CN113159832A - Time-space price elasticity estimation method for data-driven electric vehicle charging demand - Google Patents
Time-space price elasticity estimation method for data-driven electric vehicle charging demand Download PDFInfo
- Publication number
- CN113159832A CN113159832A CN202110328125.7A CN202110328125A CN113159832A CN 113159832 A CN113159832 A CN 113159832A CN 202110328125 A CN202110328125 A CN 202110328125A CN 113159832 A CN113159832 A CN 113159832A
- Authority
- CN
- China
- Prior art keywords
- charging
- price
- time
- demand
- field model
- 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 41
- 238000012549 training Methods 0.000 claims abstract description 43
- 238000011160 research Methods 0.000 claims abstract description 19
- 230000003203 everyday effect Effects 0.000 claims abstract description 9
- 230000006870 function Effects 0.000 claims description 38
- 239000013598 vector Substances 0.000 claims description 21
- 238000010606 normalization Methods 0.000 claims description 17
- 230000002354 daily effect Effects 0.000 claims description 13
- 238000012545 processing Methods 0.000 claims description 6
- 125000004122 cyclic group Chemical group 0.000 claims description 4
- 230000014509 gene expression Effects 0.000 claims description 4
- 238000012546 transfer Methods 0.000 claims description 4
- 238000007476 Maximum Likelihood Methods 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000005192 partition Methods 0.000 claims description 3
- 239000000126 substance Substances 0.000 claims description 3
- 230000005611 electricity Effects 0.000 abstract description 8
- 230000004044 response Effects 0.000 abstract description 7
- 239000011159 matrix material Substances 0.000 description 3
- 230000002123 temporal effect Effects 0.000 description 3
- 230000002159 abnormal effect Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000013486 operation strategy Methods 0.000 description 2
- 230000001105 regulatory effect Effects 0.000 description 2
- 241000764238 Isis Species 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000001276 controlling effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000007599 discharging Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000029305 taxis Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
- G06Q30/0206—Price or cost determination based on market factors
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- 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—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Strategic Management (AREA)
- Finance (AREA)
- Development Economics (AREA)
- Accounting & Taxation (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Game Theory and Decision Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- General Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- Primary Health Care (AREA)
- Tourism & Hospitality (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a space-time price elasticity estimation method for a data-driven electric vehicle charging demand, and belongs to the field of electric vehicle demand response. Firstly, acquiring and normalizing charging historical data of each region in research time to obtain the normalized charging demand and charging price of each region every day and each time period and form a training set; establishing a conditional random field model, and selecting a topological structure of the conditional random field model and a characteristic function of the conditional random field model; training the conditional random field model by using a training set to obtain a trained conditional random field model; and calculating the influence of the charging price of any time period of any region on the charging demand of any time period of each region by using the trained conditional random field model to obtain the space-time price elasticity. The invention can help the power grid to effectively master the response of the electric vehicle user to the charging price, thereby reasonably formulating the time-sharing and regional electricity price and reducing the pressure of the power grid capacity.
Description
Technical Field
The invention belongs to the field of electric vehicle demand response, and particularly relates to a time-space price elasticity estimation method for a data-driven electric vehicle charging demand.
Background
With the advancement of technology and the push of policy, the development of electric vehicles is fast. For electric vehicles providing operation services in cities (such as taxis, network appointments, and the like), a large part of the charging requirements is met by public quick charging stations. Because the load of the electric automobile is very flexible, the charging demand of the electric automobile can be transferred from one time interval to another time interval and from one charging station to another charging station under the drive of price. That is, pricing of public charging stations can have an impact on the distribution of charging demand over time and space. In order to measure the response of the charging demand to the price, the price elasticity of the charging demand needs to be quantized, so that a power grid company can reduce the capacity pressure of a power distribution network by guiding the charging load of the electric vehicle, and reasonable distribution of power resources is realized.
In the case of orderly charging, an electric vehicle can become an asset of a power grid, not just a load of the power grid. From the perspective of the charging service provider, a sufficient number of electric vehicles owned by the charging service provider can provide services for the power grid, such as peak shaving resources and services. Price-leading of the charging demand is beneficial to the grid from a grid perspective. The price signal is a means for regulating and controlling the vehicles which do not accept the power grid dispatching command, and because the electric vehicle load cannot be directly controlled, the control signal needs to be converted into the price signal, and the charging demand is indirectly regulated and controlled under the condition of exciting a compatible price strategy.
Conventionally, a method of estimating the price elasticity of the charging demand is an experimental method of observing the change of the charging demand by modifying the charging price and then solving the price elasticity of the charging demand. However, experimental methods are difficult to implement, and no price-elastic method for evaluating charging requirements directly from charging data exists in current research. The patent CN202011014332.7 is a combined microgrid operation strategy considering demand response and electric vehicles and the patent CN201910317334.4 is a method for formulating space-time charging and discharging electricity prices of electric vehicles of a power distribution network, and both the strategy and the charging price strategy of the system operation are optimized by using a price elastic matrix of the charging demand of the electric vehicles. However, in both patents, there is no specific acquisition method of the price elasticity matrix for the demand for charging.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a space-time price elasticity estimation method for the charging requirement of a data-driven electric automobile. The invention can be applied to the field of electric vehicle charging, and according to the price elasticity result of the charging demand obtained by the invention, a power grid company can more effectively master the response of the electric vehicle user to the charging price, thereby reasonably formulating the time-sharing and regional electricity price and reducing the pressure of the power grid capacity.
The invention provides a space-time price elasticity estimation method for a data-driven electric vehicle charging demand, which is characterized by comprising the following steps of firstly obtaining charging historical data of each region in research time, carrying out normalization processing on the historical data, obtaining the charging demand and the charging price of each region after each time period normalization every day, and forming a training set; then establishing a conditional random field model, wherein the charging price is used as an explicit random variable, the charging demand is used as an implicit random variable, and the topological structure of the conditional random field model and the characteristic function of the conditional random field model are determined; training the conditional random field model by using a training set to obtain a trained conditional random field model; and calculating the influence of the charging price of any time period of any region on the charging demand of any time period of each region by using the trained conditional random field model to obtain the space-time price elasticity. The method comprises the following steps:
1) dividing research areas, acquiring charging historical data of each area in research time, and performing normalization processing on the historical data to obtain the charging demand and the charging price of each area after each time period normalization every day and form a training set; the method comprises the following specific steps:
1-1) dividing the region of interest into a plurality of regions of the same size, said divided regions being grouped into sets of regions
The set comprises k regions in total;
the time of day is divided into T time intervals, and the divided time intervals are combined into
1-2) obtaining the charging history data of all charging stations in each area divided in the step 1) in the research time to obtain the daily average charging quantity of each area in the research time
Number of charging stations N in each areai(ii) a And obtaining the total daily charge per period for each zone over the study time
Daily per time period weighted electricity charge price in study time
Weighted degree electric charge price averaged per time period day in research time
Average daily charge per time period over the study period
Wherein the subscript
Indicating the ith area, superscript
Denotes the t-th period, τ denotes the τ -th day;
1-3) for each region obtained in step 1-2)
And
respectively carrying out normalization to obtain the normalized total charging quantity of each region in each time period every day, namely the normalized charging demand, and the normalized weighted electric charging price of each region, namely the normalized charging price, wherein the normalization formulas are respectively as follows:
1-4) forming the normalized charging price and charging demand into a training set
Wherein the small brackets represent charging demand and charging price pairs for one area in one time period, and the middle brackets represent charging demand and charging price pair data for all areas in all time periods on day τ;
2) establishing a conditional random field model, comprising: selecting a topological structure of a conditional random field model, and selecting a characteristic function of the conditional random field model; the method comprises the following specific steps:
2-1) establishing a conditional random field model; in the conditional random field model, taking a charging price as an explicit random variable and taking a charging demand as an implicit random variable; each time interval of each area corresponds to a charging price and a charging demand respectively; all the displayed random variables form a node set V in a graph structure, each displayed random variable has a corresponding implicit random variable, and the correlation between the random variables is represented by an edge connecting two nodes; the set of edges connecting between the charging demands is represented by an edge set E, where E ═ Ek+EtThe edge set E comprises edge sets E which are topologically connected in spacekAnd a set of edges E topologically connected in timetIn which EkA spatial correlation representing the charging demand of all the zones in the same time period, EtRepresenting correlations of charging demands at different periods in the same area; the graph formed by the node set and the edge set is G ═ V, E;
the relationship between charge price and charge demand in the conditional random field model is as follows:
wherein:
wherein the content of the first and second substances,
is a vector of normalized charging requirements;
is the normalized charging demand vector;
is an unknown parameter of the conditional random field model, it，jtRespectively a node in the graph corresponding to the t-th period of the ith area and a node in the graph corresponding to the t-th period of the jth area,
t-th area of i-th area1，t2Nodes in the graph corresponding to the time intervals; omegaiI ∈ V is a self-elastic parameter for quantifying the charging demand d of the ith nodeiAnd its corresponding charge price ρiThe correlation between the two characteristic functions is phii(di，ρi)；
Is a parameter of spatial elasticity, used to quantify the charge between the ith and jth regions of the tth time periodElectric demand
The correlation between the characteristic functions matched with the correlation is
Is a parameter of spatial elasticity for quantifying t in the i-th region1，t2Demand for charging between time periods
The correlation between the characteristic functions matched with the correlation is
Chi is the demand of charging
A discrete value set of (a); | V | is the number of elements in the edge set; d is equal to x|V|A vector d of the charging requirement is represented by selecting a | V | sub-element from χ; z (rho) is a partition function;
2-2) selecting a correlation relation of charging demands on a space to be researched, wherein the method comprises the following steps: firstly, an interested area is selected, then each interested area and an adjacent area of the interested area are connected, all connected edges represent a path of charge demand transfer, and an edge set E is formedk；
2-3) selecting a charging demand correlation relation in a time period needing to be researched, wherein the method comprises the following steps: taking each time interval of each region, connecting the charging demand nodes of each time interval of the region together in pairs, and forming an edge set E by all connected edgest；
2-4) selecting three characteristic functions of the conditional random field model
And
the expressions are respectively as follows:
ψi(di，ρi)＝diρi，i∈V
3) training the conditional random field model established in the step 2) by using the training set obtained in the step 1) to obtain a trained conditional random field model; the method comprises the following specific steps:
3-1) maximizing a likelihood function;
solving unknown parameters omega to make training set
The probability of occurrence is maximal;
the log-likelihood function of ω is:
ln L(d|ρ；ω)＝ln Pω(d|ρ)
the gradient of the log-likelihood function is, according to three characteristic functions:
where E (-) is the expectation of the probability distribution;
3-2) training the conditional random field model by adopting a random gradient ascent algorithm with forgetting to obtain a trained conditional random field model; the method comprises the following specific steps:
3-2-1) initialization parameter ω0Assuming that the iteration number N is 0, a forgetting coefficient α, a learning rate γ, the number of days of the training set N, and a data selection probability for the ith day N are set
3-2-2) in training set
Using the daily data probability piSelecting charging demand and charging price vector data (d) of day ii，ρi)；
3-2-3) calculating the gradient of the maximum likelihood function for the nth iteration
3-2-4) update parameter omegan+1＝ωn+γΔωn；
3-2-5) calculate average loss of day i data
And judging that: if the data of the day has been selected m times, then
Wherein δ (i) represents a set of corresponding iteration times n when the data of the ith day is selected;
3-2-6) probability of changing data for all days being selected:
3-2-7) determining n: if n reaches the maximum number of iterations, ω is outputnