CN116845874A - Short-term prediction method and device for power load - Google Patents
Short-term prediction method and device for power load Download PDFInfo
- Publication number
- CN116845874A CN116845874A CN202310824056.8A CN202310824056A CN116845874A CN 116845874 A CN116845874 A CN 116845874A CN 202310824056 A CN202310824056 A CN 202310824056A CN 116845874 A CN116845874 A CN 116845874A
- Authority
- CN
- China
- Prior art keywords
- load
- prediction
- fedformer
- historical
- sample
- 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 67
- 238000012549 training Methods 0.000 claims abstract description 76
- 238000007781 pre-processing Methods 0.000 claims abstract description 30
- 238000004364 calculation method Methods 0.000 claims abstract description 24
- 238000012545 processing Methods 0.000 claims description 30
- 230000006870 function Effects 0.000 claims description 27
- 238000012360 testing method Methods 0.000 claims description 24
- 230000015654 memory Effects 0.000 claims description 18
- 238000010606 normalization Methods 0.000 claims description 14
- 238000012795 verification Methods 0.000 claims description 10
- 238000004458 analytical method Methods 0.000 claims description 5
- 238000010276 construction Methods 0.000 claims description 5
- 238000005516 engineering process Methods 0.000 abstract description 4
- 230000035772 mutation Effects 0.000 description 22
- 238000010586 diagram Methods 0.000 description 19
- 239000004973 liquid crystal related substance Substances 0.000 description 12
- 239000011159 matrix material Substances 0.000 description 11
- 238000004891 communication Methods 0.000 description 9
- 230000008569 process Effects 0.000 description 9
- 238000000354 decomposition reaction Methods 0.000 description 8
- 230000002159 abnormal effect Effects 0.000 description 7
- 238000004590 computer program Methods 0.000 description 7
- 239000013598 vector Substances 0.000 description 7
- 230000000694 effects Effects 0.000 description 6
- 230000001932 seasonal effect Effects 0.000 description 6
- 238000013528 artificial neural network Methods 0.000 description 5
- 238000013500 data storage Methods 0.000 description 4
- 230000006872 improvement Effects 0.000 description 4
- 238000013480 data collection Methods 0.000 description 3
- 230000007774 longterm Effects 0.000 description 3
- 238000007726 management method Methods 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 238000012935 Averaging Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 230000000737 periodic effect Effects 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- DBOHWMPKJCJANT-UHFFFAOYSA-N brooker's merocyanine Chemical compound C1=CN(C)C=CC1=CC=C1C=CC(=O)C=C1 DBOHWMPKJCJANT-UHFFFAOYSA-N 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 238000004146 energy storage Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 238000012821 model calculation 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
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/003—Load forecast, e.g. methods or systems for forecasting future load demand
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/10—Pre-processing; Data cleansing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Power Engineering (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application relates to the technical field of power system automation, in particular to a short-term prediction method and device of power load, wherein the method comprises the steps of constructing a historical sample data set of a power system; the historical sample data set comprises historical load data, historical weather forecast data and daily load curve types; preprocessing the historical sample data set, and obtaining a preprocessing result; inputting the preprocessed historical sample data set into a target Fedformer prediction model, and training and optimizing the target Fedformer prediction model to obtain an optimal Fedformer prediction model; and obtaining a load prediction sample of the power system, and predicting the load prediction sample through the optimal Fedformer prediction model to obtain a load prediction value of the load prediction sample. The scheme can improve the prediction precision of the short-term prediction technology of the power load and shorten the calculation time of the prediction precision.
Description
Technical Field
The application relates to the technical field of power system load prediction, in particular to a short-term power load prediction method and device.
Background
As a key ring of the power system, the short-term prediction of the power load is important for the safety stability and economic reliability of the power system. Short-term prediction of power load belongs to a long-time series prediction problem, and it is generally required to predict a load level of the power system for 1-8 days (time resolution is 15 minutes, 96-768 periods) in the future, and it is difficult to ensure accuracy of short-term prediction of power load due to high uncertainty of the power system.
In recent years, with the rapid development of power systems, new technologies and new members such as virtual power plants, demand response, load aggregators and the like are introduced, and a large number of distributed elements such as electric vehicles, new energy sources and energy storage are accessed, so that the complexity and uncertainty of the power systems are more serious, and a greater challenge is brought to short-term prediction of power loads. In recent years, on the basis of machine learning, the theory and method of deep learning have been greatly developed and advanced, and because the power load data is time series data with a certain periodic variation rule, deep models such as CNN (convolutional neural network), RNN (recurrent neural network), LSTM (long short term memory artificial neural network) and the like have the most wide application in the aspect of short term load prediction, and have the problems of long term memory forgetting, high computational complexity and the like in the aspect of prediction accuracy, which are improved compared with the machine learning. The most advanced deep learning method based on the transducer model and the improved model thereof can effectively improve the problem of long-term information loss of short-term load prediction, so that the prediction precision is further improved, but the problems of lower calculation efficiency, incapability of capturing the overall change trend of the load, incapability of considering load mutation caused by unknown reasons and the like still exist.
Therefore, it is desirable to improve the prediction accuracy of the existing short-term prediction technique for power load and to shorten the calculation time.
Disclosure of Invention
The application provides a short-term prediction method of power load, which can improve the prediction precision of the short-term prediction technology of the power load and shorten the calculation time of the power load.
In one aspect, a method of short-term prediction of electrical load is provided, the method comprising:
constructing a historical sample data set of the power system; the historical sample data set comprises historical load data, historical weather forecast data and daily load curve types;
preprocessing the historical sample data set, and obtaining a preprocessing result; the preprocessing comprises feature coding processing and normalization processing;
inputting the preprocessed historical sample data set into a target FedFormer prediction model, and training and optimizing the target FedFormer prediction model to obtain an optimal FedFormer prediction model;
and obtaining a load prediction sample of the power system, and predicting the load prediction sample through the optimal Fedformer prediction model to obtain a load prediction value of the load prediction sample.
In yet another aspect, there is provided a short-term predictive device of electrical load, the device comprising:
the historical sample data set construction module is used for constructing a historical sample data set of the power system; the historical sample data set comprises historical load data, historical weather forecast data and daily load curve types;
the preprocessing result acquisition module is used for preprocessing the historical sample data set and acquiring a preprocessing result; the preprocessing comprises feature coding processing and normalization processing;
the training and optimizing module is used for inputting the preprocessed historical sample data set into a target Fedformer prediction model, and training and optimizing the target Fedformer prediction model to obtain an optimal Fedformer prediction model;
and the load prediction value acquisition module is used for acquiring a load prediction sample of the power system, and predicting the load prediction sample through the optimal Fedformer prediction model so as to acquire a load prediction value of the load prediction sample.
In a possible implementation manner, the historical sample data set construction module is further configured to:
acquiring historical load data and historical weather forecast data of a power system;
Classifying the daily historical load data according to date attributes to obtain daily load curve types; the date attribute comprises a working day, a double holiday, a legal holiday and an legal holiday;
and constructing the historical sample data set according to the historical load data, the historical weather forecast data and the daily load curve type.
In one possible embodiment, the apparatus is further for:
the historical sample data set is divided into a training sample set, a test sample set and a verification sample set according to a specified proportion.
In a possible implementation manner, the training and optimizing module is further configured to:
inputting the preprocessed training sample set to the target FedFormer prediction model, and training the target FedFormer prediction model;
and optimizing the trained target FedFormer prediction model through the preprocessed verification sample set to obtain an optimal FedFormer prediction model.
In one possible embodiment, the apparatus is further for:
inputting the preprocessed test sample set to the optimal Fedformer prediction model to obtain a load test value of the test sample set;
And evaluating the prediction accuracy of the optimal Fedformer prediction model according to the mean square error analysis result of the load test value.
In one possible implementation manner, the load prediction value acquisition module is further configured to:
preprocessing the load prediction sample to obtain load prediction characteristics of the load prediction sample;
and inputting the load prediction characteristics of the load prediction samples into the optimal Fedformer prediction model to obtain the load prediction values of the load prediction samples.
In one possible embodiment, the apparatus is further for:
acquiring historical load time series data corresponding to the historical sample data set, and performing sliding window processing of a specified proportion on the historical load time series data to acquire a training sample set; each training sample in the training sample set comprises an input load sequence and a predicted output load sequence;
carrying out load local difference calculation on the training sample, and obtaining a local difference calculation result;
and according to the local difference calculation result, assigning a loss weight to each training sample so as to improve the loss function of the target Fedformer prediction model.
In one possible embodiment, the apparatus is further for:
and reducing the loss weight of the training samples with larger absolute values of local differences.
In yet another aspect, a computer device is provided that includes a processor and a memory having at least one instruction stored therein that is loaded and executed by the processor to implement a short-term prediction method of electrical load as described above.
In yet another aspect, a computer readable storage medium having stored therein at least one instruction loaded and executed by a processor to implement a method of short-term prediction of electrical load as described above is provided.
In yet another aspect, a computer program product or computer program is provided, the computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer readable storage medium and executes the computer instructions to cause the computer device to perform a short-term prediction method of electrical load as described above.
The technical scheme provided by the application can comprise the following beneficial effects:
according to the application, a Fedformer prediction model is adopted to predict the power load in a short period, so that the prediction performance is ensured, and the prediction efficiency is improved; secondly, because the real load data has great randomness and unbalance, and the load mutation situation happens sometimes, the application proposes to improve the loss function of model training, namely, in the prediction model training process, the loss weight of normal load is improved by adopting a mode of reassigning the weight to the loss function, the loss weight of mutation load is reduced, so that the learning strength of the model to the normal load is improved, the interference of abnormal load on model learning is lightened, and the popularization capacity and the prediction effect of the prediction model are improved;
according to the method, the load curves of the working days, the double holidays, the legal holidays and the non-legal holidays are distinguished, the load curve types of the prediction days are informed to the prediction model in advance, the recognition capability of the prediction model on the holiday load mode is enhanced, and therefore the prediction accuracy of the model on the holiday load is improved.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings that are needed in the description of the embodiments or the prior art will be briefly described, and it is obvious that the drawings in the description below are some embodiments of the present application, and other drawings can be obtained according to the drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic diagram illustrating a short-term prediction system of electrical loads, according to an example embodiment.
FIG. 2 is a method flow diagram illustrating a method of short-term prediction of electrical load, according to an example embodiment.
FIG. 3 is a method flow diagram illustrating a method of short-term prediction of electrical load, according to an example embodiment.
FIG. 4 is a schematic overall flow diagram of short-term prediction of unbalanced power load based on a Fedformer predictive model, according to an example embodiment.
FIG. 5 is a sample feature stitching schematic diagram that is shown in accordance with an exemplary embodiment.
Fig. 6 is a schematic diagram of the overall structure of a Fedformer predictive model, according to an example embodiment.
Fig. 7 is a schematic diagram illustrating a configuration of a Fedformer frequency boosting module according to an exemplary embodiment.
Fig. 8 is a schematic diagram illustrating a configuration of a Fedformer frequency boosting module according to an exemplary embodiment.
Fig. 9 is a block diagram showing a structure of a short-term prediction apparatus of an electric load according to an exemplary embodiment.
Fig. 10 shows a block diagram of a computer device according to an exemplary embodiment of the present application.
Detailed Description
The following description of the embodiments of the present application will be made apparent and fully in view of the accompanying drawings, in which some, but not all embodiments of the application are shown. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.
It should be understood that the "indication" mentioned in the embodiments of the present application may be a direct indication, an indirect indication, or an indication having an association relationship. For example, a indicates B, which may mean that a indicates B directly, e.g., B may be obtained by a; it may also indicate that a indicates B indirectly, e.g. a indicates C, B may be obtained by C; it may also be indicated that there is an association between a and B.
In the description of the embodiments of the present application, the term "corresponding" may indicate that there is a direct correspondence or an indirect correspondence between the two, or may indicate that there is an association between the two, or may indicate a relationship between the two and the indicated, configured, etc.
In the embodiment of the present application, the "predefining" may be implemented by pre-storing corresponding codes, tables or other manners that may be used to indicate relevant information in devices (including, for example, terminal devices and network devices), and the present application is not limited to the specific implementation manner thereof.
FIG. 1 is a schematic diagram illustrating a short-term prediction system of electrical loads, according to an example embodiment. The short-term prediction system includes a server 110 and a power system 120.
The short-term prediction system is used for predicting the load of the power system 120 to manage energy, so as to improve the operation efficiency of the power system 120.
Optionally, the power system 120 may include a data collection device and a data storage module, where the data collection device may collect power load data and weather data during operation of the power system, and store the collected power load data in the data storage module.
Optionally, the power system 120 is in communication connection with the server 110 through a transmission network (such as a wireless communication network), the power system 120 may upload each data (such as the collected power load data and the meteorological data, i.e. the historical sample data) stored in the data storage module to the server 110 through the wireless communication network, so that the server 110 processes the collected historical sample data to construct a historical sample data set of the power system, and perform training and optimization of a prediction model according to the historical sample data set of the power system, and implement load prediction of the power system by using the optimized prediction model.
Alternatively, the server 110 may be a server cluster or a distributed system formed by a plurality of physical servers, and may also be a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDNs, and technical computing services such as big data and artificial intelligence platforms.
Optionally, the system may further include a management device, where the management device is configured to manage the system (e.g., manage a connection state between each module and the server, etc.), where the management device is connected to the server through a communication network. Optionally, the communication network is a wired network or a wireless network.
Alternatively, the wireless network or wired network described above uses standard communication techniques and/or protocols. The network is typically the internet, but may be any other network including, but not limited to, a local area network, a metropolitan area network, a wide area network, a mobile, a limited or wireless network, a private network, or any combination of virtual private networks. In some embodiments, techniques and/or formats including hypertext markup language, extensible markup language, and the like are used to represent data exchanged over a network. All or some of the links may also be encrypted using conventional encryption techniques such as secure socket layer, transport layer security, virtual private network, internet protocol security, etc. In other embodiments, custom and/or dedicated data communication techniques may also be used in place of or in addition to the data communication techniques described above.
FIG. 2 is a method flow diagram illustrating a method of short-term prediction of electrical load, according to an example embodiment. As shown in fig. 2, the short-term prediction method may include the steps of:
Step S201, constructing a historical sample data set of the power system; the historical sample data set includes historical load data, historical weather forecast data, and daily load curve types.
In one possible implementation, when the power load of the power system is predicted in a short period, historical sample data of the power system is collected first, and a historical sample data set is constructed; the historical sample data can comprise historical load data, historical weather forecast data and daily load curve types of the power system, and the historical load data, the historical weather forecast data and the daily load curve types are used for learning and training so as to achieve short-term forecast of the power load.
Step S202, preprocessing the historical sample data set, and obtaining a preprocessing result; the preprocessing includes feature encoding processing and normalization processing.
In one possible implementation, after the historical sample data set of the power system is constructed, preprocessing, such as feature encoding processing, normalization processing and the like, is required to be performed on the historical sample data set, and the feature encoding is to map original discrete features into classified feature vectors; the normalization processing is to normalize the maximum value and the minimum value of the historical sample data set, the normalization is a dimensionless processing means, the absolute values of different numerical values are changed into relative value relations, and the normalization processing is an effective method for simplifying calculation and reducing the numerical values. The preprocessed historical sample data set forms a historical sample load characteristic, and the historical sample load characteristic is input into a target Fedformer prediction model to perform load prediction.
Step S203, the preprocessed historical sample data set is input into a target Fedformer prediction model, and the target Fedformer prediction model is trained and optimized to obtain an optimal Fedformer prediction model.
In one possible implementation, the preprocessed historical sample dataset is input to a target Fedformer predictive model, which is trained and optimized based on the historical sample dataset; in addition, because the real load data has great randomness and unbalance, and the load mutation situation happens sometimes, when the prediction model is trained, the target Fedformer prediction model can be subjected to loss function improvement, the loss weight of normal load is improved, the loss weight of mutation load is reduced, the learning strength of the prediction model on the normal power load is improved, the interference of abnormal power load on the prediction model learning is reduced, and the popularization capability and the prediction effect of the prediction model are improved.
Step S204, a load prediction sample of the power system is obtained, and the load prediction sample is predicted through the optimal Fedformer prediction model, so that a load prediction value of the load prediction sample is obtained.
In one possible implementation manner, the optimal Fedformer prediction model is an improved Fedformer predictor generated through training of the prediction model, load prediction of the load prediction sample can be achieved through the improved Fedformer predictor, the predicted load characteristic of the load prediction sample is input into the improved Fedformer predictor, and the improved Fedformer predictor outputs the load prediction value of the load prediction sample.
In conclusion, the power load is predicted in a short period by adopting the Fedformer prediction model, so that the prediction performance is ensured, and the prediction efficiency is improved; secondly, because the real load data has great randomness and unbalance, and the load mutation situation happens sometimes, the application proposes to improve the loss function of model training, namely, in the prediction model training process, the loss weight of normal load is improved by adopting a mode of reassigning the weight to the loss function, the loss weight of mutation load is reduced, so that the learning strength of the model to the normal load is improved, the interference of abnormal load on model learning is lightened, and the popularization capacity and the prediction effect of the prediction model are improved;
according to the method, the load curves of the working days, the double holidays, the legal holidays and the non-legal holidays are distinguished, the load curve types of the prediction days are informed to the prediction model in advance, the recognition capability of the prediction model on the holiday load mode is enhanced, and therefore the prediction accuracy of the model on the holiday load is improved.
FIG. 3 is a method flow diagram illustrating a method of short-term prediction of electrical load, according to an example embodiment. As shown in fig. 3, the short-term prediction method may include the steps of:
step S301, constructing a historical sample data set of the power system; the historical sample data set includes historical load data, historical weather forecast data, and daily load curve types.
In one possible implementation manner, the step S301 includes: acquiring historical load data and historical weather forecast data of a power system;
classifying the daily historical load data according to date attributes to obtain daily load curve types; the date attributes include workdays, double holidays, legal holidays and non-legal holidays;
and constructing the historical sample data set according to the historical load data, the historical weather forecast data and the daily load curve type.
Further, since the load levels of the workday, the double holiday and the holiday are greatly different under normal conditions, and the load level of the legal holiday in the holiday is generally lower than that of the non-legal holiday, the load types (the load types are the date types of the load here) are classified into the workday, the double holiday, the legal holiday and the non-legal holiday according to the date attribute, the load types of the prediction day are used as the input of the prediction model, the recognition capability of the prediction model on the holiday load is enhanced, and the prediction precision of the holiday load is improved.
In one possible implementation, the historical sample data set is divided into a training sample set, a test sample set, and a validation sample set according to a specified ratio. Referring to fig. 4, which is a schematic diagram of a short-term prediction overall flow of unbalanced power load based on a Fedformer prediction model, in a data collection process, historical load data of a power system is collected from energy consumption data of a smart meter, daily load curves are classified (into workday, double-holiday, legal holiday and non-legal holiday load curve types), a historical sample data set including the historical load data, the historical weather prediction data and the daily load curve types is constructed, and the historical sample data set is divided into a training sample set, a test sample set and a verification sample set, wherein the ratio of the training sample set, the test sample set and the verification sample set can be 70%, 20% and 10% respectively.
Step S302, preprocessing the historical sample data set, and obtaining a preprocessing result; the preprocessing includes feature encoding processing and normalization processing.
Further, as shown in fig. 4, the preprocessing mainly includes two parts, namely, feature encoding processing and maximum value minimum value normalization processing. The feature encoding process maps the original discrete features into classification feature vectors with values between 1-n, where n is the total number of types of load features. For example, the temporal features may be encoded as: the values of 1-366, 1-53 and 1-12 are respectively taken on the day, week and month of the year; the values of 1-31,1-5 are respectively taken on the day and week of a month; the day of the week is the value of 1-7; the hour of the day is 1-24; the number of minutes in one hour is 1-60. Daily load curve types were encoded as: the load type of the workday is given a value of 1, the holiday is given a value of 2, the non-legal holiday is given a value of 3, and the legal holiday is given a value of 4. The normalization processing of the maximum value and the minimum value is to perform normalization processing by utilizing the maximum value and the minimum value in the data column, the normalized value is between 0 and 1, and the calculation mode can be that the difference between the data and the minimum value in the data column is calculated and divided by the extremely difference.
Further, in order to enhance the identification capability of the FedFormer prediction model on holiday load modes, daily loads are classified into workday, double holidays, legal holidays and non-legal holiday load types; that is, the load curves of working days (monday to friday) are classified into a class, the load curves of double holidays (friday and friday) are classified into a class, the legal holidays (3 days and 5 days of holidays, usually the holiday day, for example, the legal holiday during the primordial denier is 1 month 1 day; 7 days of holidays, usually the first 3 days of holiday) are classified into a class, and the non-legal holidays (all of the non-legal holidays except the legal holiday in the holiday) are classified into a class. If the holiday is repeated with the working day and the double holidays, the holiday is defined.
Assume that the historical sequence including load and weather features isWherein N is p For the number of history features, the time feature sequence is +.>Wherein N is t Representing the number of time features, the load curve type sequence of the prediction day is C E R O×1 . Because the dimensions of the P, T, C matrix are inconsistent, the samples are spliced and accumulated after being projected to the same dimension, a sample characteristic splicing schematic diagram is shown in fig. 5, and an accumulation expression is as follows:
X＝Pα p +Tα t +α c C；
Wherein, the liquid crystal display device comprises a liquid crystal display device,projection operators of P, T, C matrix, X E R I×N The sum of the projection matrices for P, T, C is used as the input of the Fedformer predictive model.
Step S303, acquiring historical load time series data corresponding to the historical sample data set, and performing sliding window processing of a specified proportion on the historical load time series data to acquire a training sample set; each training sample in the training sample set includes an input load sequence and a predicted output load sequence.
Furthermore, when training a training model, a preprocessed training sample set is required to be input into a target Fedformer prediction model to train the target Fedformer prediction model, short-term load prediction is performed by adopting the Fedformer prediction model, long-term dependence characteristics of long-sequence load data can be better mined, the problem that long-term characteristics of long-sequence load prediction are forgotten and the problem that calculation time is too long in the related technology are solved, but as the current power load presents the characteristics of diversification and flexibility and variability, the randomness and the volatility are enhanced increasingly, the unbalance is greatly increased, and in terms of the overall distribution of the long-sequence power load data, in order to ensure the prediction precision of short-term load prediction with stronger unbalance, the embodiment proposes to improve the loss function of the Fedformer prediction model, and in the model training process, the loads in a normal state and a mutation state are re-weighted to reduce the influence of the mutation load on the prediction model learning, and the generalization capability and the popularization capability of the prediction model are improved.
The principle of the Fedformer predictive model is analyzed as follows:
firstly, the Fedformer prediction model is based on time sequence frequency domain modeling, time sequence data can be modeled from two angles of time domain and frequency domain, most of time sequence prediction models are time domain models, and the Fedformer prediction models are different from other depth prediction models in that frequency domain modeling is adopted in a neural network, so that the prediction models can be used for capturing global characteristics of time sequences better. Fourier transforms are a common approach to frequency domain modeling and simply preserving all frequency components may lead to poor prediction results because many high frequency variations in the time series are caused by noise input, but preserving only low frequency components may not be suitable as some trend variations in the time series represent important events. The Fedformer predictive model represents the time series by randomly selecting a certain number of Fourier components (including high frequency components and low frequency components) to ensure the effectiveness of the predictive model calculation.
Assume that there are n time series variables, each denoted as X 1 (t),…,X n (t) subjecting any one of the n time-series variables X to Fourier transform i (t) conversion to a vector a i ＝(a i, ,a i, ,…,a i, ) T ∈R d The method comprises the steps of carrying out a first treatment on the surface of the At this time, all fourier transform vectors can be formed into one matrix a= (a) 1 ,a 2 ,…,a n ) T ∈R n×d Each row of matrix a corresponds to a different time series variable and each column corresponds to a different fourier component. While preserving all fourier components can best preserve the historical information in the time series, it can overfit the historical data, resulting in poor predictions of future signals. Therefore, a subset of fourier components needs to be selected, which should be small enough to avoid the over-fitting problem, and should be able to preserve a large part of the history information. The Fedformer predictive model is a model of uniformly and randomly selecting s components (s<d) The s components are denoted as i respectively 1 ,i 2 ,…,i s Constructing a matrix S.epsilon. {0,1} s×d If i=i k ，S i, =1, where i is any one of the original d fourier components; otherwise, S i, =0. The multivariate time series can then be represented as A ′ ＝AS T ∈R n×s . Although the Fourier basis is randomly selected, in general, A ′ Most of the information from a can be retained.
To measure A ′ To what extent information from A can be saved, each column vector of A is projected to the vector of A ′ In the subspace spanned by the middle column vector. Let P A′ (A) Is a projected result matrix, wherein P A′ (. Cndot.) is the projection operator. If A ′ Most of A information is reserved, then A and P A′ (A) There will be some error between, i.e. |A-P A′ (A) I (I); let A k As an approximation obtained by the first k-max single-valued decomposition of a, the following equation indicates that if the number s of randomly sampled fourier components is k 2 Order of magnitude, |A-P A′ (A) The I approaches the I A-A k I, the error of both is denoted by e.
Assuming that the coherence measure μ (a) of matrix a is Ω (k/n), then:
|A-P A′ (A)|≤(1+∈)|A-A k |；
if s＝O(k 2 /∈ 2 )；
since the univariates in the multivariate time series not only depend on their past values, but also are interdependent, sharing similar frequency components, the corresponding matrix a obtained by fourier transformation is typically of low rank nature. Thus, as shown in the above equation, the randomly selected subset of fourier components may suitably embody the information in the fourier matrix a.
Next, referring to the overall structure schematic diagram of the Fedformer prediction model shown in fig. 6, the Fedformer prediction model mainly includes an input module (encoder input), a frequency enhancement module, a forward neural network, a frequency enhancement attention and expert mixed decomposition module, and the like.
The Fedformer encoder adopts a multi-layer structure:wherein->Representing an input history sequence; / >Represent the firstOutputs of l encoder layers, l e {1, …, N }; the Encoder (·) is a Fedformer Encoder module, whose expression is as follows:
wherein, the liquid crystal display device comprises a liquid crystal display device,representing seasonal components after the ith mixed expert decomposition module in layer l, i e {1,2}; MOED (·), FEB (·) and FF (·) are denoted as hybrid expert decomposition module, frequency enhancement module and forward neural network module, respectively.
The Fedformer decoder also employs a multi-layer structure:wherein the method comprises the steps ofInitializing values for seasonal and trending components, respectively; />Output seasonal and trending components for the first decoder layer, respectively, l e {1, …, M }; the Decoder (-) is a Fedformer Decoder module, which has the following expression:
wherein, the liquid crystal display device comprises a liquid crystal display device, is->Respectively representing seasonal and trending components after the ith mixed expert decomposition module in the first layer; FEA (·) is a frequency-enhanced attention module, W l,i ,i∈{1,2,3}；W l,i Representing the i-th trend component +.>Is used for the projection weight of the image.
The final prediction result is the sum of the seasonal and trending components after multiple refinements, i.eWherein W is S The depth-converted seasonal component ++>Projected to the target dimension.
For the above input module, long-time sequence prediction is a sequence-to-sequence problem, and if the input length is I, the output length is O, and the number of time sequence variables is D, the input of the Fedformer encoder is an i×d matrix, and the input of the Fedformer decoder is an (I/2+O) ×d matrix.
Aiming at an input frequency enhancement module, the Fedformer frequency enhancement module adopts discrete Fourier transform to perform frequency domain conversion on time sequence data, so thatRepresenting the fourier transform and the inverse fourier transform, respectively. Given a set of time series data x n N=1, 2, … N, fourier transform is then +.>Wherein i is an imaginary unit, X l L=1, 2, …, L is a complex sequence in the frequency domain. Similarly, inverse Fourier transform->The complexity of the discrete fourier transform is O (N 2 ) If using fastFourier transforms, the computational complexity can be reduced to O (NlogN). Here a random subset of fourier is used and the scale of this subset is defined by a scalar, the computational complexity can be further reduced to O (N) if the mode index selection is performed after the discrete fourier transform and before the inverse transform.
The Fedformer frequency enhancement module is schematically shown in FIG. 7, and has an input x ε R i×D First with w E R D×D Performing linear mapping to obtain q=xw, then converting Q from time domain to frequency domain, and converting Q into Q E C after Fourier transformation N×D . In the frequency domain, only M modes selected randomly are reserved, and the calculation formula is as follows:
wherein, the liquid crystal display device comprises a liquid crystal display device,and M is less than N;
thus, the frequency enhancement module FEB (·) is denoted as:
Wherein R is C M×D×D Parameterized kernels for random initialization; order theAnd->Then the multiplicator +.: />D is the input channel o =1, 2, …, D being the output channel; before performing the inverse fourier transform->Is filled with 0 to C N×D 。
For frequency-enhanced attention, the structure diagram of the frequency-enhanced attention is shown in FIG. 8, the inputs are query, key and value, and q E R are used respectively L×D 、k∈R L×D 、v∈R L×D A representation; in cross-attention, the query comes from the decoder, through q=x en w q Obtained, wherein w is q ∈R D×D The key and value come from the encoder, which can be done by k=x de w k And v=x de w v Obtained, wherein w is k ,w v ∈R D×D The method comprises the steps of carrying out a first treatment on the surface of the A typical attention expression is as follows:
fourier transforming q, k, v and performing a similar attention mechanism in the frequency domain by randomly selecting M modes, the fourier transformed selected versions are represented as respectively Thus, the frequency enhanced attention FEA (·) is as follows:
where σ is the activation function, usually a softmax or tanh function is used, whose convergence performance varies from dataset to dataset. Order theY.epsilon.C before performing the inverse Fourier transform M×D Is needed to be used0 is filled with C L×D 。
For the hybrid expert decomposition module, extracting trends using a fixed window averaging pool can be difficult because the complex periodic patterns typically observed are coupled with the trend components of the real time series data. For this purpose, the Fedformer predictive model designs a hybrid expert decomposition module comprising a set of differently sized averaging filters for extracting a plurality of trend components from the input signal, and a set of data dependent weights for combining them into a final trend. The expression of the mixed expert decomposition module is as follows:
X trend ＝softmax(L(x))*F(x)；
Where F (x) represents the average filter set and softmax (L (x)) represents the weights used to mix the extracted trends.
Based on the basic principle of applying the Fedformer prediction model, the method mainly improves the Fedformer prediction model in two points: firstly, adopting a weight redistribution method to improve a loss function of a model; secondly, the daily load curve is simply classified in advance, the type of the load curve of the predicted day is used as input, and the model is informed in advance.
For the loss function improvement, first, given the historical load time series data l= { L 1 ,l 2 ,…,l T Where T is the length of the payload time series data. By sliding window with fixed size on L, model training sample set can be obtainedEach training sample consists of two parts: input load sequence->And predicted output load sequence->Where I, O are the lengths of the input and output load sequences, respectively. Load prediction model by learning the time dependence in L, in a given pastUnder the condition of the load sequence with the length of I, the load sequence with the length of O in the future is predicted. The improvement of the loss function is mainly used for solving the problem of unbalanced loss caused by the load input-output sequence pairs, if compared with other input-output sequence pairs, the input-output sequence pairs are- >Middle->And->There is a large difference between them, a is the moment that characterizes the abrupt change of load. Since most load time series data sets do not provide labels for mutations, prediction may be inaccurate, and the application achieves unsupervised load mutation identification through improvement of the loss function.
Step S304, carrying out load local difference calculation on the training sample, and obtaining a local difference calculation result.
Furthermore, the load local difference calculation is based on a statistical theory and is used for measuring the difference between two adjacent input and output sequencesAnd->) Differences from other input-output sequence pairs; the load local difference calculation formula is as follows:
wherein, the liquid crystal display device comprises a liquid crystal display device,respectively->And->Mean value of->Respectively->And->Standard deviation of (2).
And step S305, according to the local difference calculation result, assigning a loss weight to each training sample so as to improve the loss function of the target Fedformer prediction model.
In one possible implementation, for a training sample with a larger absolute value of local variance, the loss weight of the training sample is reduced.
Furthermore, because the definition of the load mutation may be very fuzzy, depending on the view angle, the application does not set the moment with large local difference value as the moment of the load mutation and the moment with small local difference value as the normal state of the load, but re-weights the loss function by using the local difference value to reduce the influence of the load mutation on the model learning in the training stage, and because the load training sample with larger absolute value of the local difference is closer to the mutation, the loss weight of the load training sample is reduced.
Obtaining a data set by calculating local differences of the load training sample setThe dataset comprising local variance values LD at the predicted instants t t . Then, the loss weights assigned to each training sample are as follows:
where r is a scaling factor.
Thus, the improved loss function is:
wherein, the liquid crystal display device comprises a liquid crystal display device,load prediction sequence, l, representing time t t+i 、/>The i-th true value in the load output sequence and the i-th predicted value in the load predicted sequence at the time t are respectively represented.
Step S306, the preprocessed historical sample data set is input into a target Fedformer prediction model (the target Fedformer prediction model is a Fedformer prediction model with improved loss function), and the target Fedformer prediction model is trained and optimized to obtain an optimal Fedformer prediction model.
In one possible implementation, the preprocessed training sample set is input to the target Fedformer predictive model, and the target Fedformer predictive model is trained;
optimizing the trained target FedFormer prediction model through the preprocessed verification sample set to obtain an optimal FedFormer prediction model;
inputting the preprocessed test sample set into the optimal Fedformer prediction model to obtain a load test value of the test sample set;
And evaluating the prediction accuracy of the optimal Fedformer prediction model according to the mean square error analysis result of the load test value.
Further, as shown in fig. 4, the preprocessed training sample set is input into an improved Fedformer prediction model (i.e., a target Fedformer prediction model), the target Fedformer prediction model is trained, and parameters of the trained target Fedformer prediction model are optimized by using the preprocessed verification sample set, so that an optimal Fedformer prediction model is obtained; and inputting the preprocessed test sample set into the optimal Fedformer prediction model to obtain a load prediction value of the test sample set, and evaluating the prediction accuracy of the optimal Fedformer prediction model by adopting MSE (mean square error analysis).
Step S307, a load prediction sample of the power system is obtained, and the load prediction sample is predicted through the optimal Fedformer prediction model, so as to obtain a load prediction value of the load prediction sample.
In one possible implementation, the load prediction sample is preprocessed to obtain the load prediction feature of the load prediction sample;
and inputting the load prediction characteristic of the load prediction sample into the optimal Fedformer prediction model to obtain a load prediction value of the load prediction sample.
Further, the load prediction sample is preprocessed to obtain a prediction feature corresponding to the load prediction sample, and the prediction feature of the load prediction sample is input into an improved Fedformer predictor generated through training, optimization and evaluation to obtain a load prediction value. Firstly, the short-term prediction of the power load is a long-time sequence prediction problem, and most of the existing most advanced depth models have a better prediction performance, but have a calculation efficiency problem, so the method uses the most advanced Fedformer prediction model to perform short-term prediction on the power load, and improves the prediction efficiency while guaranteeing the prediction performance; in addition, as the real load data has great randomness and unbalance, the load mutation situation happens sometimes, the application proposes to improve the loss function of model training, namely, in the model training process, the loss weight of normal load is improved by adopting a mode of reassigning the weight to the loss function, the loss weight of mutation load is reduced, so that the learning strength of the model to the normal load is improved, the interference of abnormal load on model learning is lightened, and the popularization capability and the prediction effect of the model are improved. Finally, compared with the load levels of working days and double holidays, the holiday load level is abnormal, but the load level is not really an unidentified mutation load, so that the application distinguishes the load curves of working days, double holidays, legal holidays and non-legal holidays, informs the model of the load curve types of the prediction days in advance, enhances the recognition capability of the model on the holiday load mode, and further improves the prediction accuracy of the model on the holiday load.
In conclusion, the power load is predicted in a short period by adopting the Fedformer prediction model, so that the prediction performance is ensured, and the prediction efficiency is improved; secondly, because the real load data has great randomness and unbalance, and the load mutation situation happens sometimes, the application proposes to improve the loss function of model training, namely, in the prediction model training process, the loss weight of normal load is improved by adopting a mode of reassigning the weight to the loss function, the loss weight of mutation load is reduced, so that the learning strength of the model to the normal load is improved, the interference of abnormal load on model learning is lightened, and the popularization capacity and the prediction effect of the prediction model are improved;
according to the method, the load curves of the working days, the double holidays, the legal holidays and the non-legal holidays are distinguished, the load curve types of the prediction days are informed to the prediction model in advance, the recognition capability of the prediction model on the holiday load mode is enhanced, and therefore the prediction accuracy of the model on the holiday load is improved.
Fig. 9 is a block diagram showing a structure of a short-term prediction apparatus of an electric load according to an exemplary embodiment. The device comprises:
A historical sample data set construction module 901 for constructing a historical sample data set of the power system; the historical sample data set comprises historical load data, historical weather forecast data and daily load curve types;
a preprocessing result obtaining module 902, configured to preprocess the historical sample dataset and obtain a preprocessing result; the preprocessing comprises feature coding processing and normalization processing;
the training and optimizing module 903 is configured to input the preprocessed historical sample dataset to a target Fedformer prediction model, and train and optimize the target Fedformer prediction model to obtain an optimal Fedformer prediction model;
the load prediction value obtaining module 904 is configured to obtain a load prediction sample of the power system, and predict the load prediction sample through the optimal Fedformer prediction model, so as to obtain a load prediction value of the load prediction sample.
In a possible implementation manner, the historical sample data set construction module 901 is further configured to:
acquiring historical load data and historical weather forecast data of a power system;
classifying the daily historical load data according to date attributes to obtain daily load curve types; the date attribute comprises a working day, a double holiday, a legal holiday and an legal holiday;
And constructing the historical sample data set according to the historical load data, the historical weather forecast data and the daily load curve type.
In one possible embodiment, the apparatus is further for:
the historical sample data set is divided into a training sample set, a test sample set and a verification sample set according to a specified proportion.
In one possible implementation, the training and optimization module 903 is further configured to:
inputting the preprocessed training sample set to the target FedFormer prediction model, and training the target FedFormer prediction model;
and optimizing the trained target FedFormer prediction model through the preprocessed verification sample set to obtain an optimal FedFormer prediction model.
In one possible embodiment, the apparatus is further for:
inputting the preprocessed test sample set to the optimal Fedformer prediction model to obtain a load test value of the test sample set;
and evaluating the prediction accuracy of the optimal Fedformer prediction model according to the mean square error analysis result of the load test value.
In a possible implementation manner, the load prediction value obtaining module 904 is further configured to:
Preprocessing the load prediction sample to obtain load prediction characteristics of the load prediction sample;
and inputting the load prediction characteristics of the load prediction samples into the optimal Fedformer prediction model to obtain the load prediction values of the load prediction samples.
In one possible embodiment, the apparatus is further for:
acquiring historical load time series data corresponding to the historical sample data set, and performing sliding window processing of a specified proportion on the historical load time series data to acquire a training sample set; each training sample in the training sample set comprises an input load sequence and a predicted output load sequence;
carrying out load local difference calculation on the training sample, and obtaining a local difference calculation result;
and according to the local difference calculation result, assigning a loss weight to each training sample so as to improve the loss function of the target Fedformer prediction model.
In one possible embodiment, the apparatus is further for:
and reducing the loss weight of the training samples with larger absolute values of local differences.
In conclusion, the power load is predicted in a short period by adopting the Fedformer prediction model, so that the prediction performance is ensured, and the prediction efficiency is improved; secondly, because the real load data has great randomness and unbalance, and the load mutation situation happens sometimes, the application proposes to improve the loss function of model training, namely, in the prediction model training process, the loss weight of normal load is improved by adopting a mode of reassigning the weight to the loss function, the loss weight of mutation load is reduced, so that the learning strength of the model to the normal load is improved, the interference of abnormal load on model learning is lightened, and the popularization capacity and the prediction effect of the prediction model are improved;
According to the method, the load curves of the working days, the double holidays, the legal holidays and the non-legal holidays are distinguished, the load curve types of the prediction days are informed to the prediction model in advance, the recognition capability of the prediction model on the holiday load mode is enhanced, and therefore the prediction accuracy of the model on the holiday load is improved.
Referring to fig. 10, a schematic diagram of a computer device according to an exemplary embodiment of the present application is provided, where the computer device includes a memory and a processor, and the memory is configured to store a computer program, and the computer program is executed by the processor to implement a short-term prediction method of an electrical load as described above.
The processor may be a central processing unit (Central Processing Unit, CPU). The processor may also be any other general purpose processor, digital signal processor (Digital Signal Processor, DSP), application specific integrated circuit (Application Specific Integrated Circuit, ASIC), field programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof.
The memory, as a non-transitory computer readable storage medium, may be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as program instructions/modules, corresponding to the methods in embodiments of the present application. The processor executes various functional applications of the processor and data processing, i.e., implements the methods of the method embodiments described above, by running non-transitory software programs, instructions, and modules stored in memory.
The memory may include a memory program area and a memory data area, wherein the memory program area may store an operating system, at least one application program required for a function; the storage data area may store data created by the processor, etc. In addition, the memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid state storage device. In some implementations, the memory optionally includes memory remotely located relative to the processor, the remote memory being connectable to the processor through a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
In an exemplary embodiment, a computer readable storage medium is also provided for storing at least one computer program that is loaded and executed by a processor to implement all or part of the steps of the above method. For example, the computer readable storage medium may be Read-Only Memory (ROM), random-access Memory (Random Access Memory, RAM), compact disc Read-Only Memory (CD-ROM), magnetic tape, floppy disk, optical data storage device, and the like.
Other embodiments of the application will be apparent to those skilled in the art from consideration of the specification and practice of the application disclosed herein. This application is intended to cover any variations, uses, or adaptations of the application following, in general, the principles of the application and including such departures from the present disclosure as come within known or customary practice within the art to which the application pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the application being indicated by the following claims.
It is to be understood that the application is not limited to the precise arrangements and instrumentalities shown in the drawings, which have been described above, and that various modifications and changes may be effected without departing from the scope thereof. The scope of the application is limited only by the appended claims.
Claims (10)
1. A method of short-term prediction of electrical load, the method comprising:
constructing a historical sample data set of the power system; the historical sample data set comprises historical load data, historical weather forecast data and daily load curve types;
preprocessing the historical sample data set, and obtaining a preprocessing result; the preprocessing comprises feature coding processing and normalization processing;
inputting the preprocessed historical sample data set into a target FedFormer prediction model, and training and optimizing the target FedFormer prediction model to obtain an optimal FedFormer prediction model;
and obtaining a load prediction sample of the power system, and predicting the load prediction sample through the optimal Fedformer prediction model to obtain a load prediction value of the load prediction sample.
2. The method of claim 1, wherein constructing the historical sample dataset of the power system comprises:
acquiring historical load data and historical weather forecast data of a power system;
classifying the daily historical load data according to date attributes to obtain daily load curve types; the date attribute comprises a working day, a double holiday, a legal holiday and an legal holiday;
And constructing the historical sample data set according to the historical load data, the historical weather forecast data and the daily load curve type.
3. The method of claim 1, wherein after said building a historical sample dataset for a power system, the method further comprises:
the historical sample data set is divided into a training sample set, a test sample set and a verification sample set according to a specified proportion.
4. The method of claim 3, wherein inputting the preprocessed set of historical sample data into a target Fedformer predictive model and training and optimizing the target Fedformer predictive model comprises:
inputting the preprocessed training sample set to the target FedFormer prediction model, and training the target FedFormer prediction model;
and optimizing the trained target FedFormer prediction model through the preprocessed verification sample set to obtain an optimal FedFormer prediction model.
5. The method of claim 4, wherein after the obtaining the optimal Fedformer predictive model, the method further comprises:
inputting the preprocessed test sample set to the optimal Fedformer prediction model to obtain a load test value of the test sample set;
And evaluating the prediction accuracy of the optimal Fedformer prediction model according to the mean square error analysis result of the load test value.
