CN114091776A - K-means-based multi-branch AGCNN short-term power load prediction method - Google Patents

K-means-based multi-branch AGCNN short-term power load prediction method Download PDF

Info

Publication number
CN114091776A
CN114091776A CN202111435479.8A CN202111435479A CN114091776A CN 114091776 A CN114091776 A CN 114091776A CN 202111435479 A CN202111435479 A CN 202111435479A CN 114091776 A CN114091776 A CN 114091776A
Authority
CN
China
Prior art keywords
data
power load
short
agcnn
branch
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202111435479.8A
Other languages
Chinese (zh)
Inventor
樊江川
李秋燕
于昊正
李科
全少理
郭勇
马杰
郭新志
孙义豪
皇甫霄文
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.)
Economic and Technological Research Institute of State Grid Henan Electric Power Co Ltd
Original Assignee
Economic and Technological Research Institute of State Grid Henan Electric Power 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 Economic and Technological Research Institute of State Grid Henan Electric Power Co Ltd filed Critical Economic and Technological Research Institute of State Grid Henan Electric Power Co Ltd
Priority to CN202111435479.8A priority Critical patent/CN114091776A/en
Publication of CN114091776A publication Critical patent/CN114091776A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Electricity, gas or water supply

Abstract

The invention relates to a K-means-based multi-branch AGCNN short-term power load prediction method, which comprises the following steps: importing power load historical data of a platform area, and preprocessing the data by adopting a K-nearest neighbor algorithm to obtain normalized data; carrying out K-means clustering on the normalized data to obtain the load type of the distribution room; constructing a network structure of a multi-branch AGCNN short-term power load prediction model, and preprocessing samples of different types of loads; in a multi-branch AGCNN short-term power load prediction model, network training is carried out on load data by adopting an Adam algorithm; and predicting the power load of the distribution area based on the multi-branch AGCNN short-term power load prediction model to obtain a load prediction result. The method can deeply mine the characteristics provided by the data, reduce the complexity of the characteristics and improve the prediction precision of the short-term power load of the transformer area.

Description

K-means-based multi-branch AGCNN short-term power load prediction method
Technical Field
The invention belongs to the technical field of power load prediction, and particularly relates to a K-means-based multi-branch AGCNN short-term power load prediction method.
Background
The power load prediction provides effective support for planning, running and scheduling of the power system. However, industrial electricity, commercial electricity and residential electricity in the power distribution network have the characteristics of high growth rate and diversified load characteristics, and the load characteristics of different users are significantly different due to different production processes and electricity consumption peak-valley periods among different industries, so that the change rule of distribution load in the power supply partition is obviously different along with different factors such as electricity consumption areas and user electricity consumption modes. The generalized prediction model adopted for each power supply partition can cause insufficient prediction accuracy due to the fact that the power consumption characteristics of different types of users cannot be drawn, and even can cause training divergence; meanwhile, the number of distribution transformers in one power supply partition is hundreds and thousands, and the work of performing prediction modeling on the load curve of a single distribution transformer is difficult to develop.
At present, the load prediction of a power system can be divided into short-term prediction, medium-term prediction and long-term prediction according to a prediction time scale, wherein the short-term load prediction of a distribution network platform area is an important technology for the economic and efficient operation of a distribution network, and the short-term load prediction mainly infers the change trend of a load in a short term in the future according to historical operation data of the distribution network. The existing short-term load prediction method comprises the following steps: the load is divided adaptively according to the temperature of seasonal variation, and the load data is predicted by adopting an optimized outlier robust extreme learning machine algorithm, so that the short-term load prediction effect is improved; and performing multistage clustering on the platform load based on the power consumption data, and constructing a load prediction model based on the impulse neural network to realize accurate classification prediction of the load. The two methods directly predict the mixed load, and the power utilization characteristic rule of the unused transformer area cannot be mined, so that the prediction precision is insufficient. The CNN-LSTM short-term power load prediction method based on the attention mechanism is adopted, the combination of the CNN and the LSTM can reduce the loss of historical information, but for only processing historical time sequence data, the combination of the CNN and the LSTM is too complex, so that the calculation load is increased, the prediction precision is further improved based on the dual-channel GCNN model provided on the basis, but the adopted basic convolutional neural network is not optimized for time sequence characteristics, and therefore the accuracy is not high. In order to solve the problems, the invention provides the K-means-based multi-branch AGCNN short-term power load prediction method which can deeply mine data characteristics, reduce the complexity of the characteristics, fully utilize the long-time dependency of historical data and improve the prediction accuracy.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, and provides a K-means-based multi-branch AGCNN short-term power load prediction method, which can deeply mine the characteristics provided by data, reduce the complexity of the characteristics and improve the prediction accuracy.
The technical scheme adopted by the invention is as follows: a K-means-based multi-branch AGCNN short-term power load prediction method comprises the following steps:
s1: importing power load historical data of a platform area, and preprocessing the data by adopting a K-nearest neighbor algorithm to obtain normalized data;
s2: carrying out K-means clustering on the normalized data to obtain the load type of the distribution room;
s3: constructing a network structure of a multi-branch AGCNN short-term power load prediction model, and preprocessing samples of different types of loads;
s4: in a multi-branch AGCNN short-term power load prediction model, network training is carried out on load data by adopting an Adam algorithm;
s5: and predicting the power load of the distribution area based on the multi-branch AGCNN short-term power load prediction model to obtain a load prediction result.
Specifically, in the step S1, the preprocessing of the data by using the K-nearest neighbor algorithm specifically includes: and for the historical data of the power load of the platform area, filling data in the month with missing data but not missing in the whole month by adopting a K-nearest neighbor algorithm, supplementing the missing data in the whole month by using the data of the adjacent month in the area or the same month in the adjacent area, and normalizing the data after the missing value processing.
Specifically, when the K-neighbor algorithm is adopted to fill missing data of power load historical data of the platform area, the distance of the K-neighbor algorithm is inversely proportional to the weight, the weighted average value is used as a missing value, and a vector to be filled is set
Figure BDA0003381626600000031
Wherein
Figure BDA0003381626600000032
As an integral part of the vector,
Figure BDA0003381626600000033
is a missing part in the vector, then
Figure BDA0003381626600000034
Has a fill-in value of
Figure BDA0003381626600000035
The formula is as follows:
Figure BDA0003381626600000036
in the formula, DijExpressing the distance between the vector i and the vector j, normalizing the filled complete data, and constraining the normalized complete data to be 0,1 by the data normalization used in the experiment]Within the range.
Specifically, in step S2, when classifying the normalized data by using K-means clustering, the characteristics of fast increase and diversified load characteristics are presented for the industrial power consumption, the commercial power consumption, the agricultural power consumption, and the residential power consumption in the distribution room, the load of the distribution room is classified by using K-means clustering, and if the clustering center does not change before and after one iteration, it is said that convergence has occurred, and the calculation formula of clustering division is as follows:
Figure BDA0003381626600000037
in the formula, miIs ciCluster center of class, xqIs a subject ofiThe samples in the class divide the platform area into four load types, namely commercial load, industrial load, agricultural load and residential load.
Specifically, the step S3 specifically includes: introducing an attention mechanism, constructing a network structure of a multi-branch AGCNN short-term power load prediction model by using a power load characteristic matrix, wherein the network structure comprises an input layer, a gated convolution layer, an attention layer and an output layer which are sequentially connected, and classifying and preprocessing samples of different types of loads according to a cycle, a daily cycle and a neighboring cycle, wherein after the hidden depth time characteristics of three branches are obtained, the samples are weighted by a channel attention module, and the specific expression is as follows:
Figure BDA0003381626600000041
in the formula, Pooling is global Pooling, MLP is full connection layer, Softmax is activation function,
the model is trained between layers using batch normalization operations, represented as follows:
Zl,(n|d|w)＝BN(Hl-1,(n|d|w)),l∈even
in the formula (I), the compound is shown in the specification,
Figure BDA0003381626600000042
specifically, in step S4, the load data is first divided into a training set, a testing set and a verification set, which are 80%, 15% and 5%, respectively.
Specifically, in step S4, in each training period of the network structure of the multi-branch AGCNN short-term power load prediction model, training set data is selected and input into the network for network training, and an Adam algorithm is adopted to optimize the training set data, so as to obtain an optimized multi-branch AGCNN short-term power load prediction model, where the training process of the network training is as follows:
in the training process, the mean square error is used as a loss function, and a small-batch Adam algorithm is used as an optimizer, wherein the mean square error loss function is expressed as follows:
Figure BDA0003381626600000043
wherein M is the number of samples to be processed in batch,
Figure BDA0003381626600000044
is the predicted value of the model and y is the true value.
Specifically, in step S5, the power load at multiple times of the future day or week of the distribution area is predicted by using the multi-branch AGCNN short-term power load prediction model to obtain a load prediction result, and the performance of the evaluation model is verified by calculating the prediction accuracy of the multi-branch AGCNN short-term power load prediction model using the test set.
The invention has the beneficial effects that: the method takes the characteristics of different load types of the user area into full consideration, adopts a K-means cluster analysis method and a multi-branch AGCNN short-term power load prediction model prediction method, can fully utilize the long-term dependency characteristics of historical data, screens extracted depth characteristics by adopting an attention mechanism, can reduce the complexity of the characteristics, can also deeply mine the characteristics provided by the data, and improves the prediction precision of the short-term power load of the user area.
Drawings
FIG. 1 is a flow chart of the steps of the present invention;
fig. 2 is a structural diagram of a short-term power load prediction model of a multi-branch AGCNN according to the present invention.
Detailed Description
The technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments of the present invention, belong to the protection scope of the present invention, and are specifically described below with reference to the embodiments.
As shown in fig. 1, the present invention comprises the steps of:
s1: importing historical data of the power load of the platform area, and preprocessing the data by adopting a K-nearest neighbor algorithm to obtain normalized data.
The data are preprocessed by adopting a K-nearest neighbor algorithm, specifically: for the historical data of the power load of the platform area, a K-nearest neighbor algorithm is adopted to fill the data of the month with missing data but not missing in the whole month, the data missing in the whole month is supplemented by the data of the adjacent month of the area or the same month of the adjacent area, and the normalization processing is carried out on the data after the missing value processing;
the core idea of the K-nearest neighbor algorithm is that for a new sample, the K nearest neighbors of the sample are first found by euclidean distance, and if most of the neighbors belong to the same class as the sample, the sample is classified into the class, so as to give a training data set. When the idea of the K-nearest neighbor algorithm is used for regression, the K-nearest neighbor algorithm can be used for filling missing values. The known data of the neighbors are weighted and then assigned, and the K neighbors have different influences on the sample distance because the K neighbors are different from the sample distance.
The Minkowski distance, the Ming's distance, is the distance between points generated in the clustering analysis and is based on a quantitative variationDistance measure of quantity, note xi＝(xi1,xi2,...,xip) And xj＝(xj1,xj2,...,xjp) Is the observed value for sample i and sample j, then the Ming's distance is defined as:
Figure BDA0003381626600000061
when q is 2, the second-order minkowski distance is called euclidean distance or euclidean distance, and is expressed as follows:
Figure BDA0003381626600000062
when the K-neighbor algorithm is adopted to fill missing data of the power load historical data of the transformer area, the distance of the K-neighbor algorithm is in inverse proportion to the weight, the weighted average value is used as a missing value, and a vector to be filled is set
Figure BDA0003381626600000063
Wherein
Figure BDA0003381626600000064
As an integral part of the vector,
Figure BDA0003381626600000065
is a missing part in the vector, then
Figure BDA0003381626600000066
Has a fill-in value of
Figure BDA0003381626600000067
The formula is as follows:
Figure BDA0003381626600000068
in the formula, DijRepresenting the distance between the vector i and the vector j, normalizing the filled complete dataChemical treatment, normalization of data used in experiments to constrain it to [0, 1%]Within the range.
S2: and carrying out K-means clustering on the normalized data to obtain the load type of the distribution room.
Because the load composition information of each transformer substation in the power supply subarea is different, the difference of the economic development conditions among the areas is large, different transformer areas can present different load growth trends, and the difference can exist on power consumption curves of month, quarter and year, for example, on the power consumption of month, the transformer substations of which the manufacturing industry such as chemical industry, textile industry and the like dominates the load structure present a curve of stable change; the power utilization curves of the users obviously influenced by the alternation of production cycle and season in agriculture, building industry and the like are in a multi-peak type; the users with summer peak and winter valley show unimodal variation curves. The clustering practice of the monthly power consumption curve of the transformer substation is to effectively divide the monthly power consumption curve according to industry constitution, so that the characteristics of high growth speed and diversified load characteristics are presented for industrial power consumption, commercial power consumption, agricultural power consumption and residential power consumption in a transformer area, K-means clustering is adopted to classify the transformer area load, the clustering cluster in a K-means clustering algorithm is composed of objects close to each other, so that the obtained compact and independent clustering cluster is taken as a final target, the algorithm firstly randomly selects any K objects as the center of initial clustering, initially represents one clustering cluster, in each iteration, each object remaining in the data set is assigned to the nearest cluster again according to the distance between each object and each clustering cluster, after all data objects are inspected, one iteration operation is completed, and a new clustering center is calculated, if the clustering center is not changed before and after one iteration, convergence is shown, and the calculation formula for obtaining the optimal clustering division through iteration by adopting the K-means algorithm is expressed as follows:
Figure BDA0003381626600000071
in the formula, miIs ciCluster center of class, xqIs a subject ofiSamples in class, which divide the table into commercial load and industrial loadIndustrial load, agricultural load and residential load.
S3: and constructing a network structure of a multi-branch AGCNN short-term power load prediction model, and preprocessing samples of different types of loads.
In order to fully mine the long-time and short-time characteristics of time series data, facilitate the characteristic extraction of the time series data, further refine the data periodicity relation, introduce an attention mechanism, and construct a network structure of a multi-branch AGCNN short-term power load prediction model by using a power load characteristic matrix, wherein the network structure comprises an input layer, a gate control convolution layer, an attention layer and an output layer which are sequentially connected, a structural schematic diagram of the network structure is shown in FIG. 2, samples of different types of loads are classified and preprocessed according to a cycle, a day cycle and a neighbor cycle, and the process is as follows:
1) sample pretreatment
Because the daily periodicity and the weekly periodicity of the load data are obvious, the weekly period, the daily period and the neighbor characteristics can be respectively input into a multi-branch network structure, the load power of the next week is predicted by using the load power of every beta week, the load power of the next day is predicted by using the load power of every alpha day, and the load power of the next node is predicted by using the neighbor characteristics as the load power of every m points, which is expressed as follows:
Figure BDA0003381626600000081
in the formula, Zw、Zd、ZnThe characteristics are respectively a week period characteristic, a day period characteristic and a neighbor characteristic, T is the number of time nodes in one day, and alpha, beta and m are positive integers.
Because the output of the node reaches a saturation region due to the excessively large or small data input quantity, the load value is normalized at the input layer, and the expression is as follows:
Figure BDA0003381626600000082
in the formula, zminData optimization for training setSmall value, zmaxFor the maximum value of the data in the training set,
Figure BDA0003381626600000083
for normalized values, z is the raw data value.
2) Gate control unit
Compared with the LSTM, the structure of the GCNN can be operated in parallel, the calculation time is shorter, and the expression is as follows:
Figure BDA0003381626600000091
in the formula, Zl,w、Zl,d、Zl,nRespectively a cycle characteristic channel, a day cycle characteristic channel and a neighbor characteristic input channel to obtain a first layer of input characteristics, Conv1For one-dimensional convolution operations along the time axis, Conv2Is and Conv1And performing another one-dimensional convolution operation with the same convolution kernel dimension, wherein Sigmoid is a gating unit activation function.
Batch standardization operation, namely BN operation, is used between layers, so that the model does not have gradient disappearance during training, the convergence efficiency is improved, and the expression is as follows:
Zl,(n|d|w)＝BN(Hl-1,(n|d|w)),l∈even
in the formula (I), the compound is shown in the specification,
Figure BDA0003381626600000092
3) attention mechanism
The attention mechanism is a model for simulating the attention of the human brain, reduces or even ignores the attention of the human brain to other parts by taking the attention of the human brain to some important objects as a reference, highlights more key influence factors by endowing different weights to the input characteristics of the model, helps the model to make more accurate judgment, reduces the loss of historical information, strengthens the influence of important information, and finally completes short-term load prediction.
After the depth time characteristics hidden by the three branches are obtained, the three branches are weighted by a channel attention module, the function of the channel attention module is to inhibit redundant hidden characteristics and enhance useful hidden characteristics, and the specific expression is as follows:
Figure BDA0003381626600000093
in the formula, Pooling is global Pooling, MLP is full-link layer, and Softmax is activation function.
S4: in the multi-branch AGCNN short-term power load prediction model, load data are subjected to network training by adopting an Adam algorithm.
Firstly, dividing load data into a training set, a testing set and a verification set, wherein the percentage of the training set, the testing set and the verification set is respectively 80%, 15% and 5%, then selecting the training set data to be input into a network for network training in each training period of a network structure of the multi-branch AGCNN short-term power load prediction model, and optimizing the network training iteration by adopting an Adam algorithm after the network is started to obtain an optimized multi-branch AGCNN short-term power load prediction model, wherein the training process of the network training is as follows:
after a prediction model is built, a large number of training samples are adopted for training, the mean square error is adopted as a loss function in the training process, a small-batch Adam algorithm is adopted as an optimizer, and the mean square error loss function is expressed as follows:
Figure BDA0003381626600000101
wherein M is the number of samples to be processed in batch,
Figure BDA0003381626600000102
is the predicted value of the model and y is the true value.
S5: and predicting the power load of the distribution area based on the multi-branch AGCNN short-term power load prediction model to obtain a load prediction result.
And predicting the power loads of the station area at a plurality of moments in one day or one week in the future by using the multi-branch AGCNN short-term power load prediction model to obtain a load prediction result, and calculating the prediction precision of the multi-branch AGCNN short-term power load prediction model by using the test set to verify the performance of the evaluation model so as to verify the effectiveness of the method provided by the invention.
It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential attributes thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.

Claims (8)

1. A K-means-based multi-branch AGCNN short-term power load prediction method is characterized by comprising the following steps:
s1: importing power load historical data of a platform area, and preprocessing the data by adopting a K-nearest neighbor algorithm to obtain normalized data;
s2: carrying out K-means clustering on the normalized data to obtain the load type of the distribution room;
s3: constructing a network structure of a multi-branch AGCNN short-term power load prediction model, and preprocessing samples of different types of loads;
s4: in a multi-branch AGCNN short-term power load prediction model, network training is carried out on load data by adopting an Adam algorithm;
s5: and predicting the power load of the distribution area based on the multi-branch AGCNN short-term power load prediction model to obtain a load prediction result.
2. The method for predicting the short-term power load of the multi-branch AGCNN based on K-means as claimed in claim 1, wherein the preprocessing of the data by the K-nearest neighbor algorithm in step S1 specifically comprises: and for the historical data of the power load of the platform area, filling data in the month with missing data but not missing in the whole month by adopting a K-nearest neighbor algorithm, supplementing the missing data in the whole month by using the data of the adjacent month in the area or the same month in the adjacent area, and normalizing the data after the missing value processing.
3. The method of claim 2, wherein the K-means based multi-branch AGCNN short term power load prediction method comprises: when the K-neighbor algorithm is adopted to fill missing data of the power load historical data of the transformer area, the distance of the K-neighbor algorithm is in inverse proportion to the weight, the weighted average value is used as a missing value, and a vector to be filled is set
Figure FDA0003381626590000011
Wherein
Figure FDA0003381626590000012
As an integral part of the vector,
Figure FDA0003381626590000013
is a missing part in the vector, then
Figure FDA0003381626590000016
Has a fill-in value of
Figure FDA0003381626590000015
The formula is as follows:
Figure FDA0003381626590000021
in the formula, DijExpressing the distance between the vector i and the vector j, normalizing the filled complete data, and constraining the normalized complete data to be 0,1 by the data normalization used in the experiment]Within the range.
4. The method of claim 1, wherein the K-means based multi-branch AGCNN short term power load prediction method comprises: in step S2, when the normalized data is classified by using K-means clustering, the load of the distribution room is classified by using K-means clustering, and if the clustering center does not change before and after one iteration, it indicates that convergence has occurred, and the calculation formula of clustering division is as follows:
Figure FDA0003381626590000022
in the formula, miIs ciCluster center of class, xqIs a subject ofiThe samples in the class are to be classified into,
aiming at the characteristics that the industrial electricity consumption, the commercial electricity consumption, the agricultural electricity consumption and the residential electricity consumption in a transformer area have high growth speed and diversified load characteristics, the transformer area is divided into four load types, namely commercial load, industrial load, agricultural load and residential load.
5. The method for predicting the short-term power load of the multi-branch AGCNN based on K-means as claimed in claim 1, wherein the step S3 is specifically as follows: introducing an attention mechanism, constructing a network structure of a multi-branch AGCNN short-term power load prediction model by using a power load characteristic matrix, wherein the network structure comprises an input layer, a gated convolution layer, an attention layer and an output layer which are sequentially connected, and classifying and preprocessing samples of different types of loads according to a cycle, a daily cycle and a neighboring cycle, wherein after the hidden depth time characteristics of three branches are obtained, the samples are weighted by a channel attention module, and the specific expression is as follows:
Figure FDA0003381626590000031
in the formula, Pooling is global Pooling, MLP is full connection layer, Softmax is activation function,
the model is trained between layers using batch normalization operations, represented as follows:
Zl,(n|d|w)＝BN(Hl-1,(n|d|w)),l∈even
in the formula (I), the compound is shown in the specification,
Figure FDA0003381626590000032
6. the method of claim 1, wherein the K-means based multi-branch AGCNN short term power load prediction method comprises: in step S4, the load data is first divided into training set, testing set and verification set, which are 80%, 15% and 5%, respectively.
7. The K-means based multi-branch AGCNN short-term power load prediction method according to claim 6, wherein: the step S4 is specifically that, in each training period of the network structure of the multi-branch AGCNN short-term power load prediction model, training set data is selected and input into the network for network training, and an Adam algorithm is adopted to optimize the training set data, so as to obtain an optimized multi-branch AGCNN short-term power load prediction model, where the training process of the network training is as follows:
in the training process, the mean square error is used as a loss function, and a small-batch Adam algorithm is used as an optimizer, wherein the mean square error loss function is expressed as follows:
Figure FDA0003381626590000033
wherein M is the number of samples to be processed in batch,
Figure FDA0003381626590000034
is the predicted value of the model and y is the true value.
8. The K-means based multi-branch AGCNN short-term power load prediction method according to claim 6, wherein: in step S5, the multi-branch AGCNN short-term power load prediction model is used to predict the power loads of the distribution area at multiple times of the future day or week to obtain a load prediction result, and the test set is used to calculate the prediction accuracy of the multi-branch AGCNN short-term power load prediction model to verify the performance of the evaluation model.
CN202111435479.8A 2021-11-29 2021-11-29 K-means-based multi-branch AGCNN short-term power load prediction method Pending CN114091776A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111435479.8A CN114091776A (en) 2021-11-29 2021-11-29 K-means-based multi-branch AGCNN short-term power load prediction method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111435479.8A CN114091776A (en) 2021-11-29 2021-11-29 K-means-based multi-branch AGCNN short-term power load prediction method

Publications (1)

Publication Number Publication Date
CN114091776A true CN114091776A (en) 2022-02-25

Family

ID=80305772

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111435479.8A Pending CN114091776A (en) 2021-11-29 2021-11-29 K-means-based multi-branch AGCNN short-term power load prediction method

Country Status (1)

Country Link
CN (1) CN114091776A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116632842A (en) * 2023-07-26 2023-08-22 国网山东省电力公司信息通信公司 Clustering characteristic-based method and system for predicting distribution type photovoltaic load probability of platform
CN117113159A (en) * 2023-10-23 2023-11-24 国网山西省电力公司营销服务中心 Deep learning-based power consumer side load classification method and system

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116632842A (en) * 2023-07-26 2023-08-22 国网山东省电力公司信息通信公司 Clustering characteristic-based method and system for predicting distribution type photovoltaic load probability of platform
CN116632842B (en) * 2023-07-26 2023-11-10 国网山东省电力公司信息通信公司 Clustering characteristic-based method and system for predicting distribution type photovoltaic load probability of platform
CN117113159A (en) * 2023-10-23 2023-11-24 国网山西省电力公司营销服务中心 Deep learning-based power consumer side load classification method and system

Similar Documents

Publication Publication Date Title
CN113962364B (en) Multi-factor power load prediction method based on deep learning
CN110674999A (en) Cell load prediction method based on improved clustering and long-short term memory deep learning
CN105678407A (en) Daily electricity consumption prediction method based on artificial neural network
CN108596242B (en) Power grid meteorological load prediction method based on wavelet neural network and support vector machine
CN101782743A (en) Neural network modeling method and system
US20210326696A1 (en) Method and apparatus for forecasting power demand
CN111815035A (en) Short-term load prediction method fusing morphological clustering and TCN-Attention
CN114091776A (en) K-means-based multi-branch AGCNN short-term power load prediction method
CN110598854A (en) GRU model-based transformer area line loss rate prediction method
CN112149890A (en) Comprehensive energy load prediction method and system based on user energy label
CN113554466A (en) Short-term power consumption prediction model construction method, prediction method and device
CN106779219A (en) A kind of electricity demand forecasting method and system
CN113255900A (en) Impulse load prediction method considering improved spectral clustering and Bi-LSTM neural network
CN116845889B (en) Hierarchical hypergraph neural network-based power load prediction method
CN110163444A (en) A kind of water demand prediction method based on GASA-SVR
CN109754122A (en) A kind of Numerical Predicting Method of the BP neural network based on random forest feature extraction
Chen et al. MultiCycleNet: multiple cycles self-boosted neural network for short-term electric household load forecasting
CN111882114B (en) Short-time traffic flow prediction model construction method and prediction method
CN114119273A (en) Park comprehensive energy system non-invasive load decomposition method and system
CN113591368A (en) Comprehensive energy system multi-energy load prediction method and system
CN114611757A (en) Electric power system short-term load prediction method based on genetic algorithm and improved depth residual error network
CN116826745B (en) Layered and partitioned short-term load prediction method and system in power system background
CN113887809A (en) Power distribution network supply and demand balance method, system, medium and computing equipment under double-carbon target
CN109214610A (en) A kind of saturation Methods of electric load forecasting based on shot and long term Memory Neural Networks
CN109146553A (en) Spot Price forecasting system and its method based on multi-density cluster and multicore SVM

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