CN116703464A - Electric automobile charging demand modeling method and device, electronic equipment and storage medium - Google Patents
Electric automobile charging demand modeling method and device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN116703464A CN116703464A CN202310688564.8A CN202310688564A CN116703464A CN 116703464 A CN116703464 A CN 116703464A CN 202310688564 A CN202310688564 A CN 202310688564A CN 116703464 A CN116703464 A CN 116703464A
- Authority
- CN
- China
- Prior art keywords
- charging
- subjective
- data
- lstm network
- sample
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 83
- 238000012549 training Methods 0.000 claims abstract description 94
- 238000005457 optimization Methods 0.000 claims abstract description 57
- 230000006399 behavior Effects 0.000 claims description 197
- 238000004422 calculation algorithm Methods 0.000 claims description 38
- 238000004364 calculation method Methods 0.000 claims description 34
- 238000012795 verification Methods 0.000 claims description 29
- 230000008569 process Effects 0.000 claims description 23
- 238000012937 correction Methods 0.000 claims description 14
- 238000012545 processing Methods 0.000 claims description 14
- 238000012360 testing method Methods 0.000 claims description 14
- 230000035945 sensitivity Effects 0.000 claims description 13
- 230000015654 memory Effects 0.000 claims description 8
- 238000010606 normalization Methods 0.000 claims description 8
- 230000002238 attenuated effect Effects 0.000 claims description 7
- 230000003044 adaptive effect Effects 0.000 claims description 3
- 238000004458 analytical method Methods 0.000 description 14
- 230000006870 function Effects 0.000 description 12
- 238000013528 artificial neural network Methods 0.000 description 8
- 230000000694 effects Effects 0.000 description 7
- 238000013135 deep learning Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 6
- 230000004913 activation Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 230000003993 interaction Effects 0.000 description 4
- 230000001537 neural effect Effects 0.000 description 4
- 238000004088 simulation Methods 0.000 description 4
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 210000002569 neuron Anatomy 0.000 description 3
- 230000006403 short-term memory Effects 0.000 description 3
- 230000000903 blocking effect Effects 0.000 description 2
- 238000007621 cluster analysis Methods 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 230000005284 excitation Effects 0.000 description 2
- 239000000446 fuel Substances 0.000 description 2
- 238000003064 k means clustering Methods 0.000 description 2
- 230000007787 long-term memory Effects 0.000 description 2
- 230000007774 longterm Effects 0.000 description 2
- 210000005036 nerve Anatomy 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 239000013598 vector Substances 0.000 description 2
- 238000011217 control strategy Methods 0.000 description 1
- 230000010485 coping Effects 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000003912 environmental pollution Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 239000002360 explosive Substances 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000011423 initialization method Methods 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000000178 monomer Substances 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 238000010792 warming Methods 0.000 description 1
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
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
- G06Q30/0202—Market predictions or forecasting for commercial activities
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
- G06Q10/06315—Needs-based resource requirements planning or analysis
-
- 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
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/003—Load forecast, e.g. methods or systems for forecasting future load demand
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/60—Other road transportation technologies with climate change mitigation effect
- Y02T10/70—Energy storage systems for electromobility, e.g. batteries
Abstract
The invention discloses a method and a device for modeling charging requirements of an electric automobile, electronic equipment and a storage medium, which are used for solving the problem that the prediction accuracy of an existing electric automobile charging requirement model is low. The method comprises the following steps: acquiring a plurality of groups of subjective charging behavior data of a user, and clustering the plurality of groups of subjective charging behavior data to obtain a corresponding charging behavior clustering result; constructing a deep LSTM network, acquiring a plurality of groups of objective factor data of a user, and performing associated optimization training on the deep LSTM network by adopting a plurality of groups of objective factor data, charging behavior clustering results and a plurality of groups of subjective charging behavior data to acquire a target deep LSTM network; obtaining objective factor data to be detected of a user to be analyzed, inputting the objective factor data to a target deep LSTM network, and outputting predicted subjective charging behavior data; and calculating a charging load curve of the user to be analyzed according to the predicted subjective charging behavior data, and performing charging load prediction based on the charging load curve.
Description
Technical Field
The invention relates to the technical field of electric vehicle charging analysis, in particular to an electric vehicle charging demand modeling method, an electric vehicle charging demand modeling device, electronic equipment and a storage medium.
Background
With the explosive growth of the number of automobiles, the problems of gradual exhaustion of primary energy, global warming, environmental pollution and the like are brought, and under the condition, the development of new energy automobiles is one of effective ways for coping with the problems. In recent years, the increasing speed of the number of new energy automobiles is continuously increased, the holding quantity of the new energy automobiles in the whole country reaches 1310 ten thousand by the end of 2022, and the holding quantity accounts for 4.1% of the total quantity in the whole country, but for a power grid, unordered charging of large-scale electric automobiles can cause a series of problems of insufficient capacity of a power distribution network, insufficient capacity of the power distribution network, partial heavy overload, reduced power quality and the like.
Aiming at the problems, the interaction technology of the electric automobile and the power grid gradually becomes an effective way for solving the problem of impact on the power grid caused by disordered charging of the large-scale electric automobile, and meanwhile, the method can also play the roles of peak clipping, valley filling, blocking relieving and the like, so that the real-time balance of the power system, the safety and the reliable operation of the power grid are effectively ensured.
Therefore, accurate prediction of the charging demand of the electric automobile is an important basis for realizing the interaction technology of the automobile network, the randomness of the charging demand of the electric automobile depends on the travel rule of a user, at present, the research on the travel rule of the electric automobile is mostly based on a large number of assumptions, such as replacing the travel rule of the electric automobile with the travel rule of the fuel automobile, assuming relevant parameters to obey normal distribution and the like, the setting of influence factor parameters is simple, subjectivity is strong, and large errors exist between the influence factor parameters and complex and changeable actual application scenes, so that the accuracy of a model has large limitation, and the prediction accuracy of the charging demand of the electric automobile is low.
Disclosure of Invention
The invention provides a method, a device, electronic equipment and a storage medium for modeling the charging demand of an electric automobile, which are used for solving or partially solving the technical problem that the prediction accuracy of the charging demand model of the electric automobile in the prior art is low.
The invention provides a method for modeling the charging requirement of an electric automobile, which comprises the following steps:
acquiring a plurality of groups of subjective charging behavior data of a user, and clustering the subjective charging behavior data to obtain a corresponding charging behavior clustering result;
constructing a deep LSTM network, acquiring a plurality of groups of objective factor data of the user, and performing association optimization training on the deep LSTM network by adopting the plurality of groups of objective factor data, the charging behavior clustering result and the plurality of groups of subjective charging behavior data to acquire a target deep LSTM network;
obtaining objective factor data to be detected of a user to be analyzed, inputting the objective factor data to be detected into the target deep LSTM network, and outputting predicted subjective charging behavior data;
and calculating a charging load curve of the user to be analyzed according to the predicted subjective charging behavior data, and performing charging load prediction based on the charging load curve.
Optionally, the clustering the multiple sets of subjective charging behavior data to obtain a corresponding charging behavior clustering result includes:
step S11: taking each subjective charging behavior data as a clustering sample, and initializing K initial centroids;
step S12: calculating sample distances from each clustering sample to each initial centroid;
step S13: respectively distributing each clustered sample to a sample cluster corresponding to an initial centroid with the smallest sample distance, calculating a clustered average value of clustered samples in each sample cluster, and re-determining K optimized centroids based on the clustered average value;
step S14: repeating the steps S12 to S13 until the sample cluster allocation is unchanged or the maximum iteration number is reached;
step S15: and clustering samples in each sample cluster and cluster information corresponding to each sample cluster are used as a charging behavior clustering result.
Optionally, the initializing K initial centroids includes:
step S21: randomly selecting one cluster sample from the cluster samples as a first initial centroid;
step S22: taking each cluster sample remained after the first centroid is selected as an observation value, and calculating the observation distance from each observation value to the first initial centroid;
Step S23: according to the observation distance, randomly selecting one cluster sample from the cluster samples remained after the first initial centroid is selected as a second initial centroid, wherein the selection probability is as follows:
wherein d (x m ，c 1 ) For the observed value x m To a first initial centroid c 1 The observation distance between the two is divided by the first initial centroid c 1 The number of clustering samples outside, d (x j ，c 1 ) To divide the first initial centroid c 1 Outside observation value x j To a first initial centroid c 1 The observation distance between the two;
step S24: steps S22 to S23 are repeatedly performed until K initial centroids are selected.
Optionally, before calculating the clustered samples, the method further comprises:
the normalization process is performed on each of the clustered samples by the following formula:
wherein x is mn N-th dimensional data for an mth cluster sample; mu (mu) n The nth dimension data average value of all the clustered samples is obtained; sigma (sigma) n The nth dimensional data variance for all cluster samples,and (3) the n-th dimension data of the m-th clustering sample obtained after the normalization processing.
Optionally, the subjective charging behavior data includes an initial charging time, a charging duration, a required electric quantity and a controlled preference, the objective factor data includes average month income, charging expense sensitivity, daily driving mileage, charging service unit price, incentive subsidy, outdoor weather condition, charging facility capacity distribution density and daily type, the deep LSTM network is subjected to associated optimization training by using the multiple sets of objective factor data, the charging behavior clustering result and the multiple sets of subjective charging behavior data, and a target deep LSTM network is obtained, which includes:
Setting parameters of the deep LSTM network, taking average month income of the user, the charging expense sensitivity, the daily driving mileage, the charging service unit price, the incentive subsidy, the outdoor weather condition, the charging facility capacity distribution density, the daily type and the charging behavior clustering result as input sample characteristics, and taking the initial charging time, the charging duration, the required electric quantity and the controlled preference as input sample labels;
and carrying out association training between subjective behavior and objective factors of user charging on the deep LSTM network after parameter setting by adopting the input sample characteristics and the input sample labels, and carrying out network optimization adjustment on the deep LSTM network after association training to obtain a target deep LSTM network.
Optionally, the training the association between the subjective behavior and the objective factor of the user charging on the deep LSTM network after the parameter setting by using the input sample feature and the input sample label includes:
carrying out association training between subjective behavior and objective factors of user charging on the deep LSTM network subjected to parameter setting by adopting the input sample characteristics and the input sample labels;
In the associated training process, the learning rate of the deep LSTM network is adjusted according to a piecewise constant attenuation mode, and gradient correction is carried out on the deep LSTM network based on a self-adaptive momentum random optimization mode, so that an optimized deep LSTM network is obtained;
randomly extracting data with preset proportion from the input sample characteristics to be used as a characteristic verification sample, and randomly extracting data with preset proportion from the input sample labels to be used as a label verification sample;
and carrying out accuracy verification on the optimized deep LSTM network by adopting the characteristic verification sample and the label verification sample, and carrying out network optimization adjustment on the optimized deep LSTM network to obtain a target deep LSTM network.
Optionally, the adjusting the learning rate of the deep LSTM network according to the piecewise constant attenuation mode includes:
in the iteration process of network training, the current learning rate of the deep LSTM network is attenuated to be a preset multiple of the previous learning rate after the preset iteration times, wherein the attenuated learning rate is smaller than the learning rate before the attenuation.
Optionally, the gradient correction is performed on the deep LSTM network based on the adaptive momentum random optimization mode, including:
And obtaining parameter updating difference values required by iteratively updating the deep LSTM network by adopting a self-adaptive momentum random optimization mode, wherein the calculation formula is as follows:
M t ＝β 1 M t-1 +(1-β 1 )g t
G t ＝β 2 G t-1 +(1-β 2 )g t ⊙g t
wherein beta is 1 Is the first moment attenuation coefficient, beta 2 Is a second moment attenuation coefficient, alpha is a step factor, M t Weighted average of the current index for first moment, G t Is the current exponentially weighted average of the second moment, M t-1 Is M t Corresponding last exponentially weighted average, G t-1 Is G t The corresponding last exponentially weighted average, g t For the current gradient, t is the current iteration number, θ is the parameter to be updated, and Δθ t Updating the difference value for the parameter corresponding to the current required parameter, wherein, as indicated by the ";
and carrying out gradient correction on the parameters corresponding to the deep LSTM network based on the parameter updating difference value.
Optionally, the performing network optimization adjustment on the deep LSTM network after the association training includes:
respectively inputting the subjective charging behavior data into the optimized deep LSTM network and at least one preset comparison algorithm model, and calculating a first root mean square error corresponding to the optimized deep LSTM network and a second root mean square error corresponding to at least one preset comparison algorithm model;
If a second root mean square error is smaller than the first root mean square error, determining the subjective charging behavior data as subjective charging behavior data to be optimized, replacing the optimized deep LSTM network by a preset comparison algorithm model corresponding to the second root mean square error smaller than the first root mean square error, and performing data fitting on the subjective charging behavior data to be optimized;
if at least two second root mean square errors are smaller than the first root mean square errors, determining the subjective charging behavior data as subjective charging behavior data to be optimized, replacing the optimized deep LSTM network by a preset comparison algorithm model corresponding to the minimum second root mean square errors, and performing data fitting on the subjective charging behavior data to be optimized;
if the second root mean square error is greater than or equal to the first root mean square error, continuing to perform data fitting on the subjective charging behavior data by adopting the optimized deep LSTM network;
the root mean square error is calculated as follows:
wherein RMSE is root mean square error, m t A number of test set samples representing a model or network,representation ofOutput value of model or network, y i Representing the actual value of the test set samples.
Optionally, the predicted subjective charging behavior data includes a predicted initial charging time, a predicted charging duration, a predicted required electric quantity, and a predicted controlled preference, and the calculating the charging load curve of the user to be analyzed according to the predicted subjective charging behavior data includes:
and calculating by adopting the predicted required electric quantity and the predicted charging time length to obtain an average charging load, wherein the calculation formula is as follows:
calculating the predicted controlled preference and the average charging load to obtain a controlled charging load, and drawing a corresponding charging load curve, wherein the calculation formula is as follows:
P c ′＝(1-R p )×P c
wherein P is c For average charging load, ΔE is predicted required power, Δt is predicted charging duration, P c ' is a controlled post-charge load, R p To predict controlled preferences;
the charge load curve is expressed as: at [ T ] b ,T b +△t]In the time period, the charging load corresponding to the user to be analyzed is P c ' in [ T ] b ,T b +△t]The charging load corresponding to the user to be analyzed is zero and T is the time period except the time period b To predict the initial charge time.
Optionally, the user to be analyzed is one of the users in the area to be analyzed of the electric automobile, and the method further includes:
And superposing charging load curves of all users to be analyzed in the area to be analyzed to obtain a total charging load curve, and carrying out overall charging load prediction on the electric automobile in the area to be analyzed based on the total charging load curve.
The invention also provides an electric automobile charging demand modeling device, which comprises:
the clustering calculation module is used for acquiring a plurality of groups of subjective charging behavior data of a user, and clustering the subjective charging behavior data to obtain a corresponding charging behavior clustering result;
the correlation optimization training module is used for constructing a deep LSTM network, acquiring a plurality of groups of objective factor data of the user, and carrying out correlation optimization training on the deep LSTM network by adopting the plurality of groups of objective factor data, the charging behavior clustering result and the plurality of groups of subjective charging behavior data to acquire a target deep LSTM network;
the predicted subjective charging behavior data output module is used for acquiring objective factor data to be detected of a user to be analyzed, inputting the objective factor data to be detected into the target deep LSTM network and outputting predicted subjective charging behavior data;
and the charging load prediction module is used for calculating a charging load curve of the user to be analyzed according to the predicted subjective charging behavior data and performing charging load prediction based on the charging load curve.
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 electric vehicle charging demand modeling method according to any one of the above instructions in the program code.
The present invention also provides a computer-readable storage medium for storing program code for executing the electric vehicle charging demand modeling method as described in any one of the above.
From the above technical scheme, the invention has the following advantages: aiming at the charge load demand prediction of the electric automobile, the electric automobile charge demand modeling method is provided, firstly, a plurality of groups of subjective charge behavior data of a user are obtained, and the subjective charge behavior data are clustered to obtain corresponding charge behavior clustering results, so that a large amount of data can be classified and classified into a plurality of data groups with similar values by clustering complex and various data, the training effect is better in the subsequent network training process, and the charge load prediction accuracy of the network after training optimization is higher; then constructing a deep LSTM network, acquiring a plurality of groups of objective factor data of a user, carrying out associated optimization training on the deep LSTM network by adopting a plurality of groups of objective factor data, charging behavior clustering results and a plurality of groups of subjective charging behavior data, and obtaining a target deep LSTM network, so that the charging requirement of the user is measured by quantifying the subjective charging behavior data, and meanwhile, the influence of a plurality of objective factors on the charging requirement is considered, so that the model can be more suitable for a practical complex charging scene; then obtaining objective factor data to be detected of a user to be analyzed, inputting the objective factor data to a target deep LSTM network, and outputting predicted subjective charging behavior data; and then calculating a charging load curve of the user to be analyzed according to the predicted subjective charging behavior data, and performing charging load prediction based on the charging load curve. Compared with the traditional method, the method adopts the autonomous learning capability and generalization capability of artificial intelligent algorithms such as clustering, deep learning and the like, and performs intelligent analysis on the correlation between objective influence factor data and subjective charging behavior data, so that the simulation analysis of the randomness of the charging behavior is realized, and the accuracy of the charging load prediction is further improved.
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 step flowchart of an electric vehicle charging demand modeling method according to an embodiment of the present invention;
fig. 2 is a schematic diagram of subjective charging behavior samples after clustering according to an embodiment of the present invention;
fig. 3 is a schematic structural diagram of a deep LSTM network according to an embodiment of the present invention;
fig. 4 is a flowchart illustrating steps of another method for modeling a charging requirement of an electric vehicle according to an embodiment of the present invention;
fig. 5 is a block diagram of a charging demand modeling apparatus for an electric vehicle according to an embodiment of the present invention.
Detailed Description
The embodiment of the invention provides a method, a device, electronic equipment and a storage medium for modeling the charging demand of an electric automobile, which are used for solving or partially solving the technical problem that the prediction accuracy of the charging demand model of the electric automobile in the prior art is low.
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, in recent years, the number of new energy automobiles is increasing, and by the end of 2022, the holding amount of new energy automobiles in the whole country is 1310 ten thousand, which accounts for 4.1% of the total amount in the whole country, but for a power grid, unordered charging of large-scale electric automobiles will cause a series of problems such as insufficient capacity of a power distribution network, insufficient capacity of the power distribution network, local heavy overload, and reduced power quality.
Aiming at the problems, the interaction technology of the electric automobile and the power grid gradually becomes an effective way for solving the problem of impact on the power grid caused by disordered charging of the large-scale electric automobile, and meanwhile, the method can also play the roles of peak clipping, valley filling, blocking relieving and the like, so that the real-time balance of the power system, the safety and the reliable operation of the power grid are effectively ensured.
Therefore, accurate prediction of the charging demand of the electric automobile is an important basis for realizing the interaction technology of the automobile network, the randomness of the charging demand of the electric automobile depends on the travel rule of a user, at present, the research on the travel rule of the electric automobile is mostly based on a large number of assumptions, such as replacing the travel rule of the electric automobile with the travel rule of the fuel automobile, assuming relevant parameters to obey normal distribution and the like, the setting of influence factor parameters is simple, subjectivity is strong, and large errors exist between the influence factor parameters and complex and changeable actual application scenes, so that the accuracy of a model has large limitation, and the prediction accuracy of the charging demand of the electric automobile is low. That is, the traditional charge demand modeling method adopts a hypothesized and ideal statistical rule to simulate the charge behavior rule of each user, so that the model has poor adaptability and low accuracy.
Therefore, one of the core inventions of the embodiments of the present invention is: aiming at the charge load demand prediction of the electric automobile, the electric automobile charge demand modeling method is provided, firstly, a plurality of groups of subjective charge behavior data of a user are obtained, and the subjective charge behavior data are clustered to obtain corresponding charge behavior clustering results, so that a large amount of data can be classified and classified into a plurality of data groups with similar numerical values by clustering complex and various data, the training effect is better in the subsequent network training process, and the charge load prediction accuracy of the network after training optimization is higher; then constructing a deep LSTM network, acquiring a plurality of groups of objective factor data of a user, carrying out associated optimization training on the deep LSTM network by adopting a plurality of groups of objective factor data, charging behavior clustering results and a plurality of groups of subjective charging behavior data, and obtaining a target deep LSTM network, so that the charging requirement of the user is measured by quantifying the subjective charging behavior data, and meanwhile, the influence of a plurality of objective factors on the charging requirement is considered, so that the model can be more suitable for a practical complex charging scene; then obtaining objective factor data to be detected of a user to be analyzed, inputting the objective factor data to a target deep LSTM network, and outputting predicted subjective charging behavior data; and then calculating a charging load curve of the user to be analyzed according to the predicted subjective charging behavior data, and performing charging load prediction based on the charging load curve. Compared with the traditional method, the method adopts the autonomous learning capability and generalization capability of artificial intelligent algorithms such as clustering, deep learning and the like to carry out intelligent analysis on the association between objective influence factor data and subjective charging behavior data, thereby realizing the simulation analysis of the randomness of the charging behavior and further improving the accuracy of the charging load prediction.
Referring to fig. 1, a step flowchart of an electric vehicle charging demand modeling method provided by an embodiment of the present invention may specifically include the following steps:
step 101, obtaining a plurality of groups of subjective charging behavior data of a user, and clustering the subjective charging behavior data to obtain a corresponding charging behavior clustering result;
when electric vehicles are charged, each user has different use habits, and each electric vehicle has different charging time or charging electric quantity required by the configuration of the electric vehicle or the service life of a battery, so that for each user, the related data can be collected and subjective charging behavior data corresponding to the user can be formed, then a charging area can be used as a collection range to collect multiple groups of subjective charging behavior data of multiple users, and meanwhile, the subjective charging behavior data are used as one of charging load prediction models or input sample data of network training to optimally adjust the charging load prediction network of the electric vehicle so as to achieve higher accuracy when the charging load prediction of the user is carried out in the follow-up practice.
Specifically, the subjective charging behavior data may include an initial charging time, a charging duration, a required electric quantity and a controlled preference of a user of the electric vehicle, where the initial charging time is an initial time point of charging of the electric vehicle, the charging duration is a charging duration corresponding to a period from a start of charging to an end of charging of the electric vehicle, the required electric quantity represents an electric quantity required by the electric vehicle when charging is performed, the controlled preference is understood as a coefficient for measuring a degree of the electric vehicle receiving a charging regulation strategy, a value range of the controlled preference is between (0 and 1), and the closer the value is 1, the more the user receives the charging regulation strategy, and the charging regulation strategy is also called a charging control strategy, and specifically refers to a safety and protection requirement of a battery safety extremum such as a charging maximum voltage, a maximum allowable current, a temperature limit, a monomer extremum and the like.
After subjective charging behavior data of a user are collected, because the data volume is large and the numerical value between each group of data can be large, if the data are directly adopted for model or network training, the calculated volume is very large, and the disorder of the data can cause poor training effect, the K-means algorithm can be utilized for carrying out clustering analysis on the subjective charging behavior data of the user, and after a clustering result is obtained, the clustering result, the subjective charging behavior data and objective factor data are taken as input sample data for model or network training.
Wherein, K-means (K-means clustering algorithm, K mean clustering algorithm) is a clustering analysis algorithm for iterative solution, and the core principle steps are as follows: the data to be processed is divided into K groups in advance, K objects can be randomly selected from the data to be processed to serve as initial clustering centers, then the distance between each object in the data to be processed and each clustering center is calculated, each object is distributed to the closest clustering center based on the calculation result, and the clustering center and the distributed objects corresponding to the clustering center can represent one cluster.
In the iterative solution process, each time a sample is allocated, the clustering center of the cluster is recalculated according to the existing objects in the cluster, and the process is repeated until a certain termination condition is met, for example, no (or minimum number of) objects are reassigned to different clusters after the iterative calculation is performed, no (or minimum number of) clustering centers are changed again, the square sum of errors is minimum, or the maximum number of iterations is reached.
In the embodiment of the invention, for a plurality of groups of subjective charging behavior data, each group of subjective charging behavior data can be used as a clustering sample, then clustering calculation is carried out, and a clustering result is obtained through iteration.
In a specific implementation, clustering the multiple sets of subjective charging behavior data to obtain a corresponding charging behavior clustering result may include:
step S11: taking each subjective charging behavior data as a clustering sample, and initializing K initial centroids;
in the embodiment of the invention, a group of initial centroids is determined by heuristic search in order to obtain a better initial value because the initial value of K centroids has a larger influence on the convergence performance of an algorithm in the actual clustering process, and the search principle can be simply understood that when one cluster sample is randomly determined as a first initial centroid from all cluster samples, the distance from each remaining cluster sample to the first initial centroid is calculated, then the probability is selected based on the distance calculation, and the next initial centroid, in particular, the cluster sample farthest from the current initial centroid is selected as the next initial centroid, and so on until K initial centroids are selected.
Specifically, the step of initializing K initial centroids may include:
step S21: randomly selecting one cluster sample from all the cluster samples as a first initial centroid;
step S22: taking each cluster sample remained after the first centroid is selected as an observed value, and calculating the observed distance from each observed value to the first initial centroid;
step S23: according to the observation distance, randomly selecting one cluster sample from the rest cluster samples after the first initial centroid is selected as a second initial centroid, wherein the selection probability is as follows:
wherein d (x m ，c 1 ) For the observed value x m To a first initial centroid c 1 The observation distance between the two is divided by the first initial centroid c 1 The number of clustering samples outside, d (x j ，c 1 ) To divide the first initial centroid c 1 Outside observation value x j To a first initial centroid c 1 The observation distance between the two;
step S24: steps S22 to S23 are repeatedly performed until K initial centroids are selected.
Meanwhile, before cluster analysis, in order to eliminate the influence of different dimension differences on the cluster analysis, all the cluster samples can be subjected to standardized processing.
In a specific implementation, before computing the clustered samples, each clustered sample may be normalized by the following formula:
Wherein x is mn N-th dimensional data for an mth cluster sample; mu (mu) n The nth dimension data average value of all the clustered samples is obtained; sigma (sigma) n The nth dimensional data variance for all cluster samples,and (3) the n-th dimension data of the m-th clustering sample obtained after the normalization processing.
Step S12: calculating sample distances from each cluster sample to each initial centroid;
step S13: respectively distributing each clustered sample to a sample cluster corresponding to an initial centroid with the minimum sample distance, calculating a clustered average value of clustered samples in each sample cluster, and re-determining K optimized centroids based on the clustered average value;
the sample cluster is the result representation of cluster calculation, the average value of all data in the cluster is usually called the mass center of the cluster, in a two-dimensional plane, for a certain cluster, the abscissa of the mass center corresponding to the data point of the cluster is the abscissa average value of the data point of the cluster, the ordinate of the mass center is the ordinate average value of the data point of the cluster, and for a high-dimensional space, the core task of K-means is to find out K optimal mass centers according to the set number K of the clusters, and the data closest to the mass centers are respectively distributed to the clusters represented by the mass centers.
Step S14: repeating the steps S12 to S13 until the sample cluster allocation is unchanged or the maximum iteration number is reached;
Therefore, after the clustering calculation is performed, when the distribution of the sample clusters is no longer changed or the maximum iteration number has been reached, the sample clusters corresponding to the clusters can be determined, each sample cluster can be understood as a data cluster corresponding to a data value similar to the sample cluster, for example, referring to fig. 2, a schematic diagram of a clustered subjective charging behavior sample scatter point is shown, which is provided by the embodiment of the present invention, it can be seen that, by taking the required electric quantity, the charging duration and the initial charging time of an electric automobile as the coordinate axes of a high-dimensional space in a K-means algorithm, after the clustering, a plurality of sets of subjective charging behavior data can be divided into 5 data clusters in a scatter point distribution form, corresponding to 5 sample clusters, namely, a sample cluster-1, a sample cluster-2, a sample cluster-3, a sample cluster-4 and a sample cluster-5, and, in practical application, the number of samples output after the clustering calculation and each sample cluster corresponding to the actual cluster can be different according to the actual situation, for example, the number of samples can be output 6 samples, or the number of samples can be limited according to the actual situation, and the invention is not understood.
Step S15: and clustering samples in each sample cluster and cluster information corresponding to each sample cluster are used as a charging behavior clustering result.
After a plurality of sample clusters are determined through calculation, clustering samples in each sample cluster and cluster information corresponding to each sample cluster can be used as a charging behavior clustering result, wherein the cluster information can comprise a cluster serial number, a cluster center, the number of clusters and a cluster standard deviation of each sample cluster.
Therefore, a large amount of data can be classified and classified into a plurality of data groups with similar numerical values through K-means clustering calculation, so that the training effect is better in the subsequent network training process, and the accuracy of the charge load prediction of the network after training optimization is higher.
Step 102, constructing a deep LSTM network, obtaining multiple groups of objective factor data of the user, and carrying out association optimization training on the deep LSTM network by adopting the multiple groups of objective factor data, the charging behavior clustering result and the multiple groups of subjective charging behavior data to obtain a target deep LSTM network;
a deep LSTM (Long Short-Term Memory) network for predicting the charging load of an electric vehicle can then be constructed, wherein the deep LSTM network is a time-cycled neural network specifically designed to solve the Long-Term dependency problem of a general RNN (Recurrent Neural Network, cycled neural network), all RNNs have a chained form of repeating neural network modules, which have only a very simple structure, such as a tanh (hyperbolic tangent function ) layer, in a standard RNN.
For better explanation, referring to fig. 3, a schematic structural diagram of a deep LSTM network provided by an embodiment of the present invention is shown.
It can be seen that the deep LSTM network mainly comprises 1 input layer, 3 long and short term memory layers (LSTM-1, LSTM-2, and LSTM-3, respectively), 3 random discard layers (Dropout-1, dropout-2, and Dropout-3, respectively), 1 full link layer FC (Fully Connected layers), and 1 output layer (not shown in the figure), wherein the input layer is used as an initial layer, the middle is subjected to data feature processing of the triple long and short term memory layer-random discard layer, the feature output by the last random discard layer Dropout-3 is input to the full link layer FC for processing, and the result obtained after processing by the full link layer FC is output through the output layer as a processing result of the deep LSTM network.
The long-short-period memory layer can be regarded as a cyclic neural network, and is characterized by remembering long-term information, avoiding the problem of gradient disappearance or explosion, the random discarding layer is used for preventing the occurrence of fitting conditions in the data processing process, discarding some data randomly so as to improve the performance of the neural network, the full-connection layer is used for connecting all neuron output nodes of the last layer and integrating all the extracted features of the last layer, so that the mapping of input data from high dimension to low dimension and the output of results are realized.
After the deep LSTM network is constructed, the collected subjective charging behavior data, objective factor data and the clustering result obtained by the clustering calculation in the previous embodiment can be used for constructing training data of the deep LSTM network, the training data is used as network input, and the deep LSTM network is utilized for carrying out association training between objective factors and subjective charging behaviors.
As can be seen from the foregoing, the subjective charging behavior data of the user may include an initial charging time, a charging duration, a required electric quantity, and a controlled preference of the user of the electric vehicle, and the objective factor data may include an average monthly income of the user, a charging cost sensitivity, a daily driving mileage, a charging service unit price, an incentive patch, an outdoor weather condition, a charging facility capacity distribution density, and a daily type, where the average monthly income of the user refers to an average monthly income of the user over a period of time, such as an average monthly income of the last half year or the last year, the charging cost sensitivity refers to a sensitivity to the charging cost when the electric vehicle is charged each time, the charging cost sensitivity ranges between (0, 1), the closer the value is to 1, the higher the sensitivity of the user to the charging cost, the more influenced by the charging cost, the daily driving mileage refers to a mileage on the day, the charging service unit price is a single price spent when the electric vehicle is charged, the incentive patch is a part for counteracting the charging cost, the outdoor weather condition is mainly classified as rainfall or non-rainfall, the charging facility capacity distribution density refers to a main daily distribution density of the area when the electric vehicle is charged, and the daily cost distribution is a main day, or a holiday density.
In a specific implementation, performing association optimization training on the deep LSTM network by using multiple sets of objective factor data, charging behavior clustering results and multiple sets of subjective charging behavior data to obtain a target deep LSTM network may include:
firstly, setting parameters (such as the number of neurons of each nerve layer, nerve layer parameters and the like in a deep LSTM network structure, and super parameters of each relevant network training during network training, such as division of data sets, initial learning rate, gradient threshold and the like), taking average month income, charging expense sensitivity, daily driving mileage, charging service unit price, excitation subsidy, outdoor weather conditions, charging facility capacity distribution density, daily type and charging behavior clustering result of a user as input sample characteristics, and taking initial charging time, charging duration, required electric quantity and controlled preference as input sample labels;
and then, adopting input sample characteristics and input sample labels to perform association training between subjective behavior and objective factors of user charging on the deep LSTM network after parameter setting, and performing network optimization adjustment on the deep LSTM network after association training to obtain the target deep LSTM network.
Further, by adopting the input sample characteristics and the input sample labels, the correlation training between the subjective behavior and the objective factor of the user charging can be performed on the deep LSTM network after the parameter setting, which can be as follows:
step S31: carrying out association training between subjective charging behaviors and objective factors of users on the deep LSTM network subjected to parameter setting by adopting input sample characteristics and input sample labels;
step S32: in the associated training process, the learning rate of the deep LSTM network is adjusted according to a piecewise constant attenuation mode, and gradient correction is carried out on the deep LSTM network based on a self-adaptive momentum random optimization mode, so that an optimized deep LSTM network is obtained;
the learning rate of the deep LSTM network can be adjusted according to the piecewise constant attenuation mode, which can be specifically as follows: in the iteration process of network training, the current learning rate of the deep LSTM network is attenuated to be a preset multiple of the previous learning rate after the preset iteration times, wherein the attenuated learning rate is smaller than the learning rate before the attenuation.
For example, each time T1, T2,..tm iterations attenuate the learning rate to the original ζ1, ζ2,..ζm times, where Tm represents the number of iterations, ζm <1 is the attenuation multiple, and Tm and ζm are all hyper-parameters set according to experience or actual requirements.
Gradient correction is carried out on the deep LSTM network based on a self-adaptive momentum random optimization mode, and the method can be specifically as follows: and obtaining parameter updating difference values required by iteratively updating the deep LSTM network by adopting a self-adaptive momentum random optimization mode, wherein the calculation formula is as follows:
M t ＝β 1 M t-1 +(1-β 1 )g t
G t ＝β 2 G t-1 +(1-β 2 )g t ⊙g t
wherein beta is 1 Is the first moment attenuation coefficient, beta 2 Is a second moment attenuation coefficient, alpha is a step factor, M t Weighted average of the current index for first moment, G t Is the current exponentially weighted average of the second moment, M t-1 Is M t Corresponding last exponentially weighted average, G t-1 Is G t The corresponding last exponentially weighted average, g t For the current gradient, t is the current iteration number, θ is the parameter to be updated, and Δθ t Updating the difference value for the parameter corresponding to the current required parameter, wherein, as indicated by the ";
β1 and β2 can also be understood as the rate of decay of the moving average, for example, β1 can take on a value of 0.9, β2 can take on a value of 0.99, α can control the rate of update of the weights (e.g., can take on a value of 0.001), a larger value (e.g., 0.3) will have a faster initial learning before the rate of update is learned, and a smaller value (e.g., 1.0E -5 ) Will converge the training to better performance.
When the parameter updating difference value is calculated through the formula, gradient correction can be carried out on the corresponding parameters of the deep LSTM network based on the parameter updating difference value to obtain an optimized deep LSTM network, then the accuracy of the optimized deep LSTM network obtained by sample verification can be selected, and the structure of the deep LSTM network is optimized and adjusted.
Step S33: randomly extracting data with preset proportion from input sample characteristics to be used as a characteristic verification sample, and randomly extracting data with preset proportion from input sample labels to be used as a label verification sample;
for example, data is randomly extracted from input sample features according to a ratio of 1/10 to be used as a feature verification sample, and data is randomly extracted from input sample tags according to a ratio of 1/10 to be used as a tag verification sample, it is understood that when network training is performed, the input sample data is generally divided into a training set and a test verification set, for example, in the embodiment of the invention, the input sample features are randomly extracted from the input sample features according to a ratio of 1/10 to be used as the feature verification sample, or the input sample features are divided into a feature training sample set and a feature verification sample set according to a ratio of 9:1, the feature training sample set is adopted in the early stage to perform network training, and the feature verification sample set is adopted in the later stage to perform accuracy verification on the trained network.
Step S34: and carrying out accuracy verification on the optimized deep LSTM network by adopting a feature verification sample and a label verification sample, and carrying out network optimization adjustment on the optimized deep LSTM network to obtain the target deep LSTM network.
Besides the adoption of verification sample data to verify the accuracy of the optimized deep LSTM network, the optimized deep LSTM network can be further optimized and adjusted, and the basic principle of the optimized and adjusted is as follows: for subjective charging behavior data of a certain user, comparing root mean square errors between an optimized deep LSTM network and other algorithms such as multiple linear regression, regression decision trees, support vector machines and shallow neural networks, if the root mean square error of one algorithm is smaller than that of the optimized deep LSTM network, adopting the algorithm to replace the deep LSTM network, fitting the subjective charging behavior data of the user, for example, inputting 5 groups of subjective charging behavior data, assuming that only 2 groups of data are better through calculation, only replacing the 2 groups of algorithms, and still adopting the LSTM network to fit the other groups, for example, for the 1 groups of data input, assuming that the root mean square error of more than 2 other algorithms is smaller than that of the optimized deep LSTM network through calculation, and selecting the algorithm with the smallest error value from the root mean square errors to replace.
In a specific implementation, performing network optimization adjustment on the deep LSTM network after the association training may include:
Respectively inputting subjective charging behavior data into an optimized deep LSTM network and at least one preset comparison algorithm model, and calculating a first root mean square error corresponding to the optimized deep LSTM network and a second root mean square error corresponding to the at least one preset comparison algorithm model;
if the second root mean square error is smaller than the first root mean square error, determining the subjective charging behavior data as subjective charging behavior data to be optimized, and adopting a preset comparison algorithm model corresponding to the second root mean square error smaller than the first root mean square error to replace an optimized deep LSTM network to perform data fitting on the subjective charging behavior data to be optimized;
if at least two second root mean square errors are smaller than the first root mean square errors, determining subjective charging behavior data as subjective charging behavior data to be optimized, replacing an optimized deep LSTM network by a preset comparison algorithm model corresponding to the minimum second root mean square errors, and performing data fitting on the subjective charging behavior data to be optimized;
if the second root mean square error is greater than or equal to the first root mean square error, continuing to adopt the optimized deep LSTM network to perform data fitting on the subjective charging behavior data;
The root mean square error is calculated as follows:
wherein RMSE is root mean square error, m t A number of test set samples representing a model or network,representing the output value of the model or network, y i Representing the actual value of the test set samples.
Therefore, the result output by the network has smaller error and higher accuracy by further carrying out optimization adjustment on the optimized deep LSTM network, so that the effect is better and the accuracy is higher when the deep LSTM network is adopted for charging load prediction.
Step 103, obtaining objective factor data to be detected of a user to be analyzed, inputting the objective factor data to be detected into the target deep LSTM network, and outputting predicted subjective charging behavior data;
after the target deep LSTM network is obtained through the association optimization training, the target deep LSTM network can be adopted to conduct prediction processing on data to be analyzed, specifically, objective factor data to be analyzed of a user to be analyzed can be obtained, the objective factor data to be tested is input into the target deep LSTM network, and the predicted subjective charging behavior data is output.
And 104, calculating a charging load curve of the user to be analyzed according to the predicted subjective charging behavior data, and performing charging load prediction based on the charging load curve.
Then, the charging load curve of the user may be calculated based on the predicted subjective charging behavior data, where as can be seen from the foregoing, the subjective charging behavior data includes a starting charging time, a charging duration, a required electric quantity and a controlled preference, and correspondingly, the predicted subjective charging behavior data may include a predicted starting charging time, a predicted charging duration, a predicted required electric quantity and a predicted controlled preference, and then the charging load curve of the user to be analyzed may be calculated according to the predicted subjective charging behavior data:
calculating by adopting the predicted required electric quantity and the predicted charging time length to obtain an average charging load, wherein the calculation formula is as follows:
calculating by adopting the predicted controlled preference and the average charging load, obtaining the controlled charging load, and drawing a corresponding charging load curve, wherein the calculation formula is as follows:
P c ′＝(1-R p )×P c
wherein P is c For average charging load, ΔE is predicted required power, Δt is predicted charging duration, P c ' is a controlled post-charge load, R p To predict controlled preferences;
the charge load curve is expressed as: at [ T ] b ,T b +△t]In the time period, the charging load corresponding to the user to be analyzed is P c ' in [ T ] b ,T b +△t]In the time period except the time period, the charging load corresponding to the user to be analyzed is zero, T b To predict the initial charge time.
Therefore, the charging load prediction can be intuitively performed on the user needing to perform charging analysis through the charging load curve.
In the embodiment of the invention, aiming at the charge load demand prediction of the electric automobile, the invention provides an electric automobile charge demand modeling method, which comprises the steps of firstly acquiring a plurality of groups of subjective charge behavior data of a user, clustering the subjective charge behavior data to obtain a corresponding charge behavior clustering result, so that a large amount of data can be classified and classified into a plurality of data groups with similar values by clustering complex and various data, the training effect is better in the subsequent network training process, and the accuracy of the charge load prediction of a network after training optimization is higher; then constructing a deep LSTM network, acquiring a plurality of groups of objective factor data of a user, carrying out associated optimization training on the deep LSTM network by adopting a plurality of groups of objective factor data, charging behavior clustering results and a plurality of groups of subjective charging behavior data, and obtaining a target deep LSTM network, so that the charging requirement of the user is measured by quantifying the subjective charging behavior data, and meanwhile, the influence of a plurality of objective factors on the charging requirement is considered, so that the model can be more suitable for a practical complex charging scene; then obtaining objective factor data to be detected of a user to be analyzed, inputting the objective factor data to a target deep LSTM network, and outputting predicted subjective charging behavior data; and then calculating a charging load curve of the user to be analyzed according to the predicted subjective charging behavior data, and performing charging load prediction based on the charging load curve. Compared with the traditional method, the method adopts the autonomous learning capability and generalization capability of artificial intelligent algorithms such as clustering, deep learning and the like, and performs intelligent analysis on the correlation between objective influence factor data and subjective charging behavior data, so that the simulation analysis of the randomness of the charging behavior is realized, and the accuracy of the charging load prediction is further improved.
Referring to fig. 4, a step flowchart of another method for modeling electric vehicle charging requirements according to an embodiment of the present invention is shown, which specifically may include the following steps:
step 401, obtaining multiple groups of subjective charging behavior data of a user, and clustering the multiple groups of subjective charging behavior data to obtain a corresponding charging behavior clustering result;
since this step is described in more detail in step 101 of the foregoing embodiment, details thereof will not be described herein, and reference may be made to the related content in the foregoing embodiment.
Illustratively, with reference to fig. 2, assuming 10000 sets of subjective charging behavior data, 5 sample clusters can be obtained after clustering calculation, specifically as follows:
| cluster serial number | Cluster center | Cluster number | Standard deviation of |
| 1 | (710.70,54.09,0.62,0.70) | 1558 | (79.88,14.18,0.09,0.10) |
| 2 | (718.53,59.96,0.36,0.71) | 1831 | (72.11,1.29,0.08,0.12) |
| 3 | (319.10,19.53,0.76,0.19) | 2241 | (205.02,4.32,0.09,0.30) |
| 4 | (690.13,121.07,0.56,0.79) | 1847 | (99.88,7.95,0.10,0.17) |
| 5 | (1102.39,34.01,0.51,0.00) | 2523 | (72.66,9.45,0.14,0.00) |
Table 1: clustered cluster results
In the table, the cluster serial numbers represent serial numbers of corresponding sample clusters, the corresponding data in the vectors of the cluster center and the standard deviation are (initial charging time, charging duration, required electric quantity, controlled preference), taking the cluster center (710.70,54.09,0.62,0.70) of the first sample cluster as an example, wherein for the initial charging time, 24 hours a day is taken as 1 point every 1 minute, 24×60=1440 points in total, 710.70 represents 710.70 minutes from 0:00; for the charge duration, it is expressed in minutes, as 54.09 for 54.09 minutes; for the required electric quantity, the required electric quantity value in the table refers to the percentage of the required electric quantity to the full electric quantity, for example, 0.36 refers to the required charging quantity to be 36% of the full electric quantity; the controlled preference value is (0, 1), and the controlled preference value in the data is 0.70, which indicates that the degree of the user accepting the charge regulation strategy is moderately high.
As can be seen from fig. 2 and table 1, after clustering, most data can be distributed to the corresponding cluster center, but some data are still in a scattered state (such as some data points scattered outside the sample cluster data in fig. 2), these data are data with larger deviation, and can be ignored or counted together in the sample cluster closest to the sample cluster, and meanwhile, it can be seen intuitively that the sample cluster-1, the sample cluster-2 and the sample cluster-4 are the sample clusters closer to the cluster center, so that from the two-dimensional view, some data overlap, and it can be seen that the user charging behaviors corresponding to the subjective charging behavior data samples in the sample cluster-1, the sample cluster-2 and the sample cluster-4 are closer.
Step 402, constructing a deep LSTM network, obtaining multiple sets of objective factor data of the user, and performing association optimization training on the deep LSTM network by adopting the multiple sets of objective factor data, the charging behavior clustering result and the multiple sets of subjective charging behavior data to obtain a target deep LSTM network;
since the step 102 of the foregoing embodiment is described in more detail, details thereof will not be described herein, and reference may be made to the related content of the foregoing embodiment.
For better illustration, the following table gives the input data sample structure of the deep LSTM network:
table 2: input data sample structure for deep LSTM network
As can be seen from the table, the input data sample structure of the deep LSTM network in the embodiment of the present invention may be divided into two parts, one part is an input sample feature, and the other part is an input sample tag, where specifically, an average month income, a charging cost sensitivity, a daily driving mileage, a charging service unit price, an excitation patch, an outdoor weather condition, a charging facility capacity distribution density, a daily type, and a charging behavior clustering result (a cluster center corresponding to a cluster to which the user belongs) are taken as input sample features, an actual start charging time, a charging duration (an actual end charging time minus an actual start charging time), a required electric quantity (an SOC at end minus an initial SOC) are taken as input sample tags, and in order to make the data more reference, it is to be noted that, in practical application, a person skilled in the art may decide whether to take two types of data, i.e. the newly increased driving mileage and the current charging total cost, as input sample tags according to actual situations, which it can be understood that the present invention does not limit this.
In particular, for the value of outdoor weather conditions, when the outdoor weather conditions are rainfall, the value is 0, and when the outdoor weather conditions are non-rainfall, the value is 1; for the value of the day type, when the day type is a working day, the value is 0, when the day type is a weekend, the value is 1, and when the day type is a holiday, the value is 2; SOC (State Of Charge) refers to the current remaining capacity Of the battery, and is an important parameter in the use process Of the power battery, and is usually expressed in percentage.
Illustratively, the following table is a deep LSTM network structure parameter provided in an embodiment of the present invention:
table 3: deep LSTM network structure parameters
As can be seen from fig. 3, in the embodiment of the present invention, the neural layer of the deep LSTM network structure has 9 layers, and the name and type of each neural layer and the number of neurons corresponding to each neural layer need to refer to a table, which is not described herein, where for the setting of the neural layer parameters of the random discard layer Dropout, the probability of removing the input element is set to 0.2, that is, in the random discard layer Dropout, the probability of discarding each input element is 0.2.
Further, the following table is a deep LSTM network training hyper-parameter provided in the embodiment of the present invention:
| Parameters (parameters) | Value of | Parameters (parameters) | Value of |
| Total number of data samples | 10000 | Parameter initialization | Glorot initialization |
| Number of training data samples | 9000 | Gradient threshold | 1 |
| Number of test data samples | 1000 | Small batch Scale mini-batch | 128 |
| Input data set Scale (training) | 11×9000 | Initial learning rate | 0.0001 |
| Target data set size (training) | 7×9000 | Gradient correction | Adam algorithm |
| Input data set size (test) | 11×1000 | Loss function | L2 mean square error |
| Target data set size (test) | 7×1000 | Training period | 250 |
| LSTM state activation function | tanh | Learning rate decay | Piecewise constant decay |
| LSTM gate activation function | sigmoid | Learning rate decay factor | 0.85 |
Table 4: deep LSTM network training superparameter
As can be seen from the table, in the embodiment of the present invention, when training optimization is performed on the deep LSTM network, the total number of input data samples is 10000, and the total number of data samples is divided into the number of training data samples and the number of test data samples according to a ratio of 9:1, for the training process, the input data set size is set to 11×9000, the target data set size is set to 7×9000, for the testing process, the input data set size is set to 11×1000, and the target data set size is set to 7×1000.
The state activation function of the LSTM adopts a tanh function, namely a hyperbolic tangent function; the gate activation function of LSTM uses a sigmoid function, i.e., an S-shaped growth curve function.
The parameter initialization adopts a gloriot initialization method, wherein gloriot is a common deep learning parameter initialization strategy, and specific contents are referred to gloriot related contents in the prior art, and the invention is not repeated.
The gradient threshold is set to 1; the small batch scale mini-batch was set to 128.
The initial learning rate was set to 0.0001; gradient correction adopts Adam (Adaptive momentum, a first-order optimization algorithm), wherein Adam is a self-adaptive momentum random optimization method (A method for stochastic optimization), is often used as an optimizer algorithm in deep learning, is a first-order optimization algorithm capable of replacing the traditional random gradient descent process, and can iteratively update neural network weights based on training data.
The loss function adopts L2 mean square error, wherein L2 norm refers to calculating the square sum of each element of the vector and then solving the square root, and mean square error refers to calculating the square sum of all (predicted value-actual value) and dividing the sum by the total number; the training period is set to 250.
The learning rate attenuation adopts a piecewise constant attenuation mode; the learning rate decay factor is set to 0.85.
Step 403, obtaining objective factor data to be tested of a user to be analyzed, inputting the objective factor data to be tested into the target deep LSTM network, and outputting predicted subjective charging behavior data;
Since this step is described in more detail in step 103 of the foregoing embodiment, details thereof will not be described herein, and reference may be made to the related content in the foregoing embodiment.
Step 404, calculating a charging load curve of the user to be analyzed according to the predicted subjective charging behavior data, and performing charging load prediction based on the charging load curve;
since the step 104 of the foregoing embodiment is described in more detail, details thereof will not be described herein, and reference may be made to the related content of the foregoing embodiment.
And 405, superposing charging load curves of all users to be analyzed in the area to be analyzed to obtain a total charging load curve, and carrying out overall charging load prediction on the electric automobile in the area to be analyzed based on the total charging load curve.
The electric vehicle charging demand modeling method provided by the embodiment of the invention is not only suitable for predicting the charging load demand of a single user, but also can predict the charging load demand of a certain area corresponding to a plurality of electric vehicles, namely, the charging load curves of all users in the area to be analyzed can be overlapped to obtain the total charging load curve of the area to be analyzed, and the charging load prediction is performed on the area to be analyzed based on the total charging load curve.
In a specific implementation, the user to be analyzed is one of the users in the area to be analyzed of the electric automobile, and after the charging load curves of the single user are calculated, the charging load curves of all the users to be analyzed in the area to be analyzed can be overlapped to obtain a total charging load curve, and the electric automobile in the area to be analyzed is subjected to overall charging load prediction based on the total charging load curve.
In the embodiment of the invention, aiming at the charge load demand prediction of the electric automobile, the invention provides an electric automobile charge demand modeling method, which comprises the steps of firstly acquiring a plurality of groups of subjective charge behavior data of a user, clustering the subjective charge behavior data to obtain a corresponding charge behavior clustering result, so that a large amount of data can be classified and classified into a plurality of data groups with similar values by clustering complex and various data, the training effect is better in the subsequent network training process, and the accuracy of the charge load prediction of a network after training optimization is higher; then constructing a deep LSTM network, acquiring a plurality of groups of objective factor data of a user, carrying out associated optimization training on the deep LSTM network by adopting a plurality of groups of objective factor data, charging behavior clustering results and a plurality of groups of subjective charging behavior data, and obtaining a target deep LSTM network, so that the charging requirement of the user is measured by quantifying the subjective charging behavior data, and meanwhile, the influence of a plurality of objective factors on the charging requirement is considered, so that the model can be more suitable for a practical complex charging scene; then obtaining objective factor data to be detected of a user to be analyzed, inputting the objective factor data to a target deep LSTM network, and outputting predicted subjective charging behavior data; and then calculating a charging load curve of the user to be analyzed according to the predicted subjective charging behavior data, and predicting the charging load based on the charging load curve, so that compared with the traditional method, the method adopts the autonomous learning capability and the generalization capability of artificial intelligent algorithms such as clustering, deep learning and the like, and performs intelligent analysis on the correlation between objective influence factor data and the subjective charging behavior data, thereby realizing the simulation analysis of the randomness of the charging behavior and further improving the accuracy of the charging load prediction. Meanwhile, after the charging load curves of the single user are obtained through calculation, the charging load curves of a plurality of users in a certain area can be overlapped to obtain a total charging load curve corresponding to the area, and the electric vehicle in the area can be subjected to overall charging load prediction based on the total charging load curve, so that the electric vehicle charging load demand analysis of the single user is realized, and the electric vehicle charging load total demand analysis of a certain area can be realized.
Referring to fig. 5, a block diagram of an electric vehicle charging demand modeling apparatus according to an embodiment of the present invention may specifically include:
the clustering calculation module 501 is configured to obtain multiple sets of subjective charging behavior data of a user, and cluster the multiple sets of subjective charging behavior data to obtain a corresponding charging behavior clustering result;
the association optimization training module 502 is configured to construct a deep LSTM network, obtain multiple sets of objective factor data of the user, and perform association optimization training on the deep LSTM network by using the multiple sets of objective factor data, the charging behavior clustering result, and the multiple sets of subjective charging behavior data, so as to obtain a target deep LSTM network;
the predicted subjective charging behavior data output module 503 is configured to obtain objective factor data to be detected of a user to be analyzed, input the objective factor data to be detected to the target deep LSTM network, and output predicted subjective charging behavior data;
and the charging load prediction module 504 is configured to calculate a charging load curve of the user to be analyzed according to the predicted subjective charging behavior data, and perform charging load prediction based on the charging load curve.
In an alternative embodiment, the cluster calculation module 501 includes:
an initial centroid initializing module, configured to execute step S11: taking each subjective charging behavior data as a clustering sample, and initializing K initial centroids;
a sample distance calculation module, configured to execute step S12: calculating sample distances from each clustering sample to each initial centroid;
an optimized centroid redetermining module for performing step S13: respectively distributing each clustered sample to a sample cluster corresponding to an initial centroid with the smallest sample distance, calculating a clustered average value of clustered samples in each sample cluster, and re-determining K optimized centroids based on the clustered average value;
the cluster repetition execution module is configured to execute step S14: repeating the steps S12 to S13 until the sample cluster allocation is unchanged or the maximum iteration number is reached;
the charging behavior clustering result determining module is configured to execute step S15: and clustering samples in each sample cluster and cluster information corresponding to each sample cluster are used as a charging behavior clustering result.
In an alternative embodiment, the initial centroid initializing module comprises:
the first initial centroid selection module is configured to execute step S21: randomly selecting one cluster sample from the cluster samples as a first initial centroid;
The observation distance calculation module is configured to execute step S22: taking each cluster sample remained after the first centroid is selected as an observation value, and calculating the observation distance from each observation value to the first initial centroid;
the second initial centroid selection module is configured to execute step S23: according to the observation distance, randomly selecting one cluster sample from the cluster samples remained after the first initial centroid is selected as a second initial centroid, wherein the selection probability is as follows:
wherein d (x m ，c 1 ) For the observed value x m To a first initial centroid c 1 The observation distance between the two is divided by the first initial centroid c 1 The number of clustering samples outside, d (x j ，c 1 ) To divide the first initial centroid c 1 Outside observation value x j To a first initial centroid c 1 The observation distance between the two;
the initial centroid selection repeated execution module is configured to execute step S24: steps S22 to S23 are repeatedly performed until K initial centroids are selected.
In an alternative embodiment, the apparatus further comprises:
the cluster sample normalization processing module is used for performing normalization processing on each cluster sample through the following formula:
wherein x is mn N-th dimensional data for an mth cluster sample; mu (mu) n The nth dimension data average value of all the clustered samples is obtained; sigma (sigma) n The nth dimensional data variance for all cluster samples,and (3) the n-th dimension data of the m-th clustering sample obtained after the normalization processing.
In an alternative embodiment, the subjective charging behavior data includes a starting charging time, a charging duration, a required power amount, and a controlled preference, the objective factor data includes a user average month income, a charging fee sensitivity, a daily driving mileage, a charging service unit price, an incentive subsidy, an outdoor weather condition, a charging facility capacity distribution density, and a daily type, and the association optimization training module 502 includes:
the input sample data determining module is used for setting parameters of the deep LSTM network, taking average month income of the user, charging expense sensitivity, daily driving mileage, charging service unit price, incentive subsidy, outdoor weather condition, charging facility capacity distribution density, daily type and charging behavior clustering result as input sample characteristics, and taking the initial charging time, charging duration, required electric quantity and controlled preference as input sample labels;
and the correlation optimization training sub-module is used for carrying out correlation training between subjective behavior and objective factors of user charging on the deep LSTM network after parameter setting by adopting the input sample characteristics and the input sample labels, and carrying out network optimization adjustment on the deep LSTM network after correlation training to obtain a target deep LSTM network.
In an alternative embodiment, the association optimization training submodule includes:
the association training module is used for carrying out association training between subjective behavior and objective factors of user charging on the deep LSTM network subjected to parameter setting by adopting the input sample characteristics and the input sample labels;
the optimized deep LSTM network training module is used for adjusting the learning rate of the deep LSTM network according to a piecewise constant attenuation mode in the associated training process, and carrying out gradient correction on the deep LSTM network based on a self-adaptive momentum random optimization mode to obtain an optimized deep LSTM network;
the verification sample extraction module is used for randomly extracting data with preset proportion from the input sample characteristics to be used as a characteristic verification sample, and randomly extracting data with preset proportion from the input sample labels to be used as a label verification sample;
and the network optimization adjustment module is used for verifying the accuracy of the optimized deep LSTM network by adopting the characteristic verification sample and the label verification sample, and carrying out network optimization adjustment on the optimized deep LSTM network to obtain a target deep LSTM network.
In an alternative embodiment, the optimized deep LSTM network training module includes:
And the learning rate adjustment module is used for attenuating the current learning rate of the deep LSTM network to be a preset multiple of the previous learning rate every time a preset iteration number is passed in the iteration process of network training, wherein the attenuated learning rate is smaller than the learning rate before attenuation.
In an alternative embodiment, the optimized deep LSTM network training module includes:
the parameter updating difference value calculating module is used for obtaining parameter updating difference values required by iteratively updating the deep LSTM network by adopting a self-adaptive momentum random optimization mode, and the calculating formula is as follows:
M t ＝β 1 M t-1 +(1-β 1 )g t
G t ＝β 2 G t-1 +(1-β 2 )g t ⊙g t
wherein beta is 1 Is the first moment attenuation coefficient, beta 2 Is a second moment attenuation coefficient, alpha is a step factor, M t Weighted average of the current index for first moment, G t Is the current exponentially weighted average of the second moment, mt-1 is M t Corresponding last exponentially weighted average, G t-1 Is G t The corresponding last exponentially weighted average, g t For the current gradient, t is the current iteration number, θ is the parameter to be updated, and Δθ t For the parameter update difference corresponding to the current needed update parameter, as indicated by weightPerforming an exclusive OR operation;
and the gradient correction module is used for carrying out gradient correction on the parameters corresponding to the deep LSTM network based on the parameter updating difference value.
In an alternative embodiment, the association optimization training submodule includes:
the root mean square error calculation module is used for respectively inputting the subjective charging behavior data into the optimized deep LSTM network and at least one preset comparison algorithm model, and calculating a first root mean square error corresponding to the optimized deep LSTM network and a second root mean square error corresponding to at least one preset comparison algorithm model;
the first algorithm replacing module is used for determining the subjective charging behavior data as subjective charging behavior data to be optimized if a second root mean square error is smaller than the first root mean square error, replacing the optimized deep LSTM network by adopting a preset comparison algorithm model corresponding to the second root mean square error smaller than the first root mean square error, and performing data fitting on the subjective charging behavior data to be optimized;
the second algorithm replacing module is used for determining the subjective charging behavior data as subjective charging behavior data to be optimized if at least two second root mean square errors are smaller than the first root mean square errors, replacing the optimized deep LSTM network by adopting a preset comparison algorithm model corresponding to the minimum second root mean square errors, and performing data fitting on the subjective charging behavior data to be optimized;
The optimized deep LSTM network selection module is used for continuing to adopt the optimized deep LSTM network to perform data fitting on the subjective charging behavior data if the second root mean square error is larger than or equal to the first root mean square error;
the root mean square error is calculated as follows:
/>
wherein RMSE is root mean square error, m t Representation modelThe number of test set samples of a network or a network,representing the output value of the model or network, y i Representing the actual value of the test set samples.
In an alternative embodiment, the predicted subjective charging behavior data includes a predicted starting charging time, a predicted charging duration, a predicted required power, and a predicted controlled preference, and the charging load prediction module 504 includes:
the average charging load calculation module is used for calculating the predicted required electric quantity and the predicted charging duration to obtain an average charging load, and the calculation formula is as follows:
the controlled charging load calculation module is used for calculating the average charging load by adopting the predicted controlled preference, obtaining the controlled charging load, and drawing a corresponding charging load curve, wherein the calculation formula is as follows:
P c ′＝(1-R p )×P c
wherein P is c For average charging load, ΔE is predicted required power, Δt is predicted charging duration, P c ' is a controlled post-charge load, R p To predict controlled preferences;
the charge load curve is expressed as: at [ T ] b ,T b +△t]In the time period, the charging load corresponding to the user to be analyzed is P c ' in [ T ] b ,T b +△t]The charging load corresponding to the user to be analyzed is zero and T is the time period except the time period b To predict the initial charge time.
In an alternative embodiment, the user to be analyzed is one of the users in the area to be analyzed of the electric automobile, and the apparatus further includes:
the system comprises a total charging load curve generation module, a total charging load curve prediction module and a total charging load prediction module, wherein the total charging load curve generation module is used for superposing charging load curves of all users to be analyzed in an area to be analyzed to obtain a total charging load curve, and carrying out overall charging load prediction on electric vehicles in the area to be analyzed based on the total charging load curve.
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 electric vehicle charging demand modeling method according to any embodiment of the invention according to the instructions in the program codes.
The embodiment of the application also provides a computer readable storage medium, which is used for storing program codes, and the program codes are used for executing the electric vehicle charging demand modeling method according to any embodiment of the application.
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 the present application, it should be understood that the disclosed systems, devices, and methods may be implemented in other manners. 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.
In addition, the embodiment of the invention may involve the use of user data, and in practical application, the user specific personal data may be used in the scheme described herein within the scope allowed by the applicable legal regulations under the condition of meeting the applicable legal regulations of the country (for example, the user explicitly agrees to the user, and the user is informed practically, etc.).
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 (14)
1. The electric automobile charging demand modeling method is characterized by comprising the following steps of:
acquiring a plurality of groups of subjective charging behavior data of a user, and clustering the subjective charging behavior data to obtain a corresponding charging behavior clustering result;
constructing a deep LSTM network, acquiring a plurality of groups of objective factor data of the user, and performing association optimization training on the deep LSTM network by adopting the plurality of groups of objective factor data, the charging behavior clustering result and the plurality of groups of subjective charging behavior data to acquire a target deep LSTM network;
Obtaining objective factor data to be detected of a user to be analyzed, inputting the objective factor data to be detected into the target deep LSTM network, and outputting predicted subjective charging behavior data;
and calculating a charging load curve of the user to be analyzed according to the predicted subjective charging behavior data, and performing charging load prediction based on the charging load curve.
2. The method for modeling charging requirements of an electric vehicle according to claim 1, wherein the clustering the plurality of sets of subjective charging behavior data to obtain a corresponding charging behavior clustering result includes:
step S11: taking each subjective charging behavior data as a clustering sample, and initializing K initial centroids;
step S12: calculating sample distances from each clustering sample to each initial centroid;
step S13: respectively distributing each clustered sample to a sample cluster corresponding to an initial centroid with the smallest sample distance, calculating a clustered average value of clustered samples in each sample cluster, and re-determining K optimized centroids based on the clustered average value;
step S14: repeating the steps S12 to S13 until the sample cluster allocation is unchanged or the maximum iteration number is reached;
Step S15: and clustering samples in each sample cluster and cluster information corresponding to each sample cluster are used as a charging behavior clustering result.
3. The method of modeling charging demand for an electric vehicle of claim 2, wherein initializing K initial centroids comprises:
step S21: randomly selecting one cluster sample from the cluster samples as a first initial centroid;
step S22: taking each cluster sample remained after the first centroid is selected as an observation value, and calculating the observation distance from each observation value to the first initial centroid;
step S23: according to the observation distance, randomly selecting one cluster sample from the cluster samples remained after the first initial centroid is selected as a second initial centroid, wherein the selection probability is as follows:
wherein d (x m ，c 1 ) For the observed value x m To a first initial centroid c 1 The observation distance between the two is divided by the first initial centroid c 1 The number of clustering samples outside, d (x j ，c 1 ) To divide the first initial centroid c 1 Outside observation value x j To a first initial centroid c 1 The observation distance between the two;
step S24: steps S22 to S23 are repeatedly performed until K initial centroids are selected.
4. A method of modeling electric vehicle charging demand according to claim 2 or 3, characterized in that before calculating the clustered samples, the method further comprises:
The normalization process is performed on each of the clustered samples by the following formula:
wherein x is mn N-th dimensional data for an mth cluster sample; mu (mu) n The nth dimension data average value of all the clustered samples is obtained; sigma (sigma) n The nth dimensional data variance for all cluster samples,and (3) the n-th dimension data of the m-th clustering sample obtained after the normalization processing.
5. The method of modeling charging demand for an electric vehicle according to claim 2, wherein the subjective charging behavior data includes a starting charging time, a charging duration, a required electric quantity, and a controlled preference, the objective factor data includes a user average month income, a charging fee sensitivity, a daily driving mileage, a charging service unit price, an incentive subsidy, an outdoor weather condition, a charging facility capacity distribution density, and a daily type, and the performing the associated optimization training on the deep LSTM network using the plurality of objective factor data, the charging behavior clustering result, and the plurality of subjective charging behavior data, to obtain a target deep LSTM network, includes:
setting parameters of the deep LSTM network, taking average month income of the user, the charging expense sensitivity, the daily driving mileage, the charging service unit price, the incentive subsidy, the outdoor weather condition, the charging facility capacity distribution density, the daily type and the charging behavior clustering result as input sample characteristics, and taking the initial charging time, the charging duration, the required electric quantity and the controlled preference as input sample labels;
And carrying out association training between subjective behavior and objective factors of user charging on the deep LSTM network after parameter setting by adopting the input sample characteristics and the input sample labels, and carrying out network optimization adjustment on the deep LSTM network after association training to obtain a target deep LSTM network.
6. The method for modeling the charging requirement of the electric vehicle according to claim 5, wherein the training the correlation between the subjective behavior and the objective factor of the user charging on the deep LSTM network after the parameter setting by using the input sample feature and the input sample label includes:
carrying out association training between subjective behavior and objective factors of user charging on the deep LSTM network subjected to parameter setting by adopting the input sample characteristics and the input sample labels;
in the associated training process, the learning rate of the deep LSTM network is adjusted according to a piecewise constant attenuation mode, and gradient correction is carried out on the deep LSTM network based on a self-adaptive momentum random optimization mode, so that an optimized deep LSTM network is obtained;
randomly extracting data with preset proportion from the input sample characteristics to be used as a characteristic verification sample, and randomly extracting data with preset proportion from the input sample labels to be used as a label verification sample;
And carrying out accuracy verification on the optimized deep LSTM network by adopting the characteristic verification sample and the label verification sample, and carrying out network optimization adjustment on the optimized deep LSTM network to obtain a target deep LSTM network.
7. The method of claim 6, wherein the adjusting the learning rate of the deep LSTM network according to the piecewise constant decay manner comprises:
in the iteration process of network training, the current learning rate of the deep LSTM network is attenuated to be a preset multiple of the previous learning rate after the preset iteration times, wherein the attenuated learning rate is smaller than the learning rate before the attenuation.
8. The method for modeling the charging demand of an electric vehicle according to claim 6, wherein the gradient correction of the deep LSTM network based on the adaptive momentum random optimization method comprises:
and obtaining parameter updating difference values required by iteratively updating the deep LSTM network by adopting a self-adaptive momentum random optimization mode, wherein the calculation formula is as follows:
M t ＝β 1 M t-1 +(1-β 1 )g t
G t ＝β 2 G t-1 +(1-β 2 )g t ⊙g t
wherein beta is 1 Is the first moment attenuation coefficient, beta 2 Is a second moment attenuation coefficient, alpha is a step factor, M t Weighted average of the current index for first moment, G t Is the current exponentially weighted average of the second moment, M t-1 Is M t Corresponding last exponentially weighted average, G t-1 Is G t The corresponding last exponentially weighted average, g t For the current gradient, t is the current iteration number, θ is the parameter to be updated, and Δθ t Updating the difference value for the parameter corresponding to the current required parameter, wherein, as indicated by the ";
and carrying out gradient correction on the parameters corresponding to the deep LSTM network based on the parameter updating difference value.
9. The method for modeling electric vehicle charging demand according to any one of claims 5 to 8, wherein the performing network optimization adjustment on the deep LSTM network after the association training includes:
respectively inputting the subjective charging behavior data into the optimized deep LSTM network and at least one preset comparison algorithm model, and calculating a first root mean square error corresponding to the optimized deep LSTM network and a second root mean square error corresponding to at least one preset comparison algorithm model;
if a second root mean square error is smaller than the first root mean square error, determining the subjective charging behavior data as subjective charging behavior data to be optimized, replacing the optimized deep LSTM network by a preset comparison algorithm model corresponding to the second root mean square error smaller than the first root mean square error, and performing data fitting on the subjective charging behavior data to be optimized;
If at least two second root mean square errors are smaller than the first root mean square errors, determining the subjective charging behavior data as subjective charging behavior data to be optimized, replacing the optimized deep LSTM network by a preset comparison algorithm model corresponding to the minimum second root mean square errors, and performing data fitting on the subjective charging behavior data to be optimized;
if the second root mean square error is greater than or equal to the first root mean square error, continuing to perform data fitting on the subjective charging behavior data by adopting the optimized deep LSTM network;
the root mean square error is calculated as follows:
wherein RMSE is root mean square error, m t A number of test set samples representing a model or network,representing the output value of the model or network, y i Representing the actual value of the test set samples.
10. The method for modeling charging demand of an electric vehicle according to claim 1, wherein the predicted subjective charging behavior data includes a predicted initial charging time, a predicted charging duration, a predicted required electric quantity, and a predicted controlled preference, and the calculating the charging load curve of the user to be analyzed according to the predicted subjective charging behavior data includes:
And calculating by adopting the predicted required electric quantity and the predicted charging time length to obtain an average charging load, wherein the calculation formula is as follows:
calculating the predicted controlled preference and the average charging load to obtain a controlled charging load, and drawing a corresponding charging load curve, wherein the calculation formula is as follows:
P c ′＝(1-R p )×P c
wherein P is c For average charging load, ΔE is predicted required power, Δt is predicted charging duration, P c ' is a controlled post-charge load, R p To predict controlled preferences;
the charge load curve is expressed as: at [ T ] b ，T b +△t]In the time period, the charging load corresponding to the user to be analyzed is P c ' in [ T ] b ,T b +△t]The charging load corresponding to the user to be analyzed is zero and T is the time period except the time period b To predict the initial charge time.
11. The method for modeling a charging demand of an electric vehicle according to claim 1, wherein the user to be analyzed is one of the users in the area to be analyzed of the electric vehicle, the method further comprising:
and superposing charging load curves of all users to be analyzed in the area to be analyzed to obtain a total charging load curve, and carrying out overall charging load prediction on the electric automobile in the area to be analyzed based on the total charging load curve.
12. An electric vehicle charging demand modeling apparatus, comprising:
the clustering calculation module is used for acquiring a plurality of groups of subjective charging behavior data of a user, and clustering the subjective charging behavior data to obtain a corresponding charging behavior clustering result;
the correlation optimization training module is used for constructing a deep LSTM network, acquiring a plurality of groups of objective factor data of the user, and carrying out correlation optimization training on the deep LSTM network by adopting the plurality of groups of objective factor data, the charging behavior clustering result and the plurality of groups of subjective charging behavior data to acquire a target deep LSTM network;
the predicted subjective charging behavior data output module is used for acquiring objective factor data to be detected of a user to be analyzed, inputting the objective factor data to be detected into the target deep LSTM network and outputting predicted subjective charging behavior data;
and the charging load prediction module is used for calculating a charging load curve of the user to be analyzed according to the predicted subjective charging behavior data and performing charging load prediction based on the charging load curve.
13. 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 electric vehicle charging demand modeling method according to any one of claims 1 to 11 according to instructions in the program code.
14. A computer readable storage medium for storing program code for performing the electric vehicle charging demand modeling method of any one of claims 1-11.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310688564.8A CN116703464A (en) | 2023-06-09 | 2023-06-09 | Electric automobile charging demand modeling method and device, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310688564.8A CN116703464A (en) | 2023-06-09 | 2023-06-09 | Electric automobile charging demand modeling method and device, electronic equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116703464A true CN116703464A (en) | 2023-09-05 |
Family
ID=87828815
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310688564.8A Pending CN116703464A (en) | 2023-06-09 | 2023-06-09 | Electric automobile charging demand modeling method and device, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116703464A (en) |
Cited By (2)
| 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 |
| CN117410988A (en) * | 2023-12-11 | 2024-01-16 | 广东领卓能源科技有限公司 | Charging control method and device for new energy charging station |
-
2023
- 2023-06-09 CN CN202310688564.8A patent/CN116703464A/en active Pending
Cited By (4)
| 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 |
| CN117410988A (en) * | 2023-12-11 | 2024-01-16 | 广东领卓能源科技有限公司 | Charging control method and device for new energy charging station |
| CN117410988B (en) * | 2023-12-11 | 2024-03-29 | 广东领卓能源科技有限公司 | Charging control method and device for new energy charging station |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10713597B2 (en) | Systems and methods for preparing data for use by machine learning algorithms | |
| CN116703464A (en) | Electric automobile charging demand modeling method and device, electronic equipment and storage medium | |
| CN111563706A (en) | Multivariable logistics freight volume prediction method based on LSTM network | |
| CN111476435B (en) | Charging pile load prediction method based on density peak value | |
| CN112581263A (en) | Credit evaluation method for optimizing generalized regression neural network based on wolf algorithm | |
| CN110298574B (en) | Convolutional neural network-based electricity user payment risk rating method | |
| CN108681742B (en) | Analysis method for analyzing sensitivity of driver driving behavior to vehicle energy consumption | |
| CN112381320A (en) | Community personnel flow prediction method, system, storage medium and equipment | |
| CN110264270A (en) | A kind of behavior prediction method, apparatus, equipment and storage medium | |
| CN112257935B (en) | Aviation safety prediction method based on LSTM-RBF neural network model | |
| CN112418476A (en) | Ultra-short-term power load prediction method | |
| CN115510963A (en) | Incremental equipment fault diagnosis method | |
| CN113406503A (en) | Lithium battery SOH online estimation method based on deep neural network | |
| CN115375205A (en) | Method, device and equipment for determining water user portrait | |
| CN113239638A (en) | Overdue risk prediction method for optimizing multi-core support vector machine based on dragonfly algorithm | |
| CN115310650A (en) | Low-complexity high-precision time sequence multi-step prediction method and system | |
| Urgun et al. | Composite power system reliability evaluation using importance sampling and convolutional neural networks | |
| CN111260161A (en) | Method and device for issuing crowdsourcing tasks | |
| CN113033898A (en) | Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network | |
| CN112836750A (en) | System resource allocation method, device and equipment | |
| CN117113086A (en) | Energy storage unit load prediction method, system, electronic equipment and medium | |
| CN116542701A (en) | Carbon price prediction method and system based on CNN-LSTM combination model | |
| CN116720620A (en) | Grain storage ventilation temperature prediction method based on IPSO algorithm optimization CNN-BiGRU-Attention network model | |
| CN116431346A (en) | Compensation method for main memory capacity of electronic equipment | |
| CN116231665A (en) | Power load regulation and control method and device, nonvolatile storage medium and electronic equipment |
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 |