6. The method of claim 1, wherein predicting the load prediction samples by the optimal Fedformer prediction model to obtain load prediction values for the load prediction samples comprises:
preprocessing the load prediction sample to obtain load prediction characteristics of the load prediction sample;
and inputting the load prediction characteristics of the load prediction samples into the optimal Fedformer prediction model to obtain the load prediction values of the load prediction samples.
7. The method of claim 1, wherein prior to said inputting the preprocessed historical sample dataset into a target Fedformer predictive model, the method further comprises:
acquiring historical load time series data corresponding to the historical sample data set, and performing sliding window processing of a specified proportion on the historical load time series data to acquire a training sample set; each training sample in the training sample set comprises an input load sequence and a predicted output load sequence;
Carrying out load local difference calculation on the training sample, and obtaining a local difference calculation result;
and according to the local difference calculation result, assigning a loss weight to each training sample so as to improve the loss function of the target Fedformer prediction model.
8. The method of claim 7, wherein assigning a loss weight to each of the training samples based on the local variance calculation comprises:
and reducing the loss weight of the training samples with larger absolute values of local differences.
9. A short-term predictive device for electrical load, the device comprising:
the historical sample data set construction module is used for constructing a historical sample data set of the power system; the historical sample data set comprises historical load data, historical weather forecast data and daily load curve types;
the preprocessing result acquisition module is used for preprocessing the historical sample data set and acquiring a preprocessing result; the preprocessing comprises feature coding processing and normalization processing;
the training and optimizing module is used for inputting the preprocessed historical sample data set into a target Fedformer prediction model, and training and optimizing the target Fedformer prediction model to obtain an optimal Fedformer prediction model;
And the load prediction value acquisition module is used for acquiring a load prediction sample of the power system, and predicting the load prediction sample through the optimal Fedformer prediction model so as to acquire a load prediction value of the load prediction sample.
10. A computer device comprising a processor and a memory having stored therein at least one instruction that is loaded and executed by the processor to implement a method of short-term prediction of electrical load as claimed in any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310824056.8A CN116845874A (en) | 2023-07-06 | 2023-07-06 | Short-term prediction method and device for power load |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310824056.8A CN116845874A (en) | 2023-07-06 | 2023-07-06 | Short-term prediction method and device for power load |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116845874A true CN116845874A (en) | 2023-10-03 |
Family
ID=88170291
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310824056.8A Pending CN116845874A (en) | 2023-07-06 | 2023-07-06 | Short-term prediction method and device for power load |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116845874A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117114056A (en) * | 2023-10-25 | 2023-11-24 | 城云科技（中国）有限公司 | Power load prediction model, construction method and device thereof and application |
| CN117318055A (en) * | 2023-12-01 | 2023-12-29 | 山东理工昊明新能源有限公司 | Power load prediction model processing method and device, electronic equipment and storage medium |
| CN117810997A (en) * | 2024-03-01 | 2024-04-02 | 南京师范大学 | Short-term wind power prediction method based on space-time correlation |
-
2023
- 2023-07-06 CN CN202310824056.8A patent/CN116845874A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117114056A (en) * | 2023-10-25 | 2023-11-24 | 城云科技（中国）有限公司 | Power load prediction model, construction method and device thereof and application |
| CN117114056B (en) * | 2023-10-25 | 2024-01-09 | 城云科技（中国）有限公司 | Power load prediction model, construction method and device thereof and application |
| CN117318055A (en) * | 2023-12-01 | 2023-12-29 | 山东理工昊明新能源有限公司 | Power load prediction model processing method and device, electronic equipment and storage medium |
| CN117318055B (en) * | 2023-12-01 | 2024-03-01 | 山东理工昊明新能源有限公司 | Power load prediction model processing method and device, electronic equipment and storage medium |
| CN117810997A (en) * | 2024-03-01 | 2024-04-02 | 南京师范大学 | Short-term wind power prediction method based on space-time correlation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110610280B (en) | Short-term prediction method, model, device and system for power load | |
| CN109492830B (en) | Mobile pollution source emission concentration prediction method based on time-space deep learning | |
| CN116845874A (en) | Short-term prediction method and device for power load | |
| Gilik et al. | Air quality prediction using CNN+ LSTM-based hybrid deep learning architecture | |
| Alencar et al. | Hybrid approach combining SARIMA and neural networks for multi-step ahead wind speed forecasting in Brazil | |
| CN107506868B (en) | Method and device for predicting short-time power load | |
| CN111160626B (en) | Power load time sequence control method based on decomposition fusion | |
| CN112488396A (en) | Wavelet transform-based electric power load prediction method of Holt-Winters and LSTM combined model | |
| Yu et al. | Improved convolutional neural network‐based quantile regression for regional photovoltaic generation probabilistic forecast | |
| CN115148019A (en) | Early warning method and system based on holiday congestion prediction algorithm | |
| CN113671421A (en) | Transformer state evaluation and fault early warning method | |
| CN114091361B (en) | Weather event-based transform model construction method | |
| Arpogaus et al. | Short-term density forecasting of low-voltage load using bernstein-polynomial normalizing flows | |
| Wu et al. | A novel bayesian additive regression trees ensemble model based on linear regression and nonlinear regression for torrential rain forecasting | |
| Huang et al. | A decomposition‐based multi‐time dimension long short‐term memory model for short‐term electric load forecasting | |
| CN115713044B (en) | Method and device for analyzing residual life of electromechanical equipment under multi-condition switching | |
| CN116151449A (en) | Short-term prediction method and device for power load | |
| Abdallah et al. | A vector autoregressive methodology for short-term weather forecasting: tests for Lebanon | |
| Ciaburro | Time series data analysis using deep learning methods for smart cities monitoring | |
| CN116865254A (en) | Power load index prediction method, system, equipment and medium | |
| CN114757096B (en) | Bridge temperature prediction method, device, equipment and medium based on NARX neural network | |
| CN115965160A (en) | Data center energy consumption prediction method and device, storage medium and electronic equipment | |
| CN111950752A (en) | Photovoltaic power station generating capacity prediction method, device and system and storage medium thereof | |
| CN114065996A (en) | Traffic flow prediction method based on variational self-coding learning | |
| Wang et al. | Short‐Term Load Forecasting Based on VMD and Combined Deep Learning Model |
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 |