CN110489671A - A kind of road charging pile recommended method based on encoder-decoder model - Google Patents

A kind of road charging pile recommended method based on encoder-decoder model Download PDF

Info

Publication number
CN110489671A
CN110489671A CN201910767503.4A CN201910767503A CN110489671A CN 110489671 A CN110489671 A CN 110489671A CN 201910767503 A CN201910767503 A CN 201910767503A CN 110489671 A CN110489671 A CN 110489671A
Authority
CN
China
Prior art keywords
charging pile
encoder
input
probability
decoder
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.)
Granted
Application number
CN201910767503.4A
Other languages
Chinese (zh)
Other versions
CN110489671B (en
Inventor
刘林峰
贾鉴
吴家皋
金仙力
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing Post and Telecommunication University
Nanjing University of Posts and Telecommunications
Original Assignee
Nanjing Post and Telecommunication University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nanjing Post and Telecommunication University filed Critical Nanjing Post and Telecommunication University
Priority to CN201910767503.4A priority Critical patent/CN110489671B/en
Publication of CN110489671A publication Critical patent/CN110489671A/en
Application granted granted Critical
Publication of CN110489671B publication Critical patent/CN110489671B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9537—Spatial or temporal dependent retrieval, e.g. spatiotemporal queries
    • 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
    • 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
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G06Q10/047—Optimisation of routes or paths, e.g. travelling salesman problem
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Electricity, gas or water supply

Abstract

The present invention relates to a kind of road charging pile recommended methods using encoder-decoder model, belong to depth learning technology field, using encoder-decoder model, the model uses the Move Mode analyzed in electric car historical track based on the encoder module of shot and long term memory network, and Future Trajectory sequence is generated using the decoder module of LSTM, probability of the output by softmax layers of generation Future Trajectory sequence, before then being exported using Beam Search algorithmKThe Future Trajectory sequence conduct of maximum probabilityKPath candidate, finally according to the most short principle pair of additional moving distanceKPath candidate determinesKA charging pile, what is obtainedKA charging pile and its priority recommend electric car user.

