CN114037480B - New energy vehicle charging pile demand prediction and deployment optimization method for new city - Google Patents
New energy vehicle charging pile demand prediction and deployment optimization method for new city Download PDFInfo
- Publication number
- CN114037480B CN114037480B CN202111390832.5A CN202111390832A CN114037480B CN 114037480 B CN114037480 B CN 114037480B CN 202111390832 A CN202111390832 A CN 202111390832A CN 114037480 B CN114037480 B CN 114037480B
- Authority
- CN
- China
- Prior art keywords
- city
- charging
- module
- demand prediction
- new
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000005457 optimization Methods 0.000 title claims abstract description 25
- 230000008569 process Effects 0.000 claims abstract description 7
- 238000004364 calculation method Methods 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 230000003044 adaptive effect Effects 0.000 claims description 5
- 238000010606 normalization Methods 0.000 claims description 4
- 238000011176 pooling Methods 0.000 claims description 4
- 238000012549 training Methods 0.000 claims description 4
- 230000035764 nutrition Effects 0.000 claims description 2
- 235000016709 nutrition Nutrition 0.000 claims description 2
- 230000005012 migration Effects 0.000 abstract description 4
- 238000013508 migration Methods 0.000 abstract description 4
- 238000012804 iterative process Methods 0.000 abstract description 3
- 230000001360 synchronised effect Effects 0.000 abstract description 3
- 238000004880 explosion Methods 0.000 abstract description 2
- 230000003993 interaction Effects 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 9
- 230000008901 benefit Effects 0.000 description 3
- 230000006399 behavior Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000009193 crawling Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000013526 transfer learning Methods 0.000 description 1
Abstract
The invention discloses a new energy vehicle charging pile demand prediction and deployment optimization method for a new city, which uses a city migration learning paradigm, utilizes city migration knowledge rich in historical data to help a target city predict user demands of candidate deployment schemes, and simultaneously utilizes heuristic ideas to provide a new model of prediction-planning iteration interaction and a corresponding efficient solving method, so as to quickly search an optimal solution in an explosion scheme combination and ensure the trending and convergence of the method. According to the method, prediction and planning are performed simultaneously, the optimization trend and convergence of the deployment scheme adjustment process are guaranteed, the solving efficiency is improved through a heuristic method, the time required by optimization is greatly reduced, meanwhile, the optimal solution in the iterative process is searched by using the thought of dynamic planning, and the synchronous demand prediction and the optimization of the charging pile deployment scheme are truly realized.
Description
Technical Field
The invention relates to the technical fields of city calculation, city planning and new energy vehicles, in particular to a new city-oriented new energy vehicle charging pile demand prediction and deployment optimization method.
Background
Charging piles are deployed in the New City, and charging demand prediction is needed first, and currently existing methods are mainly divided into two types: prediction from implicit data and prediction from explicit data.
1. Prediction from implicit data. The main idea of the method is as follows: the association between the parking requirements, population distribution, traffic flow, fueling records and the like and the charging requirements is modeled by using implicit data associated with the charging requirements, so that the charging requirements of a new city are calculated. The implicit data itself has different natural properties from the charging demand data, which brings substantial space-time mobility differences, and the predicted charging demand is very prone to error.
2. Prediction from explicit data. The main idea of the method is as follows: and calculating the charging demand of the whole city through the acquired partial charging demand data. For example, a charging electronic track is extracted from the running tracks of a plurality of charging automobiles, and charging demand distribution is obtained; or crawling transaction data from the charging transaction platform, and combining point of interest information (POI) corresponding to the charging station as a charging requirement. However, in a completely new city, it is almost impossible to directly obtain these explicit data.
However, when the charging vehicle is facing a new city, the charging station needs to be deployed in the new city, and the charging station is required to be deployed in the city, so that the charging station is required to be deployed according to the charging requirement of the city, but the charging requirement of the city cannot be predicted due to the lack of historical data of the charging station in the new city, and the optimal deployment scheme cannot be selected from the candidate charging station set.
The definition of the deployment optimization of the charging pile is as follows: for a group of candidate schemes, the maximum benefit which can be obtained in each scheme is automatically calculated, and then the optimal solution is found from the whole candidate scheme set. From the optimization target, the prior work mainly aims at maximizing the social benefits of the charging station, for example, meeting the charging requirements of vehicle owners as much as possible and reducing the queuing waiting time.
The invention aims to solve the problem of how to predict the charging demand and optimize the deployment scheme without charging data when the charging station is deployed in the newcastle city.
Disclosure of Invention
Aiming at the problems, the invention provides a new energy vehicle charging pile demand prediction and deployment optimization method for a new city.
In order to achieve the above object, the present invention provides the following technical solutions:
a new energy vehicle charging pile demand prediction and deployment optimization method for a new city comprises the following steps:
S1, initializing distribution of charging piles in a target city
S2, distributing charging piles according to source citiesTarget city charging pile distribution/>Training a charging demand prediction model by using the source city data D SC and the target city data D TC, and predicting the charging demand distribution/> -of the target city
S3, calculating daily average nutrition R TC of current charging pile distribution by using a prediction result;
S4, giving a revenue increase threshold value theta, if R TC is less than or equal to theta, ending, and returning to the distribution of the charging piles of the current target city Otherwise, executing the step S5;
S5, constructing a candidate set according to the current distribution of the charging piles
S6, pairingTraversing, and predicting to obtain a new charging demand by using a charging demand prediction model;
s7, searching by using dynamic planning thought Optimal solution to/>Performing primary updating; the process goes to step S2.
Further, the charging demand prediction model comprises a context module, an portrait feature module, a demand prediction module and a field self-adaption module: the context feature of each charging station in the source city and the target city is input into the context module, and the context feature f c is output; the portrait features of each charging station of the source city and the target city are input into a portrait feature module, and portrait features f p are output; the context feature f c and the portrait feature f p are combined to obtain a site joint feature f, the site joint feature f and the time intervals are input into a demand prediction module, and the charging demand in each time interval is outputSite joint characteristic f input field self-adaptive module and output field label/>Indicating the domain to which the site feature belongs.
Further, the context module is sequentially composed of a convolution block, a space-time attention module, convolution blocks and a global pooling layer, wherein each convolution block comprises a convolution layer, a normalization layer and a ReLU activation function.
Further, the portrait feature module is composed of two fully connected layers with a ReLU activation function.
Further, the demand prediction module is composed of an embedded layer, two full-connection layers and a full-connection layer, the time interval is converted into an array q through the embedded layer, the site characteristics are connected with the output result q through the two full-connection layers, and the output result is input into the full-connection layer to obtain a prediction result.
Further, the demand prediction module loses the function:
Ldemand＝(1-α)Lreg+αLrank
Where L reg represents the mean square error and L rank represents the ranking error.
Further, the calculation formula of L reg is:
S SC denotes the charging demand of the source city, And y represents the predicted value and the true value of the target city charging demand, respectively.
Further, the calculation formula of L rank is:
Where P ij represents the probability that y i is ranked higher than y j in the target city truth, Representing the probability that y i is ranked higher than y j in the predicted result, the calculation formula of P ij is:
oij＝yi-yj(yi＞yj)。
further, the domain adaptive module is composed of two fully connected layers.
Further, the loss function of the domain adaptive module is:
Wherein: d is a domain label and is used to indicate that, Is a predicted domain tag, s=s sc∪STC,/>And/>Is a weight parameter,/>And/>Is a deviation parameter.
Compared with the prior art, the invention has the beneficial effects that:
The new energy vehicle charging pile demand prediction and deployment optimization method for the new city uses a city migration learning paradigm, utilizes city migration knowledge rich in historical data to help a target city predict the user demand of a candidate deployment scheme, and simultaneously utilizes a heuristic idea to provide a new model of prediction-planning iteration interaction and a corresponding efficient solving method, so that an optimal solution is quickly searched in an explosion scheme combination, and the trending and convergence of the method are guaranteed. According to the method, prediction and planning are performed simultaneously, the optimization trend and convergence of the deployment scheme adjustment process are guaranteed, the solving efficiency is improved through a heuristic method, the time required by optimization is greatly reduced, meanwhile, the optimal solution in the iterative process is searched by using the thought of dynamic planning, and the synchronous demand prediction and the optimization of the charging pile deployment scheme are truly realized.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for the embodiments will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments described in the present application, and other drawings may be obtained according to these drawings for a person having ordinary skill in the art.
Fig. 1 is a flow chart of a new energy vehicle charging pile demand prediction and deployment optimization method for a new city.
Detailed Description
For a better understanding of the present technical solution, the method of the present invention is described in detail below with reference to the accompanying drawings.
1. Problem scenario
Now that a promotion of charging cars in a new city is required, charging stations first need to be deployed in that city. The charging station deployment is expected to bring greater benefit naturally, so that the charging station deployment is required to be deployed according to the charging requirement of the city, but the charging requirement of the city cannot be predicted due to the lack of historical data of the charging station in the newcastle city, and then the optimal deployment scheme cannot be selected from the candidate charging station set.
2. Problem definition
We define: charging stationWhere l i is the geographic location of c i,/>And/>The number of the slow filling piles and the fast filling piles in c i are respectively/>And/>The cost of deploying a slow filling pile and a fast filling pile at c i is/>, respectivelyAnd/>And C i represents the service prices of the slow fill pile and the fast fill pile in the time interval T, respectively. /(I)And/>The utilization of the slow and fast fill piles within the time interval T is denoted c i, respectively, which is used in the present case to represent the charging demand of the charging station.
We consider two cities: the source city has deployed charging stations, the collection of which is designated C sC. All attributes of each deployed charging station c i∈CSC in the source city are known. The target city, where charging stations have not been deployed, one candidate deployment scenario is denoted as C TC. Only candidate charging stations c i∈CTC in the target cityThese 5 properties are known.
We use collectionsTo represent the number of slow fills and fast fills for each site in the target city deployment scenario C TC. /(I)And the number of the slow charging piles and the fast charging piles of each station in the C SC is represented. D SC and D TC are used to represent multi-source context data, i.e., POI data, road network data, traffic data, in the source city and the target city, respectively.
We propose a charge demand prediction model f, given thatD SC and D TC, and charging requirements of source citiesUnder the condition of (1) minimizing the predicted value/>, of the charging demandThe error from the true value Y TC is:
deployment scheme for solving target city on basis of deployment scheme At a given CTC,/>Under the conditions of Y SC、DSC、DTC, the charge demand prediction method f and the preset upper limit B, the revenue R is maximized, and the deployment cost does not exceed the preset upper limit B, namely:
3. Charging demand prediction model
The architecture of the charge demand prediction model is shown in fig. 1, and the architecture is composed of four components, namely: the system comprises a context module, a portrait characteristic module, a demand prediction module and a field self-adaption module.
1) Context module
Module input: the full context features of each charging station in the source city and the target city, including POI, road network features, traffic features of the charging station and its neighbor charging stations.
And (3) module output: the context feature f c.
The module structure is as follows: convolution block + spatiotemporal attention module + convolution block + global pooling layer. Each convolution block contains one convolution layer, one normalization layer and one ReLU activation function. The input of each convolution block isThen output:
Where W c and b c are both learnable parameters, x represents convolution operations and BN represents batch normalization. To avoid over-fitting, the discard operation will be enabled after the first convolution block. The spatial attention module is used to encode the spatial correlation. The second convolution module is used to improve performance. The global pooling layer is used to reduce feature dimensions.
2) Portrait feature module
Module input: the portrait characteristics of each charging station of the source city and the target city, namely charging pile information of the charging stations of the source city and the target city and the neighboring charging stations.
And (3) module output: image feature f p.
The module structure is as follows: two fully connected layers with one ReLU activation function.
3) Demand prediction module
Module input: site association feature f (combined from f c and f p) and time interval.
And (3) module output: including charging requirements in each time interval
The module structure is as follows: the time interval is converted into an array q by an embedding layer. The site characteristics connect the output result with q through two full connection layers, and then input the output result into one full connection layer to obtain a prediction result.
Loss function: l demand＝(1-α)Lreg+αLrank
Where L reg represents the mean square error, S SC represents the charging demand of the source city,And y represents the predicted value and the true value of the target city charging demand, respectively.
L rank represents a ranking error. Where P ij represents the probability that y i is ranked higher than y j in the target city truth,Representing a higher probability of y i ranking than y j in the predicted result.
oij＝yi-yj(yi＞yj)
4) Domain adaptive module
Module input: site association feature f (derived from the combination of f c and f p)
And (3) module output: domain labelFor indicating the domain to which the site feature belongs
The module structure is as follows: two fully connected layers.
Loss function:
Wherein: d is a domain label and is used to indicate that, Is a predicted domain tag, s=s sc∪STC,/>And/>Is a weight parameter,/>And/>Is a deviation parameter.
For the whole charge demand prediction module, our goal is to minimize L reg and L rank, maximize L domain. The final joint loss function is:
L＝(1-α)Lreg+αLrank-βLdomain
4. Charging pile deployment scheme optimization module
Module target: in a given set of candidate charging station locations, the number of fast and slow piles per charging station is adjusted to maximize daily gain under conditions that meet the final total cost not exceeding the upper limit B.
Module input: revenue growth threshold θ, total budget B, source city data D SC, source city charging pile distributionTarget city data D TC
And (3) module output: charging pile distribution of target city
Constructing a candidate set: if the number of fast and slow charging piles of a certain charging station is (N SF,nS s), defining a candidate set N i as:
charging pile distribution for target city For/>
The new energy vehicle charging pile demand prediction and deployment optimization method for the new city comprises the following steps:
(1) First, initializing a charging pile distribution in a target city
(2) According toD SC、DTC training a charging demand prediction model f and predicting the charging demand distribution/>, of the target city
(3) Calculating daily average camp R TC of the current charging pile distribution by using the prediction result;
(4) If R TC is less than or equal to theta, ending and returning to the current target city charging pile distribution Otherwise, executing the step (5);
(5) Construction of candidate sets according to current charging pile distribution
(6) For a pair ofTraversing, and predicting to obtain a new charging requirement by using f;
(7) Searching by dynamic programming thought Optimal solution to/>An update is made. Jump to step (2).
The method of the application realizes the prediction of the charging demand of the target city by using the historical data of the public charging station of the source city. The implicit data set is used for calculating the charging requirement of the target city, and the predicted charging requirement is very easy to make mistakes due to the fact that natural attribute differences among data bring substantial space-time mobility differences. Explicit data is difficult to collect for new cities. According to the method, charging data of a source city is used, image features and situation features which can be distinguished are learned by combining with multi-source data such as POIs of the source city and the target city, and then charging requirements of the target city are predicted. Meanwhile, a field self-adaptive network is introduced to guide the network to learn high-level semantic features irrelevant to cities. In the aspect of model design, the application reserves the space-time characteristics of charging behaviors, considers the influence of the contextual characteristics such as road network information, POI information, traffic information and the like on the charging behaviors, and introduces the mutual influence among charging stations. Meanwhile, the domain adaptive network is introduced to solve the domain offset problem in city transfer learning, and the distribution offset of the input space caused by the change on an observation system is reduced. The method and the device realize the prediction of the charging demand under the condition of no target city charging data.
The method of the invention simultaneously predicts and plans, ensures the optimality and convergence of the deployment scheme adjustment process, and improves the solving efficiency by a heuristic method. In an iterative process of optimizing deployment schemes, each site has 5 candidate schemes, which can generate totalIt is unacceptable to train and predict each deployment scenario separately for a new deployment scenario. According to the application, only 1 training and 5|C TC prediction are carried out in one iteration process, so that the time required by optimization is greatly reduced, and meanwhile, the optimal solution in the iteration process is searched by using the thought of dynamic planning, so that the synchronous demand prediction and the optimization of the charging pile deployment scheme are truly realized.
The above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; 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 solutions described in the foregoing embodiments may be modified or some technical features may be replaced with others, which may not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims (5)
1. A new energy vehicle charging pile demand prediction and deployment optimization method for a new city is characterized by comprising the following steps:
S1, initializing distribution of charging piles in a target city
S2, distributing charging piles according to source citiesTarget city charging pile distribution/>Training a charging demand prediction model by using the source city data D SC and the target city data D TC, and predicting the charging demand distribution/> -of the target cityThe charging demand prediction model comprises a context module, an portrait feature module, a demand prediction module and a field self-adaption module: the context feature of each charging station in the source city and the target city is input into the context module, and the context feature f c is output; the portrait features of each charging station of the source city and the target city are input into a portrait feature module, and portrait features f p are output; the context feature fc and the portrait feature f p are combined to obtain a site joint feature f, the site joint feature f and the time intervals are input into a demand prediction module, and the charging demand/>, contained in each time interval, is outputSite joint characteristic f input field self-adaptive module and output field label/>Indicating the domain to which the site feature belongs; the context module sequentially comprises a convolution block, a space-time attention module, convolution blocks and a global pooling layer, wherein each convolution block comprises a convolution layer, a normalization layer and a ReLU activation function; the portrait feature module consists of two full-connection layers with a ReLU activation function; the demand prediction module consists of an embedded layer, two full-connection layers and a full-connection layer, wherein a time interval is converted into an array q through the embedded layer, site characteristics are connected with the q through the two full-connection layers, and then the output result is input into the full-connection layer to obtain a prediction result; the field self-adaptive module consists of two full-connection layers;
s3, calculating daily average nutrition R TC of current charging pile distribution by using a prediction result;
S4, giving a revenue increase threshold value theta, if R TC is less than or equal to theta, ending, and returning to the distribution of the charging piles of the current target city Otherwise, executing the step S5;
S5, constructing a candidate set according to the current distribution of the charging piles
S6, pairingTraversing, and predicting to obtain a new charging demand by using a charging demand prediction model;
s7, searching by using dynamic planning thought Optimal solution to/>Performing primary updating; the process goes to step S2.
2. The new energy vehicle charging pile demand prediction and deployment optimization method for the new city according to claim 1, wherein the demand prediction module loses a function:
Ldemand＝(1-α)Lreg+αLrank
Where L reg represents the mean square error and L rank represents the ranking error.
3. The new energy vehicle charging pile demand prediction and deployment optimization method for the new city according to claim 2, wherein the calculation formula of L reg is as follows:
S SC denotes the charging demand of the source city, And y represents the predicted value and the true value of the target city charging demand, respectively.
4. The new energy vehicle charging pile demand prediction and deployment optimization method for the new city according to claim 2, wherein the calculation formula of L rank is as follows:
Where P ij represents the probability that y i is ranked higher than y j in the target city truth, Representing the probability that y i is ranked higher than y j in the predicted result, the calculation formula of P ij is:
oij＝yi-yj(yi>yj)。
5. the new energy vehicle charging pile demand prediction and deployment optimization method for the new city according to claim 1, wherein the loss function of the field adaptive module is as follows:
Wherein: d is a domain label and is used to indicate that, Is a predicted domain tag, s=s sc∪STC,/>And/>Is a weight parameter,/>AndIs a deviation parameter.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111390832.5A CN114037480B (en) | 2021-11-23 | New energy vehicle charging pile demand prediction and deployment optimization method for new city |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111390832.5A CN114037480B (en) | 2021-11-23 | New energy vehicle charging pile demand prediction and deployment optimization method for new city |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114037480A CN114037480A (en) | 2022-02-11 |
| CN114037480B true CN114037480B (en) | 2024-06-25 |
Family
ID=
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103793758A (en) * | 2014-01-23 | 2014-05-14 | 华北电力大学 | Multi-objective optimization scheduling method for electric vehicle charging station including photovoltaic power generation system |
| CN106777835A (en) * | 2017-02-15 | 2017-05-31 | 国网江苏省电力公司苏州供电公司 | The city electric car charging network planing method of multiple optimization aims is considered simultaneously |
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103793758A (en) * | 2014-01-23 | 2014-05-14 | 华北电力大学 | Multi-objective optimization scheduling method for electric vehicle charging station including photovoltaic power generation system |
| CN106777835A (en) * | 2017-02-15 | 2017-05-31 | 国网江苏省电力公司苏州供电公司 | The city electric car charging network planing method of multiple optimization aims is considered simultaneously |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112016812B (en) | Multi-unmanned aerial vehicle task scheduling method, system and storage medium | |
| Zheng et al. | Traffic equilibrium and charging facility locations for electric vehicles | |
| CN108199100B (en) | Electric automobile long-distance operation charging planning method in intelligent traffic | |
| CN102810118B (en) | A kind of change weighs net K nearest neighbor search method | |
| CN103295074B (en) | The electric vehicle (EV) with the charging station of best located charges the method and system of infrastructure | |
| US20210233196A1 (en) | System and method for ride order dispatching | |
| Liu et al. | Control method of urban intelligent parking guidance system based on Internet of Things | |
| Kim et al. | Idle vehicle relocation strategy through deep learning for shared autonomous electric vehicle system optimization | |
| CN109087509B (en) | Road network traffic running state prediction method | |
| CN113485429B (en) | Route optimization method and device for air-ground cooperative traffic inspection | |
| CN114037175B (en) | Large-scale public transportation network hierarchical optimization method based on multi-scale clustering | |
| Tang et al. | Online operations of automated electric taxi fleets: An advisor-student reinforcement learning framework | |
| CN109063759A (en) | A kind of neural network structure searching method applied to the more attribute forecasts of picture | |
| CN108900980A (en) | A kind of resource allocation optimization method based on mobility prediction in heterogeneous network | |
| CN115545758A (en) | Method and system for self-adaptive incremental site selection of urban service facilities | |
| CN114092176A (en) | Urban commuting regular bus planning method based on bus | |
| Vasant et al. | Optimal power allocation scheme for plug-in hybrid electric vehicles using swarm intelligence techniques | |
| CN112949931A (en) | Method and device for predicting charging station data with hybrid data drive and model | |
| CN112381398A (en) | Electric vehicle charging station site selection method and system considering limited trip of user | |
| Chen et al. | Finding next high-quality passenger based on spatio-temporal big data | |
| CN114037480B (en) | New energy vehicle charging pile demand prediction and deployment optimization method for new city | |
| Wang et al. | ForETaxi: data-driven fleet-oriented charging resource allocation in large-scale electric taxi networks | |
| Chen et al. | Dynamic path optimization in sharing mode to relieve urban traffic congestion | |
| US20230153742A1 (en) | Methods for shared bicycle delivery and operation area planning in smart cities and internet of things (iot) systems thereof | |
| CN116822194A (en) | Electric vehicle charging station address selection method based on ant colony optimization algorithm |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant |