still larger than the MIC of rainfall and wind speed, indicating that the nonlinear dependence of load on humidity is stronger with respect to rainfall and wind speed. And (3) combining the analysis results of all seasons, and the load has stronger non-linear correlation with temperature and humidity relative to rainfall and wind speed.
In summary, according to the analysis results of linear correlation and nonlinear correlation, the load has stronger correlation with temperature and humidity, so that the temperature and humidity are considered to select a weather-like day set.
Step C: selecting corresponding optimal similar days to form a meteorological similar day set according to meteorological factors strongly related to the load, and taking the meteorological similar day set load, the historical load, the meteorological factors and the time factors as input data of a prediction model; the specific process is as follows:
step C1: the daily load curve differs greatly between the two types of days, the working day and the resting day. For the same type of day and the date with similar meteorological conditions, the load curve and the load size have certain similarity. According to the correlation analysis result of the load and the meteorological factors, the corresponding optimal similar days are respectively selected by the two meteorological factors of the temperature and the humidity to form a meteorological similar day set. Particularly, if the optimal similar day corresponding to the temperature and the humidity is selected as the same day, the second similar day of the temperature and the optimal similar day are taken to form a meteorological similar day set.
And step C2: selecting a weather similar day of a day to be predicted by adopting a form similar distance, wherein the form similar distance is defined as follows:
Figure BDA0003888181350000091
in the formula: l is i Real-time weather sequence, L, representing the day to be predicted j Real-time weather sequences, representing historical days,/ ik Represents the sequence L i The kth element of (1), l jk Represents the sequence L j The kth element of (1). D Euclid Representing the Euclidean distance of the two sequences, ASD representing the absolute value of the sum of the numerical differences of the two sequences, and SAD representing the Manhattan distance of the two sequences. D (L) i ,L j ) Representing the morphologically similar distance of the two sequences.
And C3: selecting a weather similar day of the day to be predicted, considering the principle of 'big-end-up and small-end-up', namely that the predicted load point is more related to recent information in the historical time period, and setting the search range of the similar day as the first 10 days of the same type of the day to be predicted.
Step D: and constructing a TCN-LSTM prediction model for prediction, and outputting a load prediction result. The specific process is as follows:
step D1: the parameters of the TCN-LSTM prediction model are: the number of TCN layers is 1, the number of convolution kernels is 32, the size of the convolution kernels is 2, the expansion coefficient is 1, 2 and 4, the number of LSTM layers is 1, the number of neurons is 32, and the activation functions are all relu. Adam is selected as the model optimizer, the learning rate is set to 0.001, the loss function is set to mae, the iteration number is set to 40, and the batch size is set to 128. The TCN-LSTM is followed by 1 layer of full-connection layer, the number of neurons is 1, and the number represents the predicted value of load at 1 moment
Step D2: and performing inverse normalization processing on the output result of the TCN-LSTM prediction model to obtain a load prediction value. The denormalization formula is:
Figure BDA0003888181350000092
in the formula: x is the number of l Is the load predicted value after the reverse normalization,
Figure BDA0003888181350000093
in order to predict the result for the model,
Figure BDA0003888181350000094
is the maximum value of the original data and,
Figure BDA0003888181350000101
is the minimum of the raw data.
In a specific implementation process, the power grid short-term load prediction method based on the meteorological similar day set selects meteorological factors strongly related to loads through Pearson coefficients and maximum information coefficients. And then selecting the optimal similar days to form a meteorological similar day set according to the meteorological factors, and taking the meteorological similar day set load, the historical load, the meteorological factors and the time factors as input data of a prediction model. Compared with other methods, the method considers the weather similar day set as the input of the prediction model, can fully mine the similar day information of different weather factors, and has better prediction effect. The input characteristics of the method presented herein are shown in table 3.
TABLE 3 input characteristics of the methods mentioned in this chapter
Figure BDA0003888181350000102
Example 2
More specifically, on the basis of example 1, to verify the scientificity and reliability of the method provided herein, actual data from 1/31/2010 to 12/31/2012 in a certain area of china are used for verification, the sampling interval of the data is 15min, that is, 96 points are sampled every day, and finally, the experimental result is compared with the predicted result of the input features of the method 1 and the method 2. The input characteristics of method 1 and method 2 are shown in table 4.
Table 4 input characterization of method 1 and method 2
Figure BDA0003888181350000103
Figure BDA0003888181350000111
In a specific implementation, the computer herein is configured to: CPU Core i7-10700, 16GB memory, windows 10 operating system, deep learning framework based on Python3.7 programming language and programming. Specific analysis is carried out on the day of 8-1/2012, corresponding optimal similar days are selected by respectively adopting a real-time temperature sequence, a real-time humidity sequence and a mixed weather sequence, the corresponding dates are respectively 20/2012/7/19/2012/7/31/2012, the load curve of the best similar day is shown in fig. 2, and it can be seen from the graph that the load curve of day 7/20 in 2012 is closest to the load curve of day 8/1 in 2012, day 7/19 in 2012, and the load curve of day 31 in 2012, day 7/31 in 2012 to be predicted. Further, the distance value between each optimal similar daily load sequence and the daily load sequence to be predicted is calculated through the morphological similarity distance to evaluate the selection result, and the result is shown in table 5.
TABLE 5 best similarity days comparison
Figure BDA0003888181350000112
Figure BDA0003888181350000121
As can be seen from table 5, the morphological similarity distance between the best similar day selected by the temperature sequence and the day to be predicted is 0.366, the morphological similarity distance between the best similar day selected by the humidity sequence and the day to be predicted is 0.459, and the morphological similarity distance between the best similar day selected by the mixed weather sequence and the day to be predicted is 0.613. Compared with the mixed weather sequence, the shape similarity distance of the optimal similar day selected by the temperature sequence and the humidity sequence is smaller, and the selected similar day is more similar, which shows that compared with the mixed weather sequence, the single weather sequence is used for selecting the similar day to avoid the influence of other weather, so that a better similar day result is obtained.
The load of the day of 2012, 8/1/day is predicted, and the verification set is 20% of the training set from the day of 2010, 1/31/day to the day before 2012, 8/1/day. The prediction result is shown in fig. 3, and it can be seen from the figure that the fitting degree of the prediction curve and the actual curve of the invention is better than that of the method 1 and the method 2, which shows that the prediction effect of the invention is more ideal. Further MAPE and MAE indices were calculated and the results are shown in table 6.
TABLE 6 prediction results of different methods
Figure BDA0003888181350000122
As can be seen from Table 6, the MAPE and MAE of method 1 were 3.77% and 319.54MW, the MAPE and MAE of method 2 were 2.67% and 232.92MW, and the MAPE and MAE of the present invention were 2.26% and 194.33MW. Compared with the method 1, the MAPE and MAE of the method 2 and the invention are smaller, which shows that the load prediction precision can be improved by considering the meteorological similar day characteristics in the load prediction. Compared with the method 2, the MAPE and MAE of the invention are reduced by 0.41 percent and 38.59MW, which shows that the influence of meteorological similar day sets is considered to be more advantageous in load prediction, and the load prediction precision can be further improved.
To further verify the universality of the invention, the load prediction before day is carried out in 2012 all year round, the training set is 20% of the training set from 1 month 31 in 2010 to 12 months 31 in 2011. The predicted results are shown in table 7.
TABLE 7 prediction results of different methods throughout the year 2012
Figure BDA0003888181350000131
As can be seen from Table 7, the MAPE and MAE of method 1 were 3.83% and 260.39MW, respectively, the MAPE and MAE of method 2 were 3.65% and 245.56MW, respectively, and the MAPE and MAE of the present invention were 3.43% and 235.81MW, respectively. Compared with the method 1 and the method 2, the MAPE and the MAE of the method are minimum, which shows that the information of different meteorological similar days can be fully mined by considering the meteorological similar day set as the input characteristic in the load prediction, the influence of the different meteorological similar days on the load prediction can be better reflected, and the load prediction precision can be favorably improved.

Claims (10)

1. A power grid short-term load prediction method based on a meteorological similar day set is characterized by comprising the following steps:
step A: collecting power grid load and related data thereof, and preprocessing the data;
and B: analyzing the correlation between the meteorological factors and the load by adopting a Pearson coefficient and a maximum information coefficient, and selecting the meteorological factors which are strongly correlated with the load;
step C: selecting corresponding optimal similar days to form a meteorological similar day set according to meteorological factors strongly related to the load, and taking the meteorological similar day set load, the historical load, the meteorological factors and the time factors as input data of a prediction model;
step D: and constructing a TCN-LSTM prediction model for prediction, and outputting a load prediction result.
2. The method for predicting the short-term load of the power grid based on the meteorological similar day set as claimed in claim 1, wherein a calculation formula of the Pearson coefficient is as follows:
Figure FDA0003888181340000011
in the formula: r is the Pearson correlation coefficient between variables, n is the total number of samples,
Figure FDA0003888181340000012
is the average value of the variable X and,
Figure FDA0003888181340000013
is the average of the variable Y.
3. The method for forecasting the short-term load of the power grid based on the meteorological similar day set, according to claim 1, wherein the maximum information coefficient is calculated according to the following formula:
Figure FDA0003888181340000014
Figure FDA0003888181340000015
in the formula: n is the number of data sets, the product of x and y needs to be less than B (n), and generally B (n) = n 0.6 。
4. The method of claim 1, wherein the meteorological factors include temperature, humidity, rainfall, wind speed.
5. The method for forecasting the short-term load of the power grid based on the meteorological similar day set as claimed in claim 1, wherein in the step A, historical load and meteorological data are collected, and the data are preprocessed, and the method comprises the following steps:
step A1: collecting power grid load and relevant data thereof, wherein the relevant data comprises meteorological data and day type data;
step A2: processing abnormal values and missing values in the load data by adopting a filling method;
step A3: processing the meteorological data by adopting a linear interpolation mode, converting the meteorological data into data with the resolution of 15min, and matching the resolution of the load data;
step A4: and carrying out minimum and maximum normalization processing on the load and the meteorological data.
6. The method for forecasting the short-term load of the power grid based on the meteorological similar day set as claimed in claim 5, wherein the supplementing method is calculated by the following formula:
Figure FDA0003888181340000021
in the formula: y' i Load value, y, restored at time i i-1 Is the load value at time i-1, y i+1 The load value at the time i + 1.
7. The method for forecasting the short-term load of the power grid based on the similar weather day set as claimed in claim 5, wherein the minimum and maximum normalization processing of the load and the weather data is obtained by the following formula:
Figure FDA0003888181340000022
in the formula: x is the number of n Is the normalized value, x is the value to be normalized, x max Is the maximum value, x, in the data min Is the minimum value in the data.
8. The method for forecasting the short-term load of the power grid based on the meteorological similar day set as claimed in claim 1, wherein in the step C, the meteorological similar day set is selected to be composed of the corresponding best similar days according to meteorological factors strongly related to the load, and the method comprises the following steps:
step C1: the daily load curve has a large difference between two types of days, namely a working day and a resting day; for dates with similar weather conditions in the same type of day, the load curve and the load size of the dates also have certain similarity; according to the correlation analysis result of the load and the meteorological factors, respectively selecting corresponding optimal similar days by the two meteorological factors of temperature and humidity to form a meteorological similar day set; when the temperature and the humidity select the corresponding optimal similar day as the same day, a second similar day of the temperature and the optimal similar day are taken to form a meteorological similar day set;
and step C2: selecting a weather similar day of a day to be predicted by adopting a form similar distance, wherein the form similar distance is defined as follows:
Figure FDA0003888181340000031
in the formula: l is i Real-time weather sequences, L, representing days to be predicted j Real-time weather sequences, representing historical days,/ ik Represents the sequence L i The kth element of (1), l jk Represents the sequence L j The kth element of (1); d Euclid Representing the Euclidean distance of two sequences, ASD representing the absolute value of the sum of the numerical differences of the two sequences, SAD representing the Manhattan distance of the two sequences, D (L) i ,L j ) Represents the morphological similarity distance of the two sequences;
and C3: selecting a weather similar day of the day to be predicted, considering the principle of 'big-end-up and small-end-up', namely that the predicted load point is more related to recent information in the historical time period, and setting the search range of the similar day as the first 10 days of the same type of the day to be predicted.
9. The method for predicting the short-term load of the power grid based on the meteorological similar day set as claimed in claim 1, wherein in the step D, a TCN-LSTM prediction model is constructed for prediction, and a load prediction result is output, and the method comprises the following steps:
step D1: the parameters of the TCN-LSTM prediction model are: the number of TCN layers is 1, the number of convolution kernels is 32, the size of the convolution kernels is 2, the expansion coefficient is 1, 2 and 4, the number of LSTM layers is 1, the number of neurons is 32, and the activation functions are all relu; the model optimizer selects Adam, the learning rate is set to be 0.001, the loss function is set to be mae, the iteration times are set to be 40 times, and the batch size is set to be 128; the rear of the TCN-LSTM is connected with 1 layer of full-connection layer, the number of neurons is 1, and the neuron represents the load predicted value at 1 moment;
step D2: and performing inverse normalization processing on the output result of the TCN-LSTM prediction model to obtain a load prediction value.
10. The method of claim 9, wherein the inverse normalization is calculated by:
Figure FDA0003888181340000041
in the formula: x is the number of l Is the load predicted value after the reverse normalization,
Figure FDA0003888181340000042
in order to predict the result for the model,
Figure FDA0003888181340000043
is the maximum value of the original data and,
Figure FDA0003888181340000044
is the minimum of the raw data.
CN202211257874.6A 2022-10-13 2022-10-13 Power grid short-term load prediction method based on meteorological similar day set Pending CN115545319A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211257874.6A CN115545319A (en) 2022-10-13 2022-10-13 Power grid short-term load prediction method based on meteorological similar day set

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211257874.6A CN115545319A (en) 2022-10-13 2022-10-13 Power grid short-term load prediction method based on meteorological similar day set

Publications (1)

Publication Number Publication Date
CN115545319A true CN115545319A (en) 2022-12-30

Family

ID=84733655

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211257874.6A Pending CN115545319A (en) 2022-10-13 2022-10-13 Power grid short-term load prediction method based on meteorological similar day set

Country Status (1)

Country Link
CN (1) CN115545319A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116562398A (en) * 2023-07-11 2023-08-08 北京东润环能科技股份有限公司 Power load prediction model training method, prediction method, electronic device and medium

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116562398A (en) * 2023-07-11 2023-08-08 北京东润环能科技股份有限公司 Power load prediction model training method, prediction method, electronic device and medium
CN116562398B (en) * 2023-07-11 2023-09-08 北京东润环能科技股份有限公司 Power load prediction model training method, prediction method, electronic device and medium

Similar Documents

Publication Publication Date Title
CN113962364B (en) Multi-factor power load prediction method based on deep learning
CN112949945B (en) Wind power ultra-short-term prediction method for improving bidirectional long-term and short-term memory network
CN112016734A (en) Stack type self-coding multi-model load prediction method and system based on LSTM
CN111260136A (en) Building short-term load prediction method based on ARIMA-LSTM combined model
CN112712209B (en) Reservoir warehousing flow prediction method and device, computer equipment and storage medium
CN113205226A (en) Photovoltaic power prediction method combining attention mechanism and error correction
CN111985719B (en) Power load prediction method based on improved long-term and short-term memory network
CN114792156A (en) Photovoltaic output power prediction method and system based on curve characteristic index clustering
CN113988477A (en) Photovoltaic power short-term prediction method and device based on machine learning and storage medium
CN111242355A (en) Photovoltaic probability prediction method and system based on Bayesian neural network
CN113822418A (en) Wind power plant power prediction method, system, device and storage medium
CN115545319A (en) Power grid short-term load prediction method based on meteorological similar day set
CN115238854A (en) Short-term load prediction method based on TCN-LSTM-AM
CN115409258A (en) Hybrid deep learning short-term irradiance prediction method
CN115456306A (en) Bus load prediction method, system, equipment and storage medium
CN116435998A (en) Prediction method of photovoltaic power generation power
CN113610328A (en) Power generation load prediction method
CN113627091A (en) Device and method for predicting energy load
CN116885699A (en) Power load prediction method based on dual-attention mechanism
CN116995670A (en) Photovoltaic power ultra-short-term prediction method based on multi-mode decomposition and multi-branch input
CN116979503A (en) Commercial building load prediction method and system based on Informir network
CN116151464A (en) Photovoltaic power generation power prediction method, system and storable medium
CN115759343A (en) E-LSTM-based user electric quantity prediction method and device
CN114399101A (en) TCN-BIGRU-based gas load prediction method and device
Xia et al. Research on Solar Radiation Estimation based on Singular Spectrum Analysis-Deep Belief Network

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