Description

A kind of road charging pile recommended method based on encoder-decoder model
Technical field
The present invention relates to a kind of road charging pile, specifically a kind of road charging pile recommended method belongs to depth Practise technical field.
Background technique
In traditional neural network model, input layer between hidden layer node and hidden layer to output layer it Between node all connect entirely, the node between every layer be it is connectionless, this performance in processing sequence problem is bad. Recognition with Recurrent Neural Network (Recurrent Neural Networks, RNNs) remembers the information of front, and is applied to current In the calculating of output, i.e., the node between hidden layer has connection, and the input of hidden layer is not only defeated including input layer It out, further include the output of last moment hidden layer.
RNNs can there are problems that relying on (Long-Term Dependencies) for a long time, i.e., most start to be input to the interior of RNNs Holding the influence exported to current time can be smaller and smaller, proposes LSTM (Long Short Term Memory) mechanism thus, LSTM is a kind of improvement to RNNs unit, it is increased multiple " doors " by the unit of hidden layer in reinforcement original RNNs, real Now to the memory of content, forgetting, so as to improve the long-term Dependence Problem of RNNs.
In encoder-decoder model, encoder module is responsible for successively reading in each unit of list entries, will It is encoded into a context vector, and decoder module is responsible for predicting output sequence in the case where given context vector. In trajectory predictions processing, encoder and decoder module has selected LSTM to realize.
Summary of the invention
The road charging pile recommended method based on encoder-decoder model that the object of the present invention is to provide a kind of, Encoder module obtains the status information about electric car, and decoder module is used to generate Future Trajectory, then utilizes softmax Layer obtains the probability of every prediction Future Trajectory, recycles Beam Search algorithm to generate K optimal candidate track sets, most K charging pile is determined according to the smallest principle of total path afterwards, K charging pile and its corresponding probability are recommended into user, so that Charging pile is recommended more accurate.
The object of the present invention is achieved like this: a kind of road charging pile recommendation based on encoder-decoder model Method, comprising the following steps:
Step 1) is at full charge to each electric taxi, and a low battery threshold value δ is arranged；
For step 2) when the electricity of some electric taxi is lower than threshold value δ, which issues the pre- of not enough power supply Alert signal starts to recommend road charging pile for the electric taxi；
80% historical track of electric taxi is input in encoder-decoder model and instructs by step 3) Practice；
The historical track of the residue 20% of electric taxi is input to encoder-decoder model and surveyed by step 4) Examination；
Step 5) is using the output result of encoder-decoder model as softmax layers of input, the softmax Probability of the layer to predict every Future Trajectory subsequence, the probability that output Future Trajectory sequence occurs, the future of every prediction The new probability formula of track subsequence are as follows:
Wherein, i represents the number of this track subsequence appearance, and j represents the number that all track subsequences occur；
Step 6) is using softmax layers of output result as the input of Beam Search algorithm, by BeamSearch algorithm The path candidate of K maximum probability before exporting；The condition probability formula of the Future Trajectory point sequence of every prediction are as follows:
Step 7) selects charging pile: to each path candidate, selecting the charging pile for meeting the following conditions: 1) electronic The dump energy of taxi is enough to maintain to be moved to the charging pile from current location；2) charging pile it is all it is eligible 1) Additional moving distance is minimum in charging pile, that is, the driving path for deviateing the path candidate is most short；The charging pile selected is as this time The corresponding charging pile of routing diameter selects the probability of the charging pile to be equal to the probability of this path candidate；
Step 8) repeats step 7, until K path candidate all selects corresponding charging pile and probability；
K charging pile and corresponding probability are recommended user by step 9).
Compared with prior art, the beneficial effects of the present invention are recommend present invention can ensure that electric taxi reaches Charging pile before will not cast anchor, and the extra path from the current location of electric car to destination is most short.The present invention can For in electric vehicle.
It is further limited as of the invention, training in step 3) method particularly includes: by the longitude and latitude of tracing point The location mode at T moment is input in decoder module, decoder module by sequence as input by encoder module The output in each stage not as next stage input, but directly using the track point sequence after the T moment as input, with This analogizes, and finally obtains a trained encoder-decoder model.The training method, which mainly passes through, to be forgotten to carry out correlation Important information preservation is got off in the update of inessential information, can thus remember key message constantly.
It is further limited as of the invention, place of the encoder module in step 4) test phase to track point sequence Reason process is as the encoder module in the training stage, and in decoder module, the prediction result in each stage will be inputted Into next stage；Finally obtain the Future Trajectory point sequence predicted by decoder module；In each LSTM structure Calculating process includes:
Forget door:
Input gate:
Candidate door:
Memory unit state vector: Ct=ft×Ct-1+it×C′t
Out gate:
Output quantity: ht=Ot×tanh(Ct)
Wherein, σ (x): sigmoid function, ft, it, Ot, C 't: gate function,Weight square Battle array, bf, bi, bc, bo: variable deviation vector, ut: electric car historical track sequence, Ct: memory unit state vector, ht: state output Vector.Each step of the test method exports all as the input information of next step, and the output result of last moment will affect down The prediction result at one moment, obtained prediction result can be more accurate.
Detailed description of the invention
Fig. 1 is the road scene figure that electric taxi selects charging pile.
Fig. 2 is flow chart of the invention.
Fig. 3 is the schematic diagram of encoder-decoder model；
It illustrates:
u1,u2,…uT: the historical track sequence of electric car；
C1,C2,…CT: the memory unit state in encoder module each stage；
h1,h2,…hT-1: the state output in encoder module each stage；
C’0,C’1,C’2,…C’T’-1: the memory unit state in deconder module each stage；
h’1,h’2,…h’T’-1: the state output in decoder module each stage；
S1,S2,…ST’: the prediction Future Trajectory sequence of electric car.
Fig. 4 is the schematic diagram of LSTM structure.
It illustrates:
ht-1: the state output of last moment；
Ct-1: the memory unit state of last moment；
ut: current time input；
ht: the state output at current time；
Ct: the memory unit state at current time.
Specific embodiment
Technical solution of the present invention is described in further detail with reference to the accompanying drawing:
A kind of road charging pile recommended method based on encoder-decoder model as shown in Figure 2, including following step Suddenly.
Step 1, at full charge to each electric taxi, and a low battery threshold value δ is set.
Step 2, when the electricity of some electric taxi is lower than threshold value 6, the state of electric taxi is as shown in figure 1 at this time Shown in automobile 1, which issues the pre-warning signal of not enough power supply, starts to recommend road charging for the electric taxi Stake.
Step 3,80% historical track of electric taxi is input in encoder-decoder model and is instructed Practice, using the longitude of tracing point and latitude sequence as the input (u in Fig. 3t), by encoder module by the cell-like at T moment State (the C in Fig. 3T) be input in decoder module, the output (S in Fig. 3 in decoder module each staget) not as under The input in one stage, but directly using the track point sequence after the T moment as input, and so on, finally obtain an instruction The encoder-decoder model perfected.
Step 4, the historical track of the residue 20% of electric taxi is tested, wherein in test phase Encoder module to the treatment process of track point sequence as the encoder module in the training stage, in decoder module In, the prediction result s in each stagetIt is input in next stage.Finally obtain the track predicted by decoder module Sequence；
Calculating process in each LSTM structure (such as Fig. 4) includes:
Forget door:
Input gate:
Candidate door:
Memory unit state vector: Ct=ft×Ct-1+it×C′t
Out gate:
Output quantity: ht=Ot×tanh(Ct)
Wherein, σ (x): sigmoid function
ft, it, Ot, C 't: gate function
Weight matrix
bf, bi, bc, bo: variable deviation vector
ut: electric car historical track sequence
Ct: memory unit state vector
ht: state output vector.
Step 5, using the output result of encoder-decoder model as softmax layers of input, Future Trajectory is exported The probability that sequence occurs.The new probability formula of the Future Trajectory subsequence of every prediction are as follows:
Wherein i represents the number of this track subsequence appearance, and j represents the number that all track subsequences occur.
Step 6, using softmax layers of output result as the input of Beam Search algorithm, by BeamSearch algorithm The path candidate of K maximum probability before exporting；The condition probability formula of the Future Trajectory point sequence of every prediction are as follows:
Step 7, it selects charging pile: to each path candidate, selecting the charging pile for meeting the following conditions: (1) is electric The dump energy of dynamic taxi is enough to maintain to be moved to the charging pile from current location；(2) charging pile is all eligible (1) additional moving distance is minimum in charging pile, that is, the driving path for deviateing the path candidate is most short.The charging pile conduct selected The corresponding charging pile of this path candidate selects the probability of the charging pile to be equal to the probability of this path candidate.
Step 8, step 7 is repeated, until K path candidate all selects corresponding charging pile and probability.
Step 9, K charging pile and corresponding probability are recommended into user.
The above, the only specific embodiment in the present invention, but scope of protection of the present invention is not limited thereto, appoints What is familiar with the people of the technology within the technical scope disclosed by the invention, it will be appreciated that expects transforms or replaces, and should all cover Within scope of the invention, therefore, the scope of protection of the invention shall be subject to the scope of protection specified in the patent claim.

Claims (3)

1. a kind of road charging pile recommended method based on encoder-decoder model, which is characterized in that including following step It is rapid:
Step 1) is at full charge to each electric taxi, and a low battery threshold value δ is arranged；
For step 2) when the electricity of some electric taxi is lower than threshold value δ, which issues the early warning letter of not enough power supply Number, start to recommend road charging pile for the electric taxi；
80% historical track of electric taxi is input in encoder-decoder model and is trained by step 3)；
The historical track of the residue 20% of electric taxi is input to encoder-decoder model and tested by step 4)；
Using the output result of encoder-decoder model as softmax layers of input, described softmax layers is used step 5) To predict the probability of every Future Trajectory subsequence, the probability that output Future Trajectory sequence occurs, the Future Trajectory of every prediction The new probability formula of subsequence are as follows:
Wherein, i represents the number of this track subsequence appearance, and j represents the number that all track subsequences occur；
Step 6) is defeated by Beam Search algorithm using softmax layers of output result as the input of Beam Search algorithm The path candidate of preceding K maximum probability out；The condition probability formula of the Future Trajectory point sequence of every prediction are as follows:
Step 7) selects charging pile: to each path candidate, selecting charging pile for meeting the following conditions: 1) electronic taxi The dump energy of vehicle is enough to maintain to be moved to the charging pile from current location；2) charging pile is in all eligible chargings 1) Additional moving distance is minimum in stake, that is, the driving path for deviateing the path candidate is most short；The charging pile selected is as this candidate road The corresponding charging pile of diameter selects the probability of the charging pile to be equal to the probability of this path candidate；
Step 8) repeats step 7, until K path candidate all selects corresponding charging pile and probability；
K charging pile and corresponding probability are recommended user by step 9).
2. a kind of road charging pile recommended method based on encoder-decoder model according to claim 1, special Sign is, trains in step 3) method particularly includes: using the longitude of tracing point and latitude sequence as input, by encoder The location mode at T moment is input in decoder module by module, and the output in decoder module each stage is not as next The input in stage, but directly using the track point sequence after the T moment as input, and so on, finally obtain a training Good encoder-decoder model.
3. a kind of road charging pile recommended method based on encoder-decoder model according to claim 2, special Sign is that the encoder module in step 4) test phase is in the treatment process and training stage of track point sequence Encoder module is the same, and in decoder module, the prediction result in each stage will be input in next stage；Finally To the Future Trajectory point sequence predicted by decoder module；Calculating process in each LSTM structure includes:
Forget door:
Input gate:
Candidate door:
Memory unit state vector: Ct=ft×Ct-1+it×C′t
Out gate:
Output quantity: ht=Ot×tanh(Ct)
Wherein, σ (x): sigmoid function, ft, it, Ot, C 't: gate function,Weight matrix, bf, bi, bc, bo: variable deviation vector, ut: electric car historical track sequence, Ct: memory unit state vector, ht: state output vector.
CN201910767503.4A 2019-08-20 2019-08-20 Road charging pile recommendation method based on encoder-decoder model Active CN110489671B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910767503.4A CN110489671B (en) 2019-08-20 2019-08-20 Road charging pile recommendation method based on encoder-decoder model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910767503.4A CN110489671B (en) 2019-08-20 2019-08-20 Road charging pile recommendation method based on encoder-decoder model

Publications (2)

Publication Number Publication Date
CN110489671A true CN110489671A (en) 2019-11-22
CN110489671B CN110489671B (en) 2022-03-11

Family

ID=68552226

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910767503.4A Active CN110489671B (en) 2019-08-20 2019-08-20 Road charging pile recommendation method based on encoder-decoder model

Country Status (1)

Country Link
CN (1) CN110489671B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111313449A (en) * 2020-03-02 2020-06-19 华北电力大学 Cluster electric vehicle power optimization management method based on machine learning

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104864881A (en) * 2014-02-25 2015-08-26 福特全球技术公司 Method and apparatus for providing a navigation route with recommended charging
CN104933466A (en) * 2015-07-13 2015-09-23 武汉大学 Charging scheduling system based on electric vehicle charging priority and method thereof
GB2542848A (en) * 2015-10-02 2017-04-05 Zapinamo Ltd Scheduling the re-charging of electric vehicles
CN107133841A (en) * 2017-04-20 2017-09-05 深圳充电网科技有限公司 A kind of charging pile website recommendation method and apparatus
CN108106626A (en) * 2017-12-18 2018-06-01 浙江工业大学 A kind of electric vehicle trip route planing method based on driving cycle
CN109635284A (en) * 2018-11-26 2019-04-16 北京邮电大学 Text snippet method and system based on deep learning associate cumulation attention mechanism
CN109670277A (en) * 2019-02-19 2019-04-23 南京邮电大学 A kind of travel time prediction method based on multi-modal data fusion and multiple-model integration

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104864881A (en) * 2014-02-25 2015-08-26 福特全球技术公司 Method and apparatus for providing a navigation route with recommended charging
CN104933466A (en) * 2015-07-13 2015-09-23 武汉大学 Charging scheduling system based on electric vehicle charging priority and method thereof
GB2542848A (en) * 2015-10-02 2017-04-05 Zapinamo Ltd Scheduling the re-charging of electric vehicles
CN107133841A (en) * 2017-04-20 2017-09-05 深圳充电网科技有限公司 A kind of charging pile website recommendation method and apparatus
CN108106626A (en) * 2017-12-18 2018-06-01 浙江工业大学 A kind of electric vehicle trip route planing method based on driving cycle
CN109635284A (en) * 2018-11-26 2019-04-16 北京邮电大学 Text snippet method and system based on deep learning associate cumulation attention mechanism
CN109670277A (en) * 2019-02-19 2019-04-23 南京邮电大学 A kind of travel time prediction method based on multi-modal data fusion and multiple-model integration

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
乔少杰: "GMTP:基于高斯混合模型的轨迹预测算法", 《软件学报》 *
周宁宁 等: "基于中介真值程度度量处理图像的应用理论研究", 《南京邮电大学学报（自然科学版）》 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111313449A (en) * 2020-03-02 2020-06-19 华北电力大学 Cluster electric vehicle power optimization management method based on machine learning
CN111313449B (en) * 2020-03-02 2022-11-29 华北电力大学 Cluster electric vehicle power optimization management method based on machine learning

Also Published As

Publication number Publication date
CN110489671B (en) 2022-03-11

Similar Documents

Publication Publication Date Title
Lu et al. An asymmetric encoder–decoder model for Zn-ion battery lifetime prediction
Li et al. Remaining useful life prediction for lithium-ion batteries based on a hybrid model combining the long short-term memory and Elman neural networks
CN107728478B (en) Fuel cell oxygen excess coefficient neural network prediction control method
CN106908736A (en) Based on the lithium battery method for predicting residual useful life that depth confidence net and Method Using Relevance Vector Machine are merged
Pan et al. Transfer learning-based hybrid remaining useful life prediction for lithium-ion batteries under different stresses
Zheng et al. Reinforcement learning-based energy management strategies of fuel cell hybrid vehicles with multi-objective control
CN111999649A (en) XGboost algorithm-based lithium battery residual life prediction method
Liu et al. Lithium-ion battery remaining useful life prediction with long short-term memory recurrent neural network
CN111275169A (en) Method for predicting building thermal load in short time
Xu et al. Fast capacity prediction of lithium-ion batteries using aging mechanism-informed bidirectional long short-term memory network
Tian et al. State-of-charge estimation for lithium-ion batteries based on attentional sequence-to-sequence architecture
CN110489671A (en) A kind of road charging pile recommended method based on encoder-decoder model
CN116298936A (en) Intelligent lithium ion battery health state prediction method in incomplete voltage range
CN111141879A (en) Deep learning air quality monitoring method, device and equipment
CN117406100A (en) Lithium ion battery remaining life prediction method and system
CN112183866A (en) Short-term wind speed exceeding probability prediction method and system
CN116541774A (en) Method for monitoring and early warning environmental factors in long tunnel asphalt pavement hole
Nair et al. AI-Driven Digital Twin Model for Reliable Lithium-Ion Battery Discharge Capacity Predictions
CN116184214A (en) Online rapid prediction method and device for service life of power battery for vehicle and storage medium
CN116609671A (en) Battery capacity prediction method, device, equipment and storage medium
CN116359773A (en) Lithium ion battery remaining life prediction method and system
CN114545279A (en) Lithium battery health state estimation method based on neural network ordinary differential equation
CN113537620A (en) Vehicle speed prediction method based on Markov model optimization and working condition recognition
CN113485313A (en) Anti-interference method and device for automatic driving vehicle
CN112132360A (en) Wind speed short-time confidence interval prediction method and system based on LSTM and GMM

Legal Events

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