CN115660228B - Power generation load prediction model training method, prediction method, device and storage medium - Google Patents

Power generation load prediction model training method, prediction method, device and storage medium Download PDF

Info

Publication number
CN115660228B
CN115660228B CN202211598178.1A CN202211598178A CN115660228B CN 115660228 B CN115660228 B CN 115660228B CN 202211598178 A CN202211598178 A CN 202211598178A CN 115660228 B CN115660228 B CN 115660228B
Authority
CN
China
Prior art keywords
power generation
month
predicted
data
prediction
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202211598178.1A
Other languages
Chinese (zh)
Other versions
CN115660228A (en
Inventor
伍歆
毛苗
马骏
王向阳
熊厚辉
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.)
Hunan Energy Big Data Center Co ltd
Original Assignee
Hunan Energy Big Data Center Co ltd
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 Hunan Energy Big Data Center Co ltd filed Critical Hunan Energy Big Data Center Co ltd
Priority to CN202211598178.1A priority Critical patent/CN115660228B/en
Publication of CN115660228A publication Critical patent/CN115660228A/en
Application granted granted Critical
Publication of CN115660228B publication Critical patent/CN115660228B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • 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

  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a power generation load prediction model training method, a prediction method, equipment and a storage medium, wherein the training method comprises the steps of obtaining various index data and power generation environment data of a power generation energy source, and forming an original data sample by the index data and the power generation environment data; carrying out maximum power generation load marking on the original data sample; normalizing the original data sample; training the constructed multiple prediction models by using a sample data set to obtain output quantity, and carrying out weighted average on the output quantity to obtain a prediction output quantity; processing the predicted output quantity by using the constructed seasonal index model to obtain a predicted value; and calculating an evaluation index according to the predicted value and the maximum power generation load, and evaluating the training precision of the prediction model by using the evaluation index. The invention can improve the training precision of the model and the prediction precision.

Description

Power generation load prediction model training method, prediction method, device and storage medium
Technical Field
The invention belongs to the technical field of power generation load prediction, and particularly relates to a power generation load prediction model training method, a power generation load prediction method, power generation load prediction equipment and a storage medium.
Background
The conversion of energy systems into more reliable, eco-friendly and cost-effective systems is a central goal of current energy policies. The power generation load prediction is to predict the maximum power generation load in a certain time period by a mathematical modeling method, and the power generation load prediction directly influences the related decisions of real-time power grid operation and long-term infrastructure expansion planning.
At present, the power generation load prediction is focused on analysis and prediction of a power utilization side, and the prediction of the power generation side basically aims at a single power generation technology, such as the power generation load prediction focusing on predicting wind power or photovoltaic, and can be seen from patent literature with application publication number of CN104376368A, namely a wind power generation short-term load prediction method and device based on frequency domain decomposition.
How to improve the applicability of the power generation load prediction and the prediction effect are the problems to be solved by those skilled in the art.
Disclosure of Invention
Aiming at the defects of the prior art, the invention aims to provide a training method, a prediction method, equipment and a storage medium for a prediction model of the maximum power generation load of a region, so as to solve the problems that the existing prediction method of the power generation load is only applicable to a single power generation technology and has poor prediction effect.
The invention solves the technical problems by the following technical scheme: a power generation load prediction model training method comprises the following steps:
acquiring various index data of electric energy sources at different historical time points and power generation environment data corresponding to the historical time points, wherein the index data and the power generation environment data corresponding to a single historical time point form an original data sample;
labeling the maximum power generation load of each original data sample, and taking the labeled maximum power generation load as a target value of the original data sample;
carrying out normalization processing on each original data sample to obtain normalized data samples; forming a sample data set by normalized data samples at different historical time points and corresponding target values;
constructing at least one prediction model and constructing a seasonal index model;
training each prediction model by using the sample data set to obtain the output quantity of the prediction model, and carrying out weighted average on the output quantity of all the prediction models to obtain the prediction output quantity;
processing the predicted output quantity by using the seasonal index model to obtain a predicted value;
calculating an evaluation index according to the predicted value and the corresponding target value, and outputting a trained predicted model when the evaluation index is smaller than a preset value; and when the evaluation index is greater than or equal to a preset value, repeating the steps of prediction model training, prediction output quantity processing and calculation and judgment of the evaluation index until the evaluation index is smaller than the preset value.
Further, the index data of the power generation energy source comprises the capacity of a general assembly machine and the actual power generation amount of different power generation modes, the theoretical maximum power generation amount, the actual power generation total amount, the total coal storage amount, the electric coal consumption amount, the local coal price, the national coal price, the coal adjustment amount, the total reservoir capacity, the average percentage of the reservoir water level and the total outsourcing electric quantity;
the power generation environment data comprise local average temperature, local average humidity, local average precipitation, local cloud cover, local average wind speed and local average irradiation intensity.
Further, the specific formula for carrying out normalization processing on each original data sample is as follows:
Figure SMS_1
wherein ,
Figure SMS_2
is the first of the original data samplesiThe data of the plurality of data,
Figure SMS_3
for data at different historical points in time
Figure SMS_4
Is used for the average value of (a),
Figure SMS_5
for data at different historical points in time
Figure SMS_6
Is a function of the variance of (a),
Figure SMS_7
is data of
Figure SMS_8
Is performed according to the normalization processing result of the (a).
Further, the number of the prediction models is 3, and the 3 prediction models are a long and short memory neural network model, a decision tree model and a support vector machine model respectively.
Further, the calculation formula of the predicted output quantity is as follows:
Figure SMS_9
wherein ,
Figure SMS_10
as a predicted output at a historical point in time,
Figure SMS_11
is the firstqThe weight value of the individual prediction model is,
Figure SMS_12
is the firstqThe output of the individual predictive models is determined,Qis the number of predictive models.
Further, the specific implementation process of processing the predicted output quantity by using the seasonal index model is as follows:
forming a single historical time series data set from the predicted output quantities of a plurality of historical time points, wherein the historical time series data set is recorded as:
Figure SMS_13
wherein ,
Figure SMS_14
as the history time series data set, the history time series data set
Figure SMS_15
Is composed ofmA set of predicted output for all days of the month,
Figure SMS_16
is the firstmMonth of the first monthnThe predicted output of the day is calculated,Nis the firstmDays of the month;
introducing a seasonal fluctuation factor into the historical time sequence data set to obtain a new time sequence data set, wherein the new time sequence data set comprises the following concrete steps:
Figure SMS_17
wherein ,
Figure SMS_18
for a new set of time-series data,
Figure SMS_19
to introduce seasonal wave factor aftermMonth of the first monthnThe predicted output of the day is calculated,
Figure SMS_20
as a factor of the fluctuation of the season,
Figure SMS_21
，
Figure SMS_22
is the firstmThe average power generation load for one month,
Figure SMS_23
is the firstmAverage annual power generation load of the year of the month;
regrouping the new time sequence data sets according to the same day of different months to obtain:
Figure SMS_24
wherein ,
Figure SMS_25
to reorganize a time series data set, the time series data set is reorganized
Figure SMS_26
Is formed by the first month of all monthsnThe set of predicted output of the day,Mis the month number;
for the reorganized time series data set
Figure SMS_27
Performing fractional order cumulative calculation to obtainrThe order cumulative sequence is:
Figure SMS_28
Figure SMS_29
wherein ,
Figure SMS_30
，
Figure SMS_31
，
Figure SMS_32
，
Figure SMS_33
for reorganising time-series data setsrA step accumulation sequence;
according torOrder cumulative sequence
Figure SMS_34
Calculating temporary intermediate variables
Figure SMS_35
And
Figure SMS_36
the method specifically comprises the following steps:
Figure SMS_37
Figure SMS_38
wherein ,
Figure SMS_39
as a result of the averaging sequence,
Figure SMS_40
，BandYis a temporary intermediate variable matrix;
to temporary intermediate variables
Figure SMS_41
And
Figure SMS_42
inserting a time sequence response function to obtain:
Figure SMS_43
wherein ,
Figure SMS_44
is the firstM+1 monthnPredictive value of day;
when (when)m＝M+1, the predicted value sequence is:
Figure SMS_45
performing dimension reduction operation on the predicted value sequence to obtain a dimension reduced predicted value sequence as follows:
Figure SMS_46
wherein ,
Figure SMS_47
and the predicted value sequence after dimension reduction is expressed as:
Figure SMS_48
predicted value sequence after dimension reduction according to all days in month
Figure SMS_49
The regrouping is:
Figure SMS_50
Figure SMS_51
Figure SMS_52
Figure SMS_53
Figure SMS_54
wherein ,
Figure SMS_55
is the predicted firstdTime series data set of month
Figure SMS_56
Is composed ofdA set of predictions for all days of the month,
Figure SMS_57
is the predicted firstdMonth of the first monthnPredictive value of day.
Further, the evaluation index is an average absolute percentage error, and a specific calculation formula of the average absolute percentage error is as follows:
Figure SMS_58
wherein MAPE is the mean absolute percentage error,
Figure SMS_59
in order to be able to predict the value,
Figure SMS_60
for the target value, N is the number of days of the predicted month.
Based on the same inventive concept, the invention also provides a power generation load prediction method, which comprises the following steps:
acquiring various index data of an electric energy source at a certain time point and generating environment data corresponding to the time point;
the index data and the power generation environment data are input into each prediction model after normalization processing, and the prediction output quantity of each prediction model is obtained;
carrying out weighted average on the predicted output quantity of each predicted model to obtain a final predicted value;
each of the prediction models is a model trained by the power generation load prediction model training method described above.
Based on the same inventive concept, an embodiment of the present invention provides an electronic device, including: a processor and a memory storing a computer program, the processor being configured to perform any of the power generation load prediction model training methods of the present invention or to perform the power generation load prediction methods of the present invention when the computer program is run.
Based on the same inventive concept, an embodiment of the present invention provides a computer readable storage medium having a computer program stored thereon, wherein the program when executed by a processor implements any one of the power generation load prediction model training methods of the present invention, or performs the power generation load prediction method of the present invention.
Advantageous effects
Compared with the prior art, the invention provides a power generation load prediction model training method, a prediction method, a device and a storage medium, wherein a plurality of prediction models are trained by utilizing a sample data set, and the output quantity of the plurality of prediction models is weighted and averaged to obtain a final prediction value, so that the prediction precision is greatly improved; meanwhile, a seasonal index model is introduced in the training process of the prediction model to process the predicted output quantity, so that the prediction precision is further improved; the power generation energy source can comprise a plurality of power generation modes, so that the power generation load prediction model can adapt to a plurality of power generation technologies and has a good prediction effect.
The maximum power generation load has more influence factors, the influence factors are also changed due to the change of economic and social environments, the interference of the influence factors on the training of the prediction model is greatly reduced through the introduction of the seasonal index model, the prediction precision of the model is improved, and the invention provides a flexible training method of the maximum power generation load prediction model.
Drawings
In order to more clearly illustrate the technical solutions of the present invention, the drawings that are needed in the description of the embodiments will be briefly introduced, it being obvious that the drawings in the following description are only one embodiment of the present invention, and that other drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flowchart of a power generation load prediction model training method in an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made more apparent and fully by reference to the accompanying drawings, in which it is shown, however, only some, but not all embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
As shown in fig. 1, the method for training the power generation load prediction model provided in this embodiment includes the following steps:
step 1: acquiring raw data
And acquiring various index data of the electric energy sources at different historical time points and the power generation environment data corresponding to the historical time points, wherein the index data and the power generation environment data corresponding to the single historical time point form an original data sample.
For example, various index data and power generation environment data of the power generation energy from 1 st 2018 to 12 nd 30 th 2018 are obtained, and the day is taken as a time point. The power generation energy sources of the embodiment include thermal power, hydroelectric power, wind power, photovoltaic power generation and other power generation, and various index data and power generation environment data of a certain historical time point (i.e. a certain day) are shown in table 1.
Table 1 various index data and power generation environment data
Figure SMS_61
In Table 1, 26 data are taken as features except for the date, and these features are all involved in training of the predictive model, and each feature or data is given as
Figure SMS_62
These 26 data or features constitute a raw data sample, noted as
Figure SMS_63
Represent the firstnDay original data samples.
Step 2: labeling the original data
Labeling the maximum power generation load of each original data sample to obtain the maximum power generation which can be born on the labeled dayLoad as the raw data sample
Figure SMS_64
Target value of (2)
Figure SMS_65
。
Step 3: normalization processing
For each of the raw data samples
Figure SMS_66
Normalization processing is carried out to obtain normalized data samples
Figure SMS_67
. The specific calculation formula of the normalization process is as follows:
Figure SMS_68
（1）
wherein ,
Figure SMS_71
for the original data sample
Figure SMS_73
The first of (3)iThe data of the plurality of data,
Figure SMS_75
for all data
Figure SMS_70
Is used for the average value of (a),
Figure SMS_72
for all data
Figure SMS_74
Is a function of the variance of (a),
Figure SMS_76
is data of
Figure SMS_69
Is performed according to the normalization processing result of the (a).
Step 4: construction of training sample data sets
Normalized data samples from different historical time points
Figure SMS_77
Corresponding target value
Figure SMS_78
Constructing a sample data set, dividing the sample data set intoQThe weight of the components is calculated by the weight,Qfor the number of predictive models, each sample dataset is used to train one predictive model.
Step 4: at least one predictive model is constructed, and a seasonal index model is constructed.
In this embodiment, the number of prediction models is 3, i.eQThe =3, 3 prediction models are long and short memory neural network model, decision tree model and support vector machine model, respectively.
Step 5: training of predictive models
And training each prediction model by using the sample data set to obtain the output quantity of the prediction model, and carrying out weighted average on the output quantity of all the prediction models to obtain the prediction output quantity.
The long and short memory neural network model, the decision tree model and the support vector machine model are all existing models, and the training process of the models is also the prior art. The calculation formula of the predicted output quantity is as follows:
Figure SMS_79
（2）
wherein ,
Figure SMS_80
as a predicted output at a historical point in time,
Figure SMS_81
is the firstqThe weight value of the individual prediction model is,
Figure SMS_82
is the firstqThe output of the individual predictive models is determined,Qis a pre-preparationThe number of models was measured.
Step 6: processing of predicted output
And processing the predicted output quantity by using the seasonal index model, wherein the specific implementation process is as follows:
step 6.1: forming a single historical time series data set from the predicted output quantities of a plurality of historical time points, wherein the historical time series data set is recorded as:
Figure SMS_83
Figure SMS_84
Figure SMS_85
Figure SMS_86
…
Figure SMS_87
wherein ,
Figure SMS_88
as the history time series data set, the history time series data set
Figure SMS_89
Is composed ofmA set of predicted output for all days of the month,
Figure SMS_90
is the firstmMonth of the first monthnThe predicted output of the day is calculated,Nis the firstmThe number of days of the month,Mis the number of months.
Step 6.2: introducing a seasonal fluctuation factor into the historical time sequence data set to obtain a new time sequence data set, wherein the new time sequence data set comprises the following concrete steps:
Figure SMS_91
Figure SMS_92
Figure SMS_93
Figure SMS_94
…
Figure SMS_95
wherein ,
Figure SMS_96
for a new set of time-series data,
Figure SMS_97
to the current year after the introduction of the seasonal wave factormMonth of the first monthnThe predicted output of the day is calculated,
Figure SMS_98
as a factor of the fluctuation of the season,
Figure SMS_99
，
Figure SMS_100
is the current yearmThe average power generation load for one month,
Figure SMS_101
is the firstmThe annual average power generation load of the year of the month.
Step 6.3: regrouping the new time sequence data sets according to the same day of different months to obtain:
Figure SMS_102
Figure SMS_103
Figure SMS_104
Figure SMS_105
…
Figure SMS_106
wherein ,
Figure SMS_107
to reorganize a time series data set, the time series data set is reorganized
Figure SMS_108
Is the first month of the current yearnThe set of predicted output of the day,Mis the number of months.
Step 6.4: for the reorganized time series data set
Figure SMS_109
Fractional order accumulation (Fractional order accumulation) calculation is performed to obtainrThe order cumulative sequence is:
Figure SMS_110
Figure SMS_111
…
Figure SMS_112
…
Figure SMS_113
Figure SMS_114
wherein ,
Figure SMS_115
，
Figure SMS_116
，
Figure SMS_117
for reorganising time-series data setsrOrder cumulative sequence.
Step 6.5: according torOrder cumulative sequence
Figure SMS_118
Calculating temporary intermediate variables
Figure SMS_119
And
Figure SMS_120
the method specifically comprises the following steps:
Figure SMS_121
Figure SMS_122
wherein ,
Figure SMS_123
as a result of the averaging sequence,
Figure SMS_124
，BandYis a temporary intermediate variable matrix.
Step 6.6: to temporary intermediate variables
Figure SMS_125
And
Figure SMS_126
inserting a time sequence response function to obtain:
Figure SMS_127
wherein ,
Figure SMS_128
is the firstMPredicted value for day +1 month;
when (when)m＝M+1, the predicted value sequence is:
Figure SMS_129
。
step 6.7: performing dimension reduction operation on the predicted value sequence to obtain a dimension reduced predicted value sequence as follows:
Figure SMS_130
wherein ,
Figure SMS_131
and the predicted value sequence after dimension reduction is expressed as:
Figure SMS_132
step 6.8: predicted value sequence after dimension reduction according to all days in month
Figure SMS_133
The regrouping is:
Figure SMS_134
Figure SMS_135
Figure SMS_136
Figure SMS_137
Figure SMS_138
wherein ,
Figure SMS_139
is the predicted firstdTime series data set of month
Figure SMS_140
Is composed ofdA set of predictions for all days of the month,
Figure SMS_141
is the predicted firstdMonth of the first monthnPredictive value of day.
Step 7: prediction model precision evaluation
Calculating an evaluation index according to the predicted value and the corresponding target value, and outputting a trained predicted model when the evaluation index is smaller than a preset value; and when the evaluation index is greater than or equal to a preset value, repeating the steps 5-7 until the evaluation index is smaller than the preset value.
And (3) processing the predicted output quantity by using the seasonal index model and evaluating the model precision once by using each pair of the prediction models in the sample data set, and outputting the trained prediction models if a plurality of continuous evaluation indexes are smaller than a preset value. Or after training the prediction model by using all samples in the sample data set, processing the prediction output quantity by using the seasonal index model, evaluating the model precision, and outputting the trained prediction model when the evaluation index is smaller than a preset value.
In this embodiment, the evaluation index is an average absolute percentage error, and a specific calculation formula of the average absolute percentage error is:
Figure SMS_142
wherein MAPE is the mean absolute percentage error,
Figure SMS_143
in order to be able to predict the value,
Figure SMS_144
for the target value, N is the number of days of the predicted month. In this embodiment, the set value is 10%.
Based on the same inventive concept, the embodiment of the invention also provides a power generation load prediction method, which comprises the following steps:
acquiring various index data of an electric energy source at a certain time point and generating environment data corresponding to the time point;
the index data and the power generation environment data are input into each prediction model after normalization processing, and the prediction output quantity of each prediction model is obtained;
carrying out weighted average on the predicted output quantity of each predicted model to obtain a final predicted value;
each of the prediction models is a model trained by the power generation load prediction model training method described above.
The foregoing is merely illustrative of the present invention, and the present invention is not limited thereto, and any person skilled in the art will readily recognize that variations or substitutions are within the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims (8)

1. The power generation load prediction model training method is characterized by comprising the following steps of:
acquiring various index data of electric energy sources at different historical time points and power generation environment data corresponding to the historical time points, wherein the index data and the power generation environment data corresponding to a single historical time point form an original data sample; the index data of the power generation energy comprises the capacity of a general loader and the actual power generation amount of different power generation modes, the theoretical maximum power generation amount, the actual power generation total amount, the total coal storage amount, the electric coal consumption amount, the local coal price, the national coal price, the coal adjustment amount, the total reservoir capacity, the average percentage of reservoir water level and the total outsourcing electric quantity; the power generation environment data comprise local average temperature, local average humidity, local average precipitation, local cloud cover, local average wind speed and local average irradiation intensity;
labeling the maximum power generation load of each original data sample, and taking the labeled maximum power generation load as a target value of the original data sample;
carrying out normalization processing on each original data sample to obtain normalized data samples; forming a sample data set by normalized data samples at different historical time points and corresponding target values;
constructing N prediction models and constructing a seasonal index model, wherein N is more than 1;
training each prediction model by using the sample data set to obtain the output quantity of the prediction model, and carrying out weighted average on the output quantity of all the prediction models to obtain the prediction output quantity;
processing the predicted output quantity by using the seasonal index model to obtain a predicted value;
calculating an evaluation index according to the predicted value and the corresponding target value, and outputting a trained predicted model when the evaluation index is smaller than a preset value; when the evaluation index is greater than or equal to a preset value, repeating the steps of prediction model training, prediction output quantity processing and calculation and judgment of the evaluation index until the evaluation index is smaller than the preset value;
the specific implementation process for processing the predicted output quantity by using the seasonal index model is as follows:
forming a single historical time series data set from the predicted output quantities of a plurality of historical time points, wherein the historical time series data set is recorded as:
Figure QLYQS_1
；
wherein ,
Figure QLYQS_2
for the history time series data set +.>
Figure QLYQS_3
Is composed ofmA set of predicted output of all days of the month,/->
Figure QLYQS_4
Is the firstmMonth of the first monthnThe predicted output of the day is calculated,Nis the firstmDays of the month;
introducing a seasonal fluctuation factor into the historical time sequence data set to obtain a new time sequence data set, wherein the new time sequence data set comprises the following concrete steps:
Figure QLYQS_5
；
wherein ,
Figure QLYQS_6
for a new time series data set, +.>
Figure QLYQS_7
To introduce seasonal wave factor aftermMonth of the first monthnPredicted output of day, < > on >>
Figure QLYQS_8
Is a seasonal wave factor->
Figure QLYQS_9
，/>
Figure QLYQS_10
Is the firstmAverage power generation load for one month, < >>
Figure QLYQS_11
Is the firstmAverage annual power generation load of the year of the month;
regrouping the new time sequence data sets according to the same day of different months to obtain:
Figure QLYQS_12
；
wherein ,
Figure QLYQS_13
for reorganizing the time series data set +.>
Figure QLYQS_14
Is formed by the first month of all monthsnThe set of predicted output of the day,Mis the month number;
for the reorganized time series data set
Figure QLYQS_15
Performing fractional order cumulative calculation to obtainrThe order cumulative sequence is:
Figure QLYQS_16
；
Figure QLYQS_17
；
wherein ,
Figure QLYQS_18
，/>
Figure QLYQS_19
for reorganising time-series data setsrA step accumulation sequence;
according torOrder cumulative sequence
Figure QLYQS_20
Calculating temporary intermediate variable +.>
Figure QLYQS_21
and />
Figure QLYQS_22
The method specifically comprises the following steps:
Figure QLYQS_23
wherein ,
Figure QLYQS_24
for average sequence, +.>
Figure QLYQS_25
，BAnd Yis a temporary intermediate variable matrix;
to temporary intermediate variables
Figure QLYQS_26
Inserting a time sequence response function to obtain:
Figure QLYQS_27
；
wherein ,
Figure QLYQS_28
is the firstM+1 monthnPredictive value of day;
when (when)m＝M+1, the predicted value sequence is:
Figure QLYQS_29
；
performing dimension reduction operation on the predicted value sequence to obtain a dimension reduced predicted value sequence as follows:
Figure QLYQS_30
；
wherein ,
Figure QLYQS_31
and the predicted value sequence after dimension reduction is expressed as:
Figure QLYQS_32
；
predicted value sequence after dimension reduction according to all days in month
Figure QLYQS_33
The regrouping is: />
Figure QLYQS_34
；
Figure QLYQS_35
；
Figure QLYQS_36
；
Figure QLYQS_37
；
Figure QLYQS_38
；
wherein ,
Figure QLYQS_39
is the predicted firstdTime series data set of month +.>
Figure QLYQS_40
Is composed ofdA set of predictions for all days of the month,/->
Figure QLYQS_41
Is the predicted firstdMonth of the first monthnPredictive value of day.
2. The power generation load prediction model training method according to claim 1, characterized in that: the specific formula for carrying out normalization processing on each original data sample is as follows:
Figure QLYQS_42
；
wherein ,
Figure QLYQS_43
is the first of the original data samplesiData of->
Figure QLYQS_44
Data for different historical time points +.>
Figure QLYQS_45
Average value of>
Figure QLYQS_46
Data for different historical time points +.>
Figure QLYQS_47
Variance of->
Figure QLYQS_48
For data->
Figure QLYQS_49
Is performed according to the normalization processing result of the (a).
3. The power generation load prediction model training method according to claim 1, characterized in that: the number of the prediction models is 3, and the 3 prediction models are a long and short memory neural network model, a decision tree model and a support vector machine model respectively.
4. The power generation load prediction model training method according to claim 1, characterized in that: the calculation formula of the predicted output quantity is as follows:
Figure QLYQS_50
；
wherein ,
Figure QLYQS_51
predicted output for historical time points, +.>
Figure QLYQS_52
Is the firstqWeight value of each prediction model, +.>
Figure QLYQS_53
Is the firstqThe output of the individual predictive models is determined,Qis the number of predictive models.
5. The power generation load prediction model training method according to claim 1, characterized in that: the evaluation index is an average absolute percentage error, and a specific calculation formula of the average absolute percentage error is as follows:
Figure QLYQS_54
；
wherein ,
Figure QLYQS_55
is the mean absolute percentage error,/>
Figure QLYQS_56
For predictive value +.>
Figure QLYQS_57
For the target value, N is the number of days of the predicted month. />
6. A power generation load prediction method, characterized by comprising the steps of:
acquiring various index data of an electric energy source at a certain time point and generating environment data corresponding to the time point;
the index data and the power generation environment data are input into each prediction model after normalization processing, and the prediction output quantity of each prediction model is obtained;
carrying out weighted average on the predicted output quantity of each predicted model to obtain a final predicted value;
wherein each prediction model is a model trained by the power generation load prediction model training method according to any one of claims 1 to 5.
7. An electronic device, comprising: a processor and a memory storing a computer program, characterized by: the processor is configured to perform the power generation load prediction model training method of any one of claims 1 to 5 or to perform the power generation load prediction method of claim 6 when running a computer program.
8. A computer-readable storage medium having stored thereon a computer program, characterized by: the program when executed by a processor realizes the power generation load prediction model training method according to any one of claims 1 to 5, or executes the power generation load prediction method according to claim 6.
CN202211598178.1A 2022-12-14 2022-12-14 Power generation load prediction model training method, prediction method, device and storage medium Active CN115660228B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211598178.1A CN115660228B (en) 2022-12-14 2022-12-14 Power generation load prediction model training method, prediction method, device and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211598178.1A CN115660228B (en) 2022-12-14 2022-12-14 Power generation load prediction model training method, prediction method, device and storage medium

Publications (2)

Publication Number Publication Date
CN115660228A CN115660228A (en) 2023-01-31
CN115660228B true CN115660228B (en) 2023-04-28

Family

ID=85023135

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211598178.1A Active CN115660228B (en) 2022-12-14 2022-12-14 Power generation load prediction model training method, prediction method, device and storage medium

Country Status (1)

Country Link
CN (1) CN115660228B (en)

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9053439B2 (en) * 2012-09-28 2015-06-09 Hewlett-Packard Development Company, L.P. Predicting near-future photovoltaic generation
CN105205326B (en) * 2015-09-24 2017-11-10 渤海大学 A kind of power plant's Real-time Load on-line prediction method
CN109242190A (en) * 2018-09-19 2019-01-18 四川大学 Mid-long term load forecasting method and system based on BFGS-FA optimization fractional order gray model
US11586706B2 (en) * 2019-09-16 2023-02-21 Oracle International Corporation Time-series analysis for forecasting computational workloads
US11900282B2 (en) * 2020-01-21 2024-02-13 Hcl Technologies Limited Building time series based prediction / forecast model for a telecommunication network
CN111626476B (en) * 2020-04-23 2023-04-07 渤海大学 Wind power plant wind power generation capacity prediction method
AU2020104000A4 (en) * 2020-12-10 2021-02-18 Guangxi University Short-term Load Forecasting Method Based on TCN and IPSO-LSSVM Combined Model
CN113379153A (en) * 2021-06-28 2021-09-10 北京百度网讯科技有限公司 Method for predicting power load, prediction model training method and device

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
何路缘.基于改进的灰色模型在中长期电力负荷预测中的应用研究.硕士电子期刊工程科技ii辑.2019,全文. *
程熙.基于行业特性的阜宁地区月度电量预测研究.中国优秀硕士论文电子期刊 工程科技Ⅱ辑.2019,全文. *
龙勇 ; 苏振宇 ; 汪於 ; .基于季节调整和BP神经网络的月度负荷预测.***工程理论与实践.2018,(04),全文. *

Also Published As

Publication number Publication date
CN115660228A (en) 2023-01-31

Similar Documents

Publication Publication Date Title
Wu et al. A novel hybrid model for short‐term forecasting in PV power generation
Andresen et al. Validation of Danish wind time series from a new global renewable energy atlas for energy system analysis
Jafarian-Namin et al. Forecasting the wind power generation using Box–Jenkins and hybrid artificial intelligence: A case study
Sarmas et al. ML-based energy management of water pumping systems for the application of peak shaving in small-scale islands
Khalid Wind power economic dispatch–impact of radial basis functional networks and battery energy storage
CN114792166A (en) Energy carbon emission optimization prediction method and device based on multiple constraints
Ameli et al. Economical load distribution in power networks that include hybrid solar power plants
Derse et al. Optimal site selection for wind energy: a case study
CN113887809A (en) Power distribution network supply and demand balance method, system, medium and computing equipment under double-carbon target
Wang et al. Multi‐stage stochastic wind‐thermal generation expansion planning with probabilistic reliability criteria
CN115660228B (en) Power generation load prediction model training method, prediction method, device and storage medium
Kareem et al. Removing seasonal effect on city based daily electricity load forecasting with linear regression
CN111126716A (en) Extreme gradient lifting algorithm-based system model for predicting electricity price
Pang et al. A photovoltaic power predicting model using the differential evolution algorithm and multi-task learning
Al-Haija et al. Time-series model for forecasting short-term future additions of renewable energy to worldwide capacity
Rekhade et al. Forecasting sector-wise electricity consumption for India using various regression models.
Hasche et al. Effects of improved wind forecasts on operational costs in the German electricity system
Ceylan et al. Harmony search algorithm for transport energy demand modeling
CN113723717A (en) Method, device, equipment and readable storage medium for predicting short-term load before system day
Dehghanzadeh et al. Mid-term load forecasting for Iran power system using seasonal autoregressive integrated moving average model (SARIMA)
Abeywickrama et al. Integrating weather patterns into machine learning models for improved electricity demand forecasting in Sri Lanka
Shendryk et al. Short-term Solar Power Generation Forecasting for Microgrid
Islam et al. Deep Learning Technique for Forecasting Solar Radiation and Wind Speed for Dynamic Microgrid Analysis.
Wang et al. [Retracted] A Stochastic Rolling Horizon‐Based Approach for Power Generation Expansion Planning
Mo et al. Modeling and quantifying the importance of snow storage information for the nordic power system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant