CN116523681A - Load decomposition method and device for electric automobile, electronic equipment and storage medium - Google Patents

Load decomposition method and device for electric automobile, electronic equipment and storage medium Download PDF

Info

Publication number
CN116523681A
CN116523681A CN202310491742.8A CN202310491742A CN116523681A CN 116523681 A CN116523681 A CN 116523681A CN 202310491742 A CN202310491742 A CN 202310491742A CN 116523681 A CN116523681 A CN 116523681A
Authority
CN
China
Prior art keywords
power consumption
load
sequence
data
convolution
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
CN202310491742.8A
Other languages
Chinese (zh)
Inventor
尹雁和
李宾
林雄锋
彭石丰
阮志杰
李国号
余俊杰
白一鸣
江清楷
钟毅
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.)
Guangdong Power Grid Co Ltd
Zhongshan Power Supply Bureau of Guangdong Power Grid Co Ltd
Original Assignee
Guangdong Power Grid Co Ltd
Zhongshan Power Supply Bureau of Guangdong Power Grid 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 Guangdong Power Grid Co Ltd, Zhongshan Power Supply Bureau of Guangdong Power Grid Co Ltd filed Critical Guangdong Power Grid Co Ltd
Priority to CN202310491742.8A priority Critical patent/CN116523681A/en
Publication of CN116523681A publication Critical patent/CN116523681A/en
Pending legal-status Critical Current

Links

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
    • G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Electricity, gas or water supply
    • 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
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06Q50/40—

Abstract

The invention discloses a load decomposition method, a device, electronic equipment and a storage medium of an electric automobile, which are used for solving the technical problem of low load decomposition accuracy in the existing load decomposition method. The method comprises the following steps: the method comprises the steps of obtaining a power sequence data set of a power utilization area, dividing the power sequence data set into a combined training set and a load decomposition test set according to a certain proportion, respectively carrying out data preprocessing on the combined training set and the load decomposition test set to obtain a corresponding power consumption training set and a power consumption test set so as to improve model training accuracy, establishing an electric automobile load decomposition model based on a VGG-16 convolutional neural network, carrying out model training by adopting the power consumption training set and the power consumption test set, obtaining total load power consumption information of the power utilization area, and inputting the total load power consumption information into the trained electric automobile load decomposition model to carry out load decomposition, so that an electric automobile power sequence is extracted rapidly, and the load decomposition accuracy is improved.

Description

Load decomposition method and device for electric automobile, electronic equipment and storage medium
Technical Field
The present invention relates to the field of non-invasive load decomposition technologies, and in particular, to a load decomposition method and apparatus for an electric vehicle, an electronic device, and a storage medium.
Background
In recent years, along with the increasing of the power demand of people and the increasing of the quality of living standard of people, emerging devices such as smart home and photovoltaic power generation gradually appear in the public view and become an important component of daily life of people, and the load characteristics of the emerging devices and the load characteristics of traditional devices often have great differences, so that the load structure becomes more and more complex, and meanwhile, along with the wide use of distributed energy sources (such as electric automobiles and buses), the charging demand of the vehicles also greatly increases the power grid load, and the problems of peak-valley load difference increase, line flow violations and the like may be caused. In order to ensure the normal operation of the power system and the power transmission line, power operation staff needs to monitor the load so as to acquire the power information of the load and help the operation staff to make a demand response plan, and because the nature of the load and the use mode are continuously changed, load modeling is needed to be performed in a targeted manner when the load is monitored, so that the time-varying nature of model parameters is also gradually highlighted.
For load monitoring methods, there are currently two main categories, invasive load monitoring (Intrusive Load Monitoring, ILM) and Non-invasive load monitoring (Non-Intrusive Load Monitoring, NILM). The invasive load monitoring method has the advantages that the invasive load monitoring method is better in safety and expandability than the invasive load monitoring method, and extra equipment and cost are not needed. In the related art, a non-invasive load modeling method based on manual experience analysis is generally adopted for modeling at present, and load decomposition is carried out through an established model so as to realize load monitoring of electric equipment, and by adopting the method, manual intervention is needed, so that the problem of low load decomposition accuracy is easily caused.
Disclosure of Invention
The invention provides a load decomposition method, a device, electronic equipment and a storage medium of an electric automobile, which are used for solving or partially solving the technical problem of low load decomposition accuracy in the existing load decomposition method.
The invention provides a load decomposition method of an electric automobile, which comprises the following steps:
acquiring a power sequence data set of a power utilization area, and dividing the power sequence data set into a combined training set and a load decomposition test set according to a certain proportion;
respectively carrying out data preprocessing on the combined training set and the load decomposition test set to obtain a power consumption training set corresponding to the combined training set and a power consumption test set corresponding to the load decomposition test set;
establishing an electric vehicle load decomposition model based on a VGG-16 convolutional neural network, and training the electric vehicle load decomposition model by adopting the power consumption training set and the power consumption testing set;
and acquiring the total load power consumption information of the electricity utilization area, inputting the total load power consumption information into a trained electric vehicle load decomposition model to carry out load decomposition, and extracting an electric vehicle power sequence of the electricity utilization area.
Optionally, the electric automobile load decomposition model includes the input layer, with 5 convolution blocks and 3 total connecting blocks of input layer series connection in proper order, with total connecting block end-to-end connection's normalization mapping layer, will total load consumption information input to electric automobile load decomposition model after training carries out the load and decomposes, extracts the electric automobile power sequence in power consumption region includes:
inputting the region total power consumption information into an input layer for normalization processing to obtain a to-be-decomposed total power consumption sequence;
transmitting the total power consumption sequence to be decomposed to the convolution block, sequentially carrying out convolution pooling treatment according to a connection sequence, and outputting a total power consumption characteristic diagram;
transmitting the total power consumption characteristic diagram to the full connecting block, and sequentially carrying out nonlinear characteristic integration according to a connecting sequence to obtain an output characteristic sequence diagram;
inputting the output characteristic sequence diagram to the normalization mapping layer for normalization mapping processing, and converting the output characteristic sequence diagram into output probability distribution corresponding to the output characteristic sequence diagram, wherein the output probability distribution represents the probability of load classification of each input load in the region total power consumption information;
and carrying out load decomposition on the total power consumption sequence to be decomposed according to the output probability distribution, and extracting an electric vehicle power sequence corresponding to the electric vehicle in the electricity utilization area.
Optionally, the 5 convolution blocks are a first convolution module, a second convolution module, a third convolution module, a fourth convolution module and a fifth convolution module, and the transmitting the to-be-decomposed total power consumption sequence to the convolution blocks sequentially performs convolution pooling processing according to a connection sequence, and outputs a total power consumption feature map, including:
transmitting the total power consumption sequence to be decomposed to a first convolution module for continuous standard convolution twice, carrying out pooling treatment, outputting a first intermediate feature map to a second convolution module for continuous standard convolution twice, and carrying out pooling treatment to obtain a second intermediate feature map;
and transmitting the second characteristic sequence diagram to a third convolution module for continuous three times of standard convolution, carrying out pooling treatment, outputting a third intermediate characteristic diagram to a fourth convolution module for continuous three times of standard convolution, carrying out pooling treatment, outputting a fourth intermediate characteristic diagram to a fifth convolution module for continuous three times of standard convolution, and carrying out pooling treatment to obtain a total power consumption characteristic diagram.
Optionally, each convolution module includes a plurality of convolution layers, where the convolution layers are used to perform standard convolution processing on a feature map input to the convolution layers, and the step of standard convolution processing includes:
Determining an input feature map, scanning the input feature map through a convolution kernel of the convolution layer, multiplying the input feature map, and adding the input feature map with a receptive field deviation of the convolution layer to obtain an output feature map, wherein a specific calculation formula is as follows:
wherein Z is l Input feature map representing the (l+1) th convolution layer, Z l+1 Output characteristic diagram representing (l+1) th convolution layer, w l+1 The weight coefficient of the (l+1) th convolution layer is represented, b is the receptive field deviation;
the input volume of the input feature map is obtained, and the output volume of the output feature map is calculated according to the input volume, wherein the specific calculation formula is as follows:
wherein V is l The spatial size of the input volume for the (l+1) -th convolution layer, V l+1 The spatial size of the output volume of the (l+1) th convolution layer, k is the kernel size of the neuron in the corresponding convolution layer, p is the zero filling quantity at the edge, s is the step length, and the number of units converted by the filter each time is represented;
and linearly correcting the output characteristic diagram by adopting a ReLU activation function to obtain a corrected output characteristic diagram.
Optionally, the 3 full connecting blocks correspond to a first full connecting block, a second full connecting block and a third full connecting block, the total power consumption feature map is transferred to the full connecting blocks, nonlinear feature integration is sequentially performed according to a connection sequence, and an output feature sequence map is obtained, and the method comprises the following steps:
Inputting the total power consumption characteristic diagram to the first full connecting block for characteristic combination classification processing, outputting a first classification characteristic diagram, linearly correcting the first classification characteristic diagram by adopting a ReLU activation function, and outputting a first corrected classification characteristic diagram to the second full connecting block;
performing feature combination classification processing on the first corrected classification feature map, outputting a second classification feature map, performing linear correction on the second classification feature map by adopting a ReLU activation function, and outputting the second corrected classification feature map to the third full connecting block;
and carrying out feature combination classification processing on the second corrected classification feature map to obtain an output feature sequence map.
Optionally, the load decomposition is performed on the total power consumption sequence to be decomposed according to the output probability distribution, and the extracting of the electric vehicle power sequence corresponding to the electric vehicle in the electricity consumption area includes:
and carrying out load decomposition on the total power consumption sequence to be decomposed by the following calculation formula, and extracting an electric automobile power sequence corresponding to the electric automobile in the electricity utilization area:
wherein P is agg (T) represents the measured value at t= { T on the low-voltage side of the substation 1 ,t 2 ,...,t n A total power consumption sequence to be decomposed at the moment; n represents the total number of sampling points; n represents the load type according to the refinement level; p (P) i (t) is a power sequence corresponding to the i-class load, expressed as:
P i (t)＝{p i,1 ,p i,2 ,...p i,n }
wherein P is i,t Representing the power consumption of the i-th class of load at the sampling point t.
Optionally, before the acquiring the power sequence data set of the power usage region, the method further comprises:
collecting residual load data of different power utilization access points in a power utilization area, wherein the residual load data comprises household load data, business load data and electric vehicle load data;
carrying out combined addition processing on the household load data, the commercial load data and the electric vehicle load data to form a power sequence data set of the power utilization area, wherein the power sequence data set comprises area total power consumption data and electric vehicle power consumption data;
the dividing the power sequence data set into a joint training set and a load decomposition test set according to a certain proportion comprises the following steps:
dividing the regional total power consumption data and the electric vehicle power consumption data into a combined training set and a load decomposition test set according to a certain proportion, wherein the combined training set comprises the electric vehicle power consumption data and a part of the regional total power consumption data, and the load decomposition test set comprises the rest of the regional total power consumption data.
Optionally, the data preprocessing is performed on the combined training set and the load decomposition test set respectively to obtain a power consumption training set corresponding to the combined training set and a power consumption test set corresponding to the load decomposition test set, which includes:
and respectively carrying out substitution-normalization processing on the power sequence data corresponding to the combined training set and the load decomposition test set, wherein the substitution-normalization processing comprises the following steps of:
step S1: for the problem value of the power sequence data, replacing the problem value by adopting the average value of the two data before and after the problem value, wherein the problem value comprises a missing value and an abnormal value;
step S2: mapping the power sequence data from-1 to 1 by adopting a maximum-minimum normalization method, wherein the specific calculation formula is as follows:
wherein x is min Representing the minimum value, x, of the power sequence data max Represents the maximum value of the power sequence data, x represents the power sequence data before normalization processing, x norm Representing the power sequence data after normalization processing;
respectively segmenting the time sequence data corresponding to the combined training set and the load decomposition test set subjected to substitution-normalization processing by adopting a sliding window method, and dividing the time sequence data into a plurality of sequence data segments with preset fixed lengths, wherein each sequence data segment comprises a preset fixed number of sequence data;
And taking the plurality of sequence data segments corresponding to the combined training set as a power consumption training set, and taking the plurality of sequence data segments corresponding to the load decomposition testing set as a power consumption testing set.
Optionally, the power consumption training set includes an electric vehicle power consumption sequence after data preprocessing and a first region total power consumption sequence, the power consumption testing set includes a second region total power consumption sequence after data preprocessing, the power consumption training set and the power consumption testing set are adopted to train the electric vehicle load decomposition model, and the method includes:
performing model training on the electric vehicle load decomposition model by adopting the electric vehicle power consumption sequence and the first region total power consumption sequence;
and testing the electric automobile load decomposition model after model training through the second region total power consumption sequence, and performing model evaluation based on a test result.
The invention also provides a load decomposition device of the electric automobile, which comprises:
the power sequence data set acquisition module is used for acquiring a power sequence data set of a power utilization area and dividing the power sequence data set into a combined training set and a load decomposition test set according to a certain proportion;
The data preprocessing module is used for respectively preprocessing the data of the combined training set and the load decomposition test set to obtain a power consumption training set corresponding to the combined training set and a power consumption test set corresponding to the load decomposition test set;
the electric automobile load decomposition model training module is used for establishing an electric automobile load decomposition model based on a VGG-16 convolutional neural network, and training the electric automobile load decomposition model by adopting the power consumption training set and the power consumption testing set;
the load decomposition module is used for acquiring the total load power consumption information of the electricity utilization area, inputting the total load power consumption information into the trained electric vehicle load decomposition model to carry out load decomposition, and extracting an electric vehicle power sequence of the electricity utilization area.
The invention also provides an electronic device comprising a processor and a memory:
the memory is used for storing program codes and transmitting the program codes to the processor;
the processor is configured to execute the load decomposition method of the electric automobile according to any one of the above according to the instructions in the program code.
The present invention also provides a computer-readable storage medium for storing a program code for executing the load splitting method of an electric vehicle as set forth in any one of the above.
From the above technical scheme, the invention has the following advantages: the method comprises the steps of constructing an electric vehicle load decomposition model by utilizing a VGG-16-based convolutional neural network, realizing non-invasive decomposition of an electric vehicle based on the electric vehicle load decomposition model, specifically, combining and adding residual load data in a power utilization area to form a power sequence data set containing regional overall power consumption and electric vehicle power consumption, dividing the power sequence data set into a training set and a test set according to a certain proportion, preprocessing the data to improve model training accuracy, training the electric vehicle load decomposition model by adopting the preprocessed data set, carrying out load decomposition on the overall power consumption information needing to be subjected to load monitoring by utilizing the trained model, and rapidly extracting a power sequence corresponding to the electric vehicle, thereby realizing measurement of the electric vehicle without influencing the use experience of a user, providing higher decomposition accuracy, improving the load management capability of a power system under the condition of high-proportion electric vehicle load injection, and providing powerful support for realizing efficient and reliable operation of the electric system.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions of the prior art, the drawings which are used in the description of the embodiments or the prior art will be briefly described, it being obvious that the drawings in the description below are only some embodiments of the invention, and that other drawings can be obtained from these drawings without inventive faculty for a person skilled in the art.
Fig. 1 is a schematic flow chart of a method for decomposing load of an electric vehicle according to an embodiment of the present invention;
fig. 2 is a schematic structural diagram of an electric vehicle load decomposition model based on a VGG-16 convolutional neural network according to an embodiment of the present invention;
fig. 3 is a flowchart illustrating steps of a load decomposition method of an electric vehicle according to an embodiment of the present invention;
fig. 4 is a schematic diagram of a load type structure of different power utilization access points in a power utilization area according to an embodiment of the present invention;
fig. 5 is a step flowchart of a method for performing load decomposition by using an electric vehicle load decomposition model according to an embodiment of the present invention;
fig. 6 is a block diagram of a load splitting device of an electric vehicle according to an embodiment of the present invention.
Detailed Description
The embodiment of the invention provides a load decomposition method and device of an electric automobile, electronic equipment and a storage medium, which are used for solving or partially solving the technical problem of low load decomposition accuracy in the existing load decomposition method.
In order to make the objects, features and advantages of the present invention more comprehensible, the technical solutions in the embodiments of the present invention are described in detail below with reference to the accompanying drawings, and it is apparent that the embodiments described below are only some embodiments of the present invention, but not all embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
As an example, with the widespread use of distributed energy sources (such as electric vehicles and buses), the corresponding charging demands also greatly increase the load of the power grid, which may cause problems of increased peak-to-valley load difference, line flow violations, etc., the popularization of electric vehicles brings new load management challenges to the electric power system, and effective monitoring and management of the load of the electric vehicles are required to ensure the normal operation of the electric power system and the transmission line. In the related art, a non-invasive load modeling method based on manual experience analysis is generally adopted for modeling at present, and load decomposition is carried out through an established model so as to realize load monitoring of electric equipment, and in the mode, manual intervention is needed, so that the problem of low accuracy of load decomposition is easily caused, the existing non-invasive load decomposition method is mostly applied to household appliances, new energy sources, electric automobiles and the like are not considered, and for load decomposition of the electric automobiles, the traditional load decomposition method needs to carry out invasive measurement on the electric automobiles, so that the use experience of users is influenced, and the problems of potential safety hazards, high cost and the like exist.
Therefore, one of the core inventions of the embodiments of the present invention is: the method comprises the steps of constructing an electric vehicle load decomposition model by utilizing a convolutional neural network based on VGG-16 (Visual Geometry Group-16, a 16-layer visual geometry group), realizing non-invasive decomposition of an electric vehicle based on the electric vehicle load decomposition model, specifically combining and adding residual load data in a power utilization area to form a power sequence data set containing the overall power consumption of the area and the power consumption of the electric vehicle, dividing the power sequence data set into a training set and a test set according to a certain proportion and preprocessing the data to improve the model training accuracy, then training the electric vehicle load decomposition model by adopting the preprocessed data set, carrying out load decomposition on the overall power consumption information required to be subjected to load monitoring by utilizing the trained model, and rapidly extracting a power sequence corresponding to the electric vehicle, so that measurement of the electric vehicle is not required, the use experience of a user is not influenced, meanwhile, the higher decomposition accuracy is provided, the load management capability of a power system is improved under the condition of high-proportion electric vehicle load injection, and a powerful support is provided for realizing efficient and reliable operation of the power system.
Referring to fig. 1, a flow chart of an electric vehicle load decomposition method according to an embodiment of the present invention is shown.
Firstly, for a certain electricity consumption area needing to be monitored by electric load, total electricity consumption load information (the total electricity consumption load information comprises electric vehicle load information, household load information and business load information) and electric vehicle load information in the electricity consumption area can be collected respectively, the collected information is processed to obtain a power sequence data set of the electricity consumption area, the power sequence data set is divided into a joint training set and a load decomposition test set according to a certain division mode, and the joint training set can comprise electric vehicle load information and a part of total electricity consumption load information, and the load decomposition test set comprises another part of total electricity consumption load information.
And respectively preprocessing data of the combined training set and the load decomposition test set, wherein for the missing value and the abnormal value of the power sequence data in the combined training set and the load decomposition test set, a front data average value and a rear data average value substitution mode can be adopted to replace data, if the 8 th data in the sequence is missing or abnormal, the 7 th data and the 9 th data are adopted to calculate the average value, the average value is used as a new data value, the original data of the 8 th data are replaced, and then the maximum-minimum normalization method can be adopted to map the power sequence data from-1 to 1 so as to eliminate the influence of different load scales.
The data segmentation processing is performed on the preprocessed power consumption training set and the preprocessed power consumption testing set which are output after the data preprocessing is performed by adopting a window sliding method, and the power consumption training set and the power consumption testing set which comprise a plurality of data segments are obtained, wherein a sliding window 1 represents a data sequence corresponding to a first working period, a sliding window 2 represents a data sequence corresponding to a second working period, and the like, and for convenience of explanation, the data sequences of the preprocessed power consumption training set and the preprocessed power consumption testing set in the figure are only illustrated by taking 0 to 4M-1 as examples, and only the data sequences corresponding to two working periods are illustrated, wherein the step length (namely, the window length) of each sliding is set to be 2M, after the window sliding processing is performed, the first data segment sequence corresponds to [0, 1..2M-2,2M-1 ], and the second data segment sequence corresponds to [ M, M+1..3M-2,3M-1 ], but in the actual processing, the length of each data sequence (namely, the working period of the electric power system) is not limited by the step length of the sliding segment, and the sliding step length of the data can be set as required.
Aiming at the characteristics of power load data, the sliding window method provided by the embodiment of the invention not only considers the correlation between the time sequence relation of the load data and adjacent data, but also utilizes the periodic characteristics of the power load data, and particularly, the method can divide the whole time sequence data into a plurality of periods, the length of each period is equal to the working period of a power system, then for each period, the sliding window method is used for dividing the data in the period into a plurality of windows, and the characteristic extraction and classification of each window are carried out through a convolutional neural network algorithm.
As an alternative embodiment, the sliding window method may include the following steps:
firstly, periodically dividing power load data corresponding to a preprocessing power consumption training set and a preprocessing power consumption testing set according to a working period; for each period, dividing the data in the period into a plurality of windows by utilizing a sliding window method; then, adopting a convolutional neural network (Convolutional Neural Networks, CNN) algorithm to extract and classify the data corresponding to each window; and extracting the correlation characteristics of the data between adjacent periods by using a convolutional neural network algorithm, and incorporating the correlation characteristics into an electric automobile load decomposition model.
It can be seen that the above method can minimize the portion of repeated calculation by utilizing the correlation between the periodic characteristics and the adjacent periods, unlike the conventional sliding window method, thereby improving the calculation efficiency.
And then inputting the power consumption training set into an electric vehicle load decomposition model based on the VGG-16 convolutional neural network to perform model training, and then inputting the power consumption testing set into the trained electric vehicle load decomposition model to perform effect testing to determine a load decomposition model for decomposing the load of the electric vehicle.
And acquiring total load power consumption information required to be subjected to load decomposition in a power consumption area, inputting the total load power consumption information into an electric vehicle load decomposition model subjected to training test to carry out load decomposition, and rapidly extracting an electric vehicle power sequence from the total load power consumption information to realize non-invasive load decomposition of the electric vehicle power sequence.
According to the electric vehicle load decomposition method, the method is combined with the electric vehicle load decomposition model provided by the embodiment of the invention, the residual load data in the electricity utilization area is combined and added to form the power sequence data set containing the overall power consumption of the area and the power consumption of the electric vehicle, the power sequence data set is divided into the training set and the testing set according to a certain proportion and is subjected to data preprocessing, so that the model training accuracy is improved, the preprocessed data set is adopted to train the electric vehicle load decomposition model, the trained model is utilized to carry out load decomposition on the overall power consumption information needing to be subjected to load monitoring, and the power sequence corresponding to the electric vehicle is rapidly extracted, so that the electric vehicle is not required to be measured, the use experience of a user is not affected, the higher decomposition accuracy is provided, the load management capability of the electric vehicle system is improved under the condition of high-proportion electric vehicle load injection, and the effective and reliable operation of the electric vehicle system is provided.
Referring to fig. 2, a schematic structural diagram of an electric vehicle load decomposition model based on a VGG-16 convolutional neural network according to an embodiment of the present invention is shown.
As can be seen from the figure, the electric automobile load decomposition model mainly comprises 5 convolution blocks and 3 full connection blocks, specifically: the first convolution module Conv1, the second convolution module Conv2, the third convolution module Conv3, the fourth convolution module Conv4, the fifth convolution module Conv5, the first full connection block fc6 (full connection), the second full connection block fc7 and the third full connection block fc8.
The first convolution module Conv1 comprises 2 convolution layers with a size of 224×224×64 and a max-pooling layer MP (Max Pooling) with a size of 112×112×128; the second convolution module Conv2 comprises 2 convolution layers with the size of 112×112×128 and a maximum pooling layer MP with the size of 56×56×256; the third convolution module Conv3 comprises 3 convolution layers with a size of 56×56×256 and a maximum pooling layer MP with a size of 28×28×512; the fourth convolution module Conv4 comprises 3 convolution layers with a size of 28×28×512 and a maximum pooling layer MP with a size of 14×14×512; the fifth convolution module Conv5 comprises 3 convolution layers with a size of 14×14×512 and a maximum pooling layer MP with a size of 7×7×512.
Wherein, the filter size of each convolution layer is 33, the step is 1, the filling is 1, the number of filters is gradually increased from 64 to 512, so as to reduce the size of data and the number of network parameters. In the maximum pooling layer MP, the output of the neuron clusters in the previous layer is combined into a single neuron by using a preset pooling function to reduce the dimension of the feature map, and each maximum pooling layer MP has a size of 2×2 and a stride of 2, so as to perform further feature selection and data filtering, reduce the image size and reduce the overfitting.
The first full connection block fc6 and the second full connection block fc7 each comprise a full connection layer with the size of 1×1×4096 (4096 neurons are shown), and a random discarding layer Dropout, and the random discarding layer Dropout can be used for randomly discarding some neurons to avoid overfitting; the third full connection block fc8 comprises a full connection layer (1000 neurons are shown) of 1×1×1000, the prediction result output after being processed by the full connection layer in the third full connection block fc8 is transmitted to a normalization mapping layer connected with the prediction result, the normalization mapping layer can normalize the prediction result by using a softmax activation function, and the output prediction result is converted into probability distribution, wherein each full connection layer adopts a filter with the size of 11, and all the neurons are connected with each other.
An input layer is further arranged in front of the first convolution module Conv1, and an output layer is further arranged behind the normalization mapping layer, so that the input layer and the output layer are not drawn in the figure for simplifying the description, and meanwhile, on the basis of the electric vehicle load decomposition model structure, reLU (Rectified Linear Unit, linear rectification function) can be used as an activation function in each convolution layer and the full connection layer to improve the load decomposition accuracy of the electric vehicle load decomposition model.
For better explanation, the load decomposition method of the electric vehicle will be described in further detail below in conjunction with the electric vehicle load decomposition model in the above embodiment.
Referring to fig. 3, a step flowchart of a load decomposition method of an electric automobile provided by an embodiment of the present invention may specifically include the following steps:
step 301, acquiring a power sequence data set of a power utilization area, and dividing the power sequence data set into a combined training set and a load decomposition test set according to a certain proportion;
before the electric automobile load decomposition is carried out on the total power consumption information of the electricity utilization area by adopting the model, the built model is required to be trained and tested by adopting the related data set, so that the power sequence data set of the electricity utilization area can be obtained, and the power sequence data set is divided into a combined training set and a load decomposition test set according to a certain proportion, so that model training is carried out through the divided data set.
Specifically, before acquiring the power sequence data set of the electricity consumption area, the measurement device may collect the residual load data of different electricity consumption access points in the electricity consumption area, or acquire the residual load data of different residences through the Pecan Street (an open source data set) data set, and further, the residual load data may include household load data, business load data and electric automobile load data.
Referring to fig. 4, an exemplary schematic load type structure of different power utilization access points in a power utilization area according to an embodiment of the present invention is shown, where a measurement device for measuring load data (such as power information) of electric equipment may be disposed on a low voltage side (i.e., an output side) of a power transmission network of a substation, the load data of the electric equipment may be measured and collected by the measurement device, home appliances such as an air conditioner, a dish washer, a refrigerator, and a heater in a residential area are used as household load types, commercial appliances such as a central air conditioner, a lighting, an office, and a commercial hotel in a commercial area are used as commercial load types, and meanwhile, electric automobile loads are collected as a single data type, so that an electric automobile load decomposition model may be trained based on the collected data.
Then, the household load data, the commercial load data and the electric vehicle load data can be combined and added to form a power sequence data set of the electricity utilization area, wherein the power sequence data set contains area total power consumption data and electric vehicle power consumption data, when model training is carried out, a training set containing the electric vehicle power consumption data and partial area total power consumption data can be adopted to carry out model training on an electric vehicle load decomposition model, then a test set containing the other partial area total power consumption data is adopted to carry out model test on the trained model, and model evaluation or further model optimization is carried out based on a test result, so that the power sequence data set is divided into a joint training set and a load decomposition test set according to a certain proportion, and the method specifically comprises the following steps: the method comprises the steps of dividing regional total power consumption data and electric vehicle power consumption data into a combined training set and a load decomposition test set according to a certain proportion, wherein the combined training set comprises electric vehicle power consumption data and a part of regional total power consumption data, and the load decomposition test set comprises the rest regional total power consumption data, for example, the regional total power consumption data can be divided into the combined training set and the load decomposition test set according to a division proportion of 9:1, the electric vehicle power consumption data can be completely divided into the combined training set, or part of electric vehicle power consumption data can be randomly selected to be divided into the combined training set, or the front half part/the rear half part of electric vehicle power consumption data can be selected to be divided into the combined training set, and the like.
Step 302, respectively performing data preprocessing on the combined training set and the load decomposition test set to obtain a power consumption training set corresponding to the combined training set and a power consumption test set corresponding to the load decomposition test set;
noise and outliers can be removed from an original combined training set and a load decomposition test set, downsampling processing is carried out on data, and then a sliding window method with a fixed window length is adopted for data segmentation, so that non-invasive load decomposition of regional electric vehicle charging loads can be realized only by adopting low-frequency power data in the process of carrying out load decomposition on an actual application model, data acquisition cost is effectively reduced, and the method can be applied to monitoring of substation-level electric vehicle charging behaviors so as to support high-proportion injection of distributed energy sources.
As an alternative embodiment, the data substitution correction, normalization processing and other modes can be adopted to respectively perform data preprocessing on the combined training set and the load decomposition test set, so as to obtain a power consumption training set corresponding to the combined training set and a power consumption test set corresponding to the load decomposition test set.
In a specific implementation, the substitution-normalization processing may be performed on the power sequence data corresponding to the combined training set and the load decomposition test set, respectively, and specifically may be:
Step S1: for the problem value of the power sequence data, replacing the problem value by adopting the average value of the two data before and after the problem value, wherein the problem value comprises a missing value and an abnormal value;
step S2: the maximum-minimum normalization method is adopted to map the power sequence data from-1 to 1 so as to eliminate the influence of different load scales, and the specific calculation formula is as follows:
wherein x is min Representing the minimum value, x, of the power sequence data max Represents the maximum value of the power sequence data, x represents the power sequence data before normalization processing, x norm Representing the power sequence data after normalization processing;
then, the sliding window method may be used to segment the time series data corresponding to the combined training set and the load decomposition test set after the substitution-normalization process, and divide the time series data into a plurality of sequence data segments with preset fixed lengths, where each sequence data segment includes a preset fixed number of sequence data.
After the data segmentation processing is performed, a plurality of sequence data segments corresponding to the combined training set can be used as a power consumption training set, and a plurality of sequence data segments corresponding to the load decomposition testing set can be used as a power consumption testing set, so that a more accurate training set and testing set can be obtained through data preprocessing and used for training and testing a model, and the accuracy can be improved when the model is used for load decomposition in the follow-up process.
Step 303, an electric automobile load decomposition model based on a VGG-16 convolutional neural network is established, and the electric automobile load decomposition model is trained by adopting the power consumption training set and the power consumption testing set;
in a specific implementation, an electric vehicle load decomposition model based on a VGG-16 convolutional neural network can be established, and the electric vehicle load decomposition model is trained by adopting the power consumption training set and the power consumption testing set obtained in the step 302.
After the data processing in the previous step, it can be obtained that the power consumption training set includes the power consumption sequence of the electric vehicle after the data preprocessing and the total power consumption sequence of the first area, and the power consumption testing set includes the total power consumption sequence of the second area after the data preprocessing, and then the power consumption training set and the power consumption testing set are adopted to train the electric vehicle load decomposition model, which can be as follows: firstly, model training is carried out on an electric vehicle load decomposition model by adopting an electric vehicle power consumption sequence and a first region total power consumption sequence to jointly extract characteristics, then the electric vehicle load decomposition model after model training is tested and verified by adopting a second region total power consumption sequence, and model evaluation is carried out on the basis of a test result.
And step 304, acquiring the total load power consumption information of the electricity utilization area, inputting the total load power consumption information into a trained electric vehicle load decomposition model to carry out load decomposition, and extracting an electric vehicle power sequence of the electricity utilization area.
And then, carrying out load decomposition on the total load power consumption information of the power utilization area by adopting a trained model so as to extract a power sequence corresponding to the electric automobile.
As can be seen from the foregoing description of the embodiment of the model structure, the electric vehicle load decomposition model provided by the embodiment of the present invention includes an input layer, 5 convolution blocks sequentially connected in series with the input layer, and 3 full connection blocks, and a normalized mapping layer connected to the end of the full connection blocks, for better explanation, referring to fig. 5, a step flowchart of a load decomposition method using the electric vehicle load decomposition model provided by the embodiment of the present invention is shown, and the step of inputting total load power consumption information into the trained electric vehicle load decomposition model for load decomposition, and extracting an electric vehicle power sequence in an electricity consumption area may include:
step 501, inputting the region total power consumption information into an input layer for normalization processing to obtain a total power consumption sequence to be decomposed;
The total power consumption information of the region needing to be subjected to load decomposition can be input into an input layer, and normalization processing is carried out on the input layer by adopting a gradient descent algorithm, so that a total power consumption sequence to be decomposed is obtained.
Step 502, transmitting the total power consumption sequence to be decomposed to the convolution block, sequentially carrying out convolution pooling treatment according to a connection sequence, and outputting a total power consumption characteristic diagram;
from the foregoing, it can be seen that the 5 convolution blocks may be correspondingly a first convolution module, a second convolution module, a third convolution module, a fourth convolution module, and a fifth convolution module, and further, the total power consumption sequence to be decomposed is transmitted to the convolution blocks, and convolution pooling processing is sequentially performed according to a connection order, so as to output a total power consumption feature map, which may be:
firstly, transmitting a total power consumption sequence to be decomposed to a first convolution module to carry out continuous standard convolution twice, carrying out pooling treatment, outputting a first intermediate feature image to a second convolution module to carry out continuous standard convolution twice, and carrying out pooling treatment to obtain a second intermediate feature image;
and then the second characteristic sequence diagram is transmitted to a third convolution module to carry out continuous three times of standard convolution, then pooling treatment is carried out, a third intermediate characteristic diagram is output to a fourth convolution module to carry out continuous three times of standard convolution, then pooling treatment is carried out, a fourth intermediate characteristic diagram is output to a fifth convolution module to carry out continuous three times of standard convolution, and then pooling treatment is carried out, so that a total power consumption characteristic diagram is obtained.
The standard convolution may include performing convolution calculation on a convolution layer and performing linear correction operation using a ReLU (Rectified Linear Unit, linear rectification function) as an activation function to improve the load decomposition accuracy of the electric vehicle load decomposition model.
Still further, each convolution module includes a plurality of convolution layers, each convolution layer includes a plurality of learnable kernels for extracting features of input data, each element of the convolution kernels corresponds to a weight coefficient and a bias vector, the neurons are similar to neurons of a feedforward neural network, in each convolution layer, each neuron is connected to a small area of an input volume so as to realize a sparse local connection mode between neurons of adjacent layers, a range of the connection area is called as a "receptive field", the range is determined by a size of the convolution kernel, the convolution layer is used for performing standard convolution processing on a feature map of the input convolution layer, and then the steps of standard convolution processing include:
firstly, determining an input feature map, scanning the input feature map through a convolution kernel of a convolution layer in a forward process, multiplying the input feature map, and adding the multiplied input feature map with a receptive field deviation of the convolution layer to obtain an output feature map, wherein a specific calculation formula is as follows:
Wherein Z is l Input feature map representing the (l+1) th convolution layer, Z l+1 Output characteristic diagram representing (l+1) th convolution layer, w l+1 The weight coefficient of the (l+1) th convolution layer is represented, b is the receptive field deviation;
then, the input volume of the input feature map is obtained, and the output volume of the output feature map is calculated according to the input volume, wherein the specific calculation formula is as follows:
wherein V is l The spatial size of the input volume for the (l+1) -th convolution layer, V l+1 The spatial size of the output volume of the (l+1) th convolution layer, k is the kernel size of the neuron in the corresponding convolution layer, p is the zero filling quantity at the edge, s is the step length, and the number of units converted by the filter each time is represented;
and then, linear correction can be carried out on the output characteristic diagram by adopting a ReLU activation function to obtain a corrected output characteristic diagram.
Step 503, transmitting the total power consumption feature map to the full connection block, and sequentially performing nonlinear feature integration according to a connection sequence to obtain an output feature sequence map;
from the foregoing, it can be seen that the 3 full connection blocks may be corresponding to a first full connection block, a second full connection block, and a third full connection block, and in a specific implementation, the total power consumption feature map is transmitted to the full connection blocks, and nonlinear feature integration is sequentially performed according to a connection order, so as to obtain an output feature sequence map, which may be:
Firstly, inputting a total power consumption characteristic diagram into a first full connecting block to perform full connection operation so as to realize characteristic combination classification, outputting a first classification characteristic diagram, linearly correcting the first classification characteristic diagram by adopting a ReLU activation function, and outputting a first corrected classification characteristic diagram to a second full connecting block;
performing full connection operation on the first correction classification characteristic map to realize characteristic combination classification, outputting a second classification characteristic map, performing linear correction on the second classification characteristic map by adopting a ReLU activation function, and outputting the second correction classification characteristic map to a third full connection block;
and then performing full-connection operation on the second modified classification feature map to realize feature combination classification, and obtaining an output feature sequence map.
Meanwhile, a random discarding layer Dropout can be added between the first full connecting block and the second full connecting block and between the second full connecting block and the third full connecting block, and some neurons are randomly discarded by adopting the random discarding layer Dropout, so that overfitting in the data processing process is avoided.
Step 504, inputting the output characteristic sequence diagram to the normalization mapping layer for normalization mapping processing, and converting the output characteristic sequence diagram into output probability distribution corresponding to the output characteristic sequence diagram, wherein the output probability distribution represents the probability of load classification of each input load in the region total power consumption information;
Specifically, the normalization processing can be performed at the normalization mapping layer by adopting a softmax activation function, so that the output characteristic sequence diagram is converted into probability distribution.
And 505, carrying out load decomposition on the total power consumption sequence to be decomposed according to the output probability distribution, and extracting an electric vehicle power sequence corresponding to the electric vehicle in the electricity utilization area.
In a specific implementation, the step of performing load decomposition on the total power consumption sequence to be decomposed according to the output probability distribution and extracting the power sequence of the electric vehicle corresponding to the electric vehicle in the power utilization area may be:
carrying out load decomposition on the total power consumption sequence to be decomposed through the following calculation formula, and extracting an electric vehicle power sequence corresponding to the electric vehicle in the electricity utilization area:
wherein P is agg (T) represents the measured value at t= { T on the low-voltage side of the substation 1 ,t 2 ,...,t n A total power consumption sequence to be decomposed at the moment; n represents the total number of sampling points; n represents the load type according to the refinement level; p (P) i (t) is a power sequence corresponding to the i-class load, expressed as:
P i (t)＝{p i,1 ,p i,2 ,...p i,n }
wherein P is i,t Representing the power consumption of the i-th class of load at the sampling point t.
Therefore, the electric vehicle load decomposition model can automatically carry out load decomposition on the total power consumption information, and the power sequence corresponding to the electric vehicle can be rapidly and accurately extracted, so that the effect that the electric vehicle load decomposition can be carried out only by relying on power data is realized, namely, the power consumption of the electric vehicle can be decomposed by inputting 'one region total power consumption' into a trained neural network.
In the embodiment of the invention, a convolutional neural network based on VGG-16 is provided to construct an electric vehicle load decomposition model, and a non-invasive decomposition method for an electric vehicle is realized based on the electric vehicle load decomposition model, specifically, a power sequence data set containing the overall power consumption of a power utilization area and the power consumption of the electric vehicle is formed by combining and adding residual load data in the power utilization area, the power sequence data set is divided into a training set and a testing set according to a certain proportion and is subjected to data preprocessing, so that the model training accuracy is improved, then the preprocessed data set is adopted to train the electric vehicle load decomposition model, and the trained model is utilized to carry out load decomposition on the overall power consumption information required to be subjected to load monitoring, so that the power sequence corresponding to the electric vehicle is rapidly extracted, the electric vehicle is not required to be measured, the use experience of a user is not influenced, meanwhile, the higher decomposition accuracy is provided, the load management capability of a power system is improved under the condition of high-proportion electric vehicle load injection, and a powerful support is provided for realizing the high-efficiency and reliable operation of the power system.
Referring to fig. 6, a block diagram of a load splitting device of an electric automobile according to an embodiment of the present invention may specifically include:
The power sequence data set acquisition module 601 is configured to acquire a power sequence data set of a power consumption area, and divide the power sequence data set into a joint training set and a load decomposition test set according to a certain proportion;
the data preprocessing module 602 is configured to perform data preprocessing on the combined training set and the load decomposition test set respectively, so as to obtain a power consumption training set corresponding to the combined training set and a power consumption test set corresponding to the load decomposition test set;
the electric automobile load decomposition model training module 603 is configured to establish an electric automobile load decomposition model based on a VGG-16 convolutional neural network, and train the electric automobile load decomposition model by adopting the power consumption training set and the power consumption testing set;
the load decomposition module 604 is configured to obtain total load power consumption information of the electricity consumption area, input the total load power consumption information to a trained electric vehicle load decomposition model to perform load decomposition, and extract an electric vehicle power sequence of the electricity consumption area.
In an alternative embodiment, the electric automobile load decomposition model includes an input layer, 5 convolution blocks and 3 full connection blocks connected in series with the input layer in sequence, and a normalized mapping layer connected with the end of the full connection blocks, and the load decomposition module 604 includes:
The to-be-decomposed total power consumption sequence generation module is used for inputting the regional total power consumption information into an input layer for normalization processing to obtain a to-be-decomposed total power consumption sequence;
the total power consumption characteristic diagram output module is used for transmitting the total power consumption sequence to be decomposed to the convolution block, sequentially carrying out convolution pooling treatment according to a connection sequence, and outputting a total power consumption characteristic diagram;
the output characteristic sequence diagram generation module is used for transmitting the total power consumption characteristic diagram to the full connecting block, and sequentially carrying out nonlinear characteristic integration according to a connecting sequence to obtain an output characteristic sequence diagram;
the normalization mapping processing module is used for inputting the output characteristic sequence diagram to the normalization mapping layer for normalization mapping processing and converting the output characteristic sequence diagram into output probability distribution corresponding to the output characteristic sequence diagram, wherein the output probability distribution represents the probability of load classification of each input load in the region total power consumption information;
and the electric automobile power sequence extraction module is used for carrying out load decomposition on the total power consumption sequence to be decomposed according to the output probability distribution, and extracting an electric automobile power sequence corresponding to the electric automobile in the electricity utilization area.
In an alternative embodiment, the 5 convolution blocks are a first convolution module, a second convolution module, a third convolution module, a fourth convolution module, and a fifth convolution module, and the total power consumption feature map output module includes:
the second intermediate feature map generating module is used for transmitting the total power consumption sequence to be decomposed to the first convolution module to carry out continuous standard convolution twice, carrying out pooling treatment, outputting the first intermediate feature map to the second convolution module to carry out continuous standard convolution twice, and carrying out pooling treatment to obtain a second intermediate feature map;
the total power consumption characteristic diagram generating module is used for transmitting the second characteristic sequence diagram to the third convolution module to perform continuous three times of standard convolution, then performing pooling treatment, outputting the third intermediate characteristic diagram to the fourth convolution module to perform continuous three times of standard convolution, then performing pooling treatment, outputting the fourth intermediate characteristic diagram to the fifth convolution module to perform continuous three times of standard convolution, and then performing pooling treatment to obtain the total power consumption characteristic diagram.
In an alternative embodiment, each convolution module includes a plurality of convolution layers, and the convolution layers are used for performing standard convolution processing on a feature map input into the convolution layers, and the apparatus includes:
The output characteristic diagram generating module is used for determining an input characteristic diagram, scanning the input characteristic diagram through a convolution kernel of the convolution layer, multiplying the input characteristic diagram, and adding the input characteristic diagram with a receptive field deviation of the convolution layer to obtain the output characteristic diagram, wherein a specific calculation formula is as follows:
wherein Z is l Input feature map representing the (l+1) th convolution layer, Z l+1 Output characteristic diagram representing (l+1) th convolution layer, w l+1 The weight coefficient of the (l+1) th convolution layer is represented, b is the receptive field deviation;
the output volume calculation module is used for obtaining the input volume of the input feature map and calculating the output volume of the output feature map according to the input volume, and the specific calculation formula is as follows:
wherein V is l The spatial size of the input volume for the (l+1) -th convolution layer, V l+1 The spatial size of the output volume of the (l+1) th convolution layer, k is the kernel size of the neuron in the corresponding convolution layer, p is the zero filling quantity at the edge, s is the step length, and the number of units converted by the filter each time is represented;
and the linear correction processing module is used for carrying out linear correction on the output characteristic diagram by adopting a ReLU activation function to obtain a corrected output characteristic diagram.
In an optional embodiment, the 3 full connection blocks are a first full connection block, a second full connection block, and a third full connection block, and the output feature sequence diagram generating module includes:
The first correction classification characteristic diagram output module is used for inputting the total power consumption characteristic diagram to the first full connecting block for characteristic combination classification processing, outputting a first classification characteristic diagram, linearly correcting the first classification characteristic diagram by adopting a ReLU activation function, and outputting a first correction classification characteristic diagram to the second full connecting block;
the second correction classification characteristic diagram output module is used for carrying out characteristic combination classification processing on the first correction classification characteristic diagram, outputting a second classification characteristic diagram, carrying out linear correction on the second classification characteristic diagram by adopting a ReLU activation function, and outputting a second correction classification characteristic diagram to the third full connection block;
and the second correction classification characteristic diagram processing module is used for carrying out characteristic combination classification processing on the second correction classification characteristic diagram to obtain an output characteristic sequence diagram.
In an alternative embodiment, the electric vehicle power sequence extraction module is specifically configured to:
and carrying out load decomposition on the total power consumption sequence to be decomposed by the following calculation formula, and extracting an electric automobile power sequence corresponding to the electric automobile in the electricity utilization area:
wherein P is agg (T) represents the measured value at t= { T on the low-voltage side of the substation 1 ,t 2 ,...,t n A total power consumption sequence to be decomposed at the moment; n represents the total number of sampling points; n represents the load type according to the refinement level; p (P) i (t) is a power sequence corresponding to the i-class load, expressed as:
P i (t)＝{p i,1 ,p i,2 ,...p i,n }
wherein P is i,t Representing the power consumption of the i-th class of load at the sampling point t.
In an alternative embodiment, the apparatus further comprises:
the residual load data collection module is used for collecting residual load data of different power utilization access points in a power utilization area, wherein the residual load data comprises household load data, business load data and electric vehicle load data;
the power sequence data set generation module is used for carrying out combined addition processing on the household load data, the commercial load data and the electric vehicle load data to form a power sequence data set of the power utilization area, wherein the power sequence data set comprises area total power consumption data and electric vehicle power consumption data;
the power sequence data set acquisition module 601 is specifically configured to:
dividing the regional total power consumption data and the electric vehicle power consumption data into a combined training set and a load decomposition test set according to a certain proportion, wherein the combined training set comprises the electric vehicle power consumption data and a part of the regional total power consumption data, and the load decomposition test set comprises the rest of the regional total power consumption data.
In an alternative embodiment, the data preprocessing module 602 includes:
the substitution-normalization processing module is used for respectively carrying out substitution-normalization processing on the power sequence data corresponding to the combined training set and the load decomposition test set, and specifically comprises the following steps:
step S1: for the problem value of the power sequence data, replacing the problem value by adopting the average value of the two data before and after the problem value, wherein the problem value comprises a missing value and an abnormal value;
step S2: mapping the power sequence data from-1 to 1 by adopting a maximum-minimum normalization method, wherein the specific calculation formula is as follows:
wherein x is min Representing the minimum value, x, of the power sequence data max Represents the maximum value of the power sequence data, x represents the power sequence data before normalization processing, x norm Representing the power sequence data after normalization processing;
the data segmentation processing module is used for respectively carrying out data segmentation on the time sequence data corresponding to the combined training set and the load decomposition test set which are subjected to substitution-normalization processing by adopting a sliding window method, dividing the time sequence data into a plurality of sequence data segments with preset fixed lengths, wherein each sequence data segment comprises a preset fixed number of sequence data;
And the sequence data segment distribution module is used for taking a plurality of sequence data segments corresponding to the combined training set as a power consumption training set and taking a plurality of sequence data segments corresponding to the load decomposition testing set as a power consumption testing set.
In an alternative embodiment, the power consumption training set includes an electric vehicle power consumption sequence after data preprocessing and a first area total power consumption sequence, the power consumption test set includes a second area total power consumption sequence after data preprocessing, and the electric vehicle load decomposition model training module 603 includes:
the load decomposition model training module is used for carrying out model training on the electric automobile load decomposition model by adopting the electric automobile power consumption sequence and the first region total power consumption sequence;
and the load decomposition model test module is used for testing the electric automobile load decomposition model after model training through the second region total power consumption sequence and carrying out model evaluation based on a test result.
For the device embodiments, since they are substantially similar to the method embodiments, the description is relatively simple, and reference is made to the description of the foregoing method embodiments for relevant points.
The embodiment of the invention also provides electronic equipment, which comprises a processor and a memory:
the memory is used for storing the program codes and transmitting the program codes to the processor;
the processor is used for executing the load decomposition method of the electric automobile according to any embodiment of the invention according to the instructions in the program codes.
The embodiment of the invention also provides a computer readable storage medium, which is used for storing program codes, and the program codes are used for executing the load decomposition method of the electric automobile of any embodiment of the invention.
It will be clear to those skilled in the art that, for convenience and brevity of description, specific working procedures of the above-described systems, apparatuses and units may refer to corresponding procedures in the foregoing method embodiments, which are not repeated herein.
In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods may be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative, e.g., the division of the units is merely a logical function division, and there may be additional divisions when actually implemented, e.g., multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other form.
The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
In addition, each functional unit in the embodiments of the present invention may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit. The integrated units may be implemented in hardware or in software functional units.
The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied essentially or in part or all of the technical solution or in part in the form of a software product stored in a storage medium, including instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
The above embodiments are only for illustrating the technical solution of the present invention, and not for limiting the same; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims (12)

1. A load split method of an electric vehicle, comprising:
acquiring a power sequence data set of a power utilization area, and dividing the power sequence data set into a combined training set and a load decomposition test set according to a certain proportion;
respectively carrying out data preprocessing on the combined training set and the load decomposition test set to obtain a power consumption training set corresponding to the combined training set and a power consumption test set corresponding to the load decomposition test set;
establishing an electric vehicle load decomposition model based on a VGG-16 convolutional neural network, and training the electric vehicle load decomposition model by adopting the power consumption training set and the power consumption testing set;
And acquiring the total load power consumption information of the electricity utilization area, inputting the total load power consumption information into a trained electric vehicle load decomposition model to carry out load decomposition, and extracting an electric vehicle power sequence of the electricity utilization area.
2. The load decomposition method according to claim 1, wherein the electric vehicle load decomposition model includes an input layer, 5 convolution blocks and 3 full connection blocks sequentially connected in series with the input layer, and a normalized mapping layer connected to the end of the full connection blocks, and the inputting the total load power consumption information into the trained electric vehicle load decomposition model to perform load decomposition, and extracting an electric vehicle power sequence of the power utilization area includes:
inputting the region total power consumption information into an input layer for normalization processing to obtain a to-be-decomposed total power consumption sequence;
transmitting the total power consumption sequence to be decomposed to the convolution block, sequentially carrying out convolution pooling treatment according to a connection sequence, and outputting a total power consumption characteristic diagram;
transmitting the total power consumption characteristic diagram to the full connecting block, and sequentially carrying out nonlinear characteristic integration according to a connecting sequence to obtain an output characteristic sequence diagram;
Inputting the output characteristic sequence diagram to the normalization mapping layer for normalization mapping processing, and converting the output characteristic sequence diagram into output probability distribution corresponding to the output characteristic sequence diagram, wherein the output probability distribution represents the probability of load classification of each input load in the region total power consumption information;
and carrying out load decomposition on the total power consumption sequence to be decomposed according to the output probability distribution, and extracting an electric vehicle power sequence corresponding to the electric vehicle in the electricity utilization area.
3. The method of load decomposition according to claim 2, wherein the 5 convolution blocks correspond to a first convolution module, a second convolution module, a third convolution module, a fourth convolution module, and a fourth convolution module
The five convolution modules are used for transmitting the total power consumption sequence to be decomposed to the convolution blocks, sequentially carrying out convolution pooling processing according to the connection sequence, and outputting a total power consumption characteristic diagram, and the five convolution modules comprise:
transmitting the total power consumption sequence to be decomposed to a first convolution module for continuous standard convolution twice, carrying out pooling treatment, outputting a first intermediate feature map to a second convolution module for continuous standard convolution twice, and carrying out pooling treatment to obtain a second intermediate feature map;
And transmitting the second characteristic sequence diagram to a third convolution module for continuous three times of standard convolution, carrying out pooling treatment, outputting a third intermediate characteristic diagram to a fourth convolution module for continuous three times of standard convolution, carrying out pooling treatment, outputting a fourth intermediate characteristic diagram to a fifth convolution module for continuous three times of standard convolution, and carrying out pooling treatment to obtain a total power consumption characteristic diagram.
4. A method of load decomposition according to claim 3, wherein each convolution module includes a plurality of convolution layers, said convolution layers being configured to perform a standard convolution process on a feature map input to said convolution layers, said standard convolution process comprising:
determining an input feature map, scanning the input feature map through a convolution kernel of the convolution layer, multiplying the input feature map, and adding the input feature map with a receptive field deviation of the convolution layer to obtain an output feature map, wherein a specific calculation formula is as follows:
wherein Z is l Input feature map representing the (l+1) th convolution layer, Z l+1 Output characteristic diagram representing (l+1) th convolution layer, w l+1 The weight coefficient of the (l+1) th convolution layer is represented, b is the receptive field deviation;
the input volume of the input feature map is obtained, and the output volume of the output feature map is calculated according to the input volume, wherein the specific calculation formula is as follows:
Wherein V is l The spatial size of the input volume for the (l+1) -th convolution layer, V l+1 The spatial size of the output volume of the (l+1) th convolution layer, k is the kernel size of the neuron in the corresponding convolution layer, p is the zero filling quantity at the edge, s is the step length, and the number of units converted by the filter each time is represented;
and linearly correcting the output characteristic diagram by adopting a ReLU activation function to obtain a corrected output characteristic diagram.
5. The load decomposition method according to claim 2, wherein the 3 full connection blocks are corresponding to a first full connection block, a second full connection block and a third full connection block, the transmitting the total power consumption feature map to the full connection blocks, and sequentially performing nonlinear feature integration according to a connection order, to obtain an output feature sequence map, including:
inputting the total power consumption characteristic diagram to the first full connecting block for characteristic combination classification processing, outputting a first classification characteristic diagram, linearly correcting the first classification characteristic diagram by adopting a ReLU activation function, and outputting a first corrected classification characteristic diagram to the second full connecting block;
performing feature combination classification processing on the first corrected classification feature map, outputting a second classification feature map, performing linear correction on the second classification feature map by adopting a ReLU activation function, and outputting the second corrected classification feature map to the third full connecting block;
And carrying out feature combination classification processing on the second corrected classification feature map to obtain an output feature sequence map.
6. The method for decomposing the load according to claim 2, wherein the performing the load decomposition on the total power consumption sequence to be decomposed according to the output probability distribution, and extracting the power sequence of the electric vehicle corresponding to the electric vehicle in the electricity utilization area, includes:
and carrying out load decomposition on the total power consumption sequence to be decomposed by the following calculation formula, and extracting an electric automobile power sequence corresponding to the electric automobile in the electricity utilization area:
wherein P is agg (T) represents the measured value at t= { T on the low-voltage side of the substation 1 ,t 2 ,...,t n A total power consumption sequence to be decomposed at the moment; n represents the total number of sampling points; n represents the load type according to the refinement level; p (P) i (t) is a power sequence corresponding to the i-class load, expressed as:
P i (t)＝{p i,1 ,p i,2 ,...p i,n }
wherein P is i,t Representing the power consumption of the i-th class of load at the sampling point t.
7. The load splitting method of claim 1, wherein prior to the acquiring the power sequence data set for the power usage region, the method further comprises:
collecting residual load data of different power utilization access points in a power utilization area, wherein the residual load data comprises household load data, business load data and electric vehicle load data;
Carrying out combined addition processing on the household load data, the commercial load data and the electric vehicle load data to form a power sequence data set of the power utilization area, wherein the power sequence data set comprises area total power consumption data and electric vehicle power consumption data;
the dividing the power sequence data set into a joint training set and a load decomposition test set according to a certain proportion comprises the following steps:
dividing the regional total power consumption data and the electric vehicle power consumption data into a combined training set and a load decomposition test set according to a certain proportion, wherein the combined training set comprises the electric vehicle power consumption data and a part of the regional total power consumption data, and the load decomposition test set comprises the rest of the regional total power consumption data.
8. The method of load decomposition according to claim 7, wherein the performing data preprocessing on the combined training set and the load decomposition test set to obtain a power consumption training set corresponding to the combined training set and a power consumption test set corresponding to the load decomposition test set respectively includes:
and respectively carrying out substitution-normalization processing on the power sequence data corresponding to the combined training set and the load decomposition test set, wherein the substitution-normalization processing comprises the following steps of:
Step S1: for the problem value of the power sequence data, replacing the problem value by adopting the average value of the two data before and after the problem value, wherein the problem value comprises a missing value and an abnormal value;
step S2: mapping the power sequence data from-1 to 1 by adopting a maximum-minimum normalization method, wherein the specific calculation formula is as follows:
wherein x is min Representing the minimum value, x, of the power sequence data max Representing powerMaximum value of sequence data, x represents power sequence data before normalization processing, x norm Representing the power sequence data after normalization processing;
respectively segmenting the time sequence data corresponding to the combined training set and the load decomposition test set subjected to substitution-normalization processing by adopting a sliding window method, and dividing the time sequence data into a plurality of sequence data segments with preset fixed lengths, wherein each sequence data segment comprises a preset fixed number of sequence data;
and taking the plurality of sequence data segments corresponding to the combined training set as a power consumption training set, and taking the plurality of sequence data segments corresponding to the load decomposition testing set as a power consumption testing set.
9. The method of claim 7, wherein the power consumption training set includes an electric vehicle power consumption sequence after data preprocessing and a first region total power consumption sequence, the power consumption test set includes a second region total power consumption sequence after data preprocessing, and the training the electric vehicle load decomposition model using the power consumption training set and the power consumption test set includes:
Performing model training on the electric vehicle load decomposition model by adopting the electric vehicle power consumption sequence and the first region total power consumption sequence;
and testing the electric automobile load decomposition model after model training through the second region total power consumption sequence, and performing model evaluation based on a test result.
10. A load split device for an electric vehicle, comprising:
the power sequence data set acquisition module is used for acquiring a power sequence data set of a power utilization area and dividing the power sequence data set into a combined training set and a load decomposition test set according to a certain proportion;
the data preprocessing module is used for respectively preprocessing the data of the combined training set and the load decomposition test set to obtain a power consumption training set corresponding to the combined training set and a power consumption test set corresponding to the load decomposition test set;
the electric automobile load decomposition model training module is used for establishing an electric automobile load decomposition model based on a VGG-16 convolutional neural network, and training the electric automobile load decomposition model by adopting the power consumption training set and the power consumption testing set;
The load decomposition module is used for acquiring the total load power consumption information of the electricity utilization area, inputting the total load power consumption information into the trained electric vehicle load decomposition model to carry out load decomposition, and extracting an electric vehicle power sequence of the electricity utilization area.
11. An electronic device, the device comprising a processor and a memory:
the memory is used for storing program codes and transmitting the program codes to the processor;
the processor is configured to execute the load splitting method of the electric vehicle according to any one of claims 1 to 9 according to instructions in the program code.
12. A computer-readable storage medium storing program code for executing the load splitting method of the electric vehicle according to any one of claims 1 to 9.
CN202310491742.8A 2023-05-04 2023-05-04 Load decomposition method and device for electric automobile, electronic equipment and storage medium Pending CN116523681A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310491742.8A CN116523681A (en) 2023-05-04 2023-05-04 Load decomposition method and device for electric automobile, electronic equipment and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310491742.8A CN116523681A (en) 2023-05-04 2023-05-04 Load decomposition method and device for electric automobile, electronic equipment and storage medium

Publications (1)

Publication Number Publication Date
CN116523681A true CN116523681A (en) 2023-08-01

Family

ID=87400712

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310491742.8A Pending CN116523681A (en) 2023-05-04 2023-05-04 Load decomposition method and device for electric automobile, electronic equipment and storage medium

Country Status (1)

Country Link
CN (1) CN116523681A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116872780A (en) * 2023-09-08 2023-10-13 国网浙江省电力有限公司杭州供电公司 Electric automobile charging supply control method, device, terminal and medium

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116872780A (en) * 2023-09-08 2023-10-13 国网浙江省电力有限公司杭州供电公司 Electric automobile charging supply control method, device, terminal and medium
CN116872780B (en) * 2023-09-08 2023-12-15 国网浙江省电力有限公司杭州供电公司 Electric automobile charging supply control method, device, terminal and medium

Similar Documents

Publication Publication Date Title
CN110610280B (en) Short-term prediction method, model, device and system for power load
Catalão et al. Short-term wind power forecasting in Portugal by neural networks and wavelet transform
US20170351288A1 (en) Non-invasive online real-time electric load identification method and identification system
CN109492667A (en) A kind of feature selecting discrimination method for non-intrusive electrical load monitoring
CN111242276B (en) One-dimensional convolutional neural network construction method for load current signal identification
CN111880044B (en) Online fault positioning method for distribution network containing distributed power supply
CN109190950B (en) Energy consumption abnormity identification method for communicating electric meter and water meter files
CN115170000B (en) Remote monitoring method and system based on electric energy meter communication module
Li et al. Energy data generation with wasserstein deep convolutional generative adversarial networks
CN116523681A (en) Load decomposition method and device for electric automobile, electronic equipment and storage medium
CN113141008B (en) Data-driven power distribution network distributed new energy consumption capability assessment method
CN114722873A (en) Non-invasive load decomposition method based on residual convolution and attention mechanism
CN112598303A (en) Non-invasive load decomposition method based on combination of 1D convolutional neural network and LSTM
CN113036759A (en) Power consumer load fine-grained identification method and identification system
CN116451117A (en) Power data anomaly detection method based on federal learning
CN111654392A (en) Low-voltage distribution network topology identification method and system based on mutual information
Dai et al. Fault diagnosis of data-driven photovoltaic power generation system based on deep reinforcement learning
CN115065078A (en) Energy storage capacity configuration method and system in micro-grid environment
CN112307675B (en) Neural network-based temperature-sensitive load separation identification method and system
CN112884532A (en) Low-power consumption thing networking collection system
Aman et al. Learning to reduce: A reduced electricity consumption prediction ensemble
CN116205544A (en) Non-invasive load identification system based on deep neural network and transfer learning
CN111611747B (en) Online state estimation method and device for hybrid energy storage system
CN115563583A (en) Non-invasive load monitoring method based on multi-physical quantity fusion
CN111753097A (en) Deep learning-based data analysis method and device for electric power spot transaction clearing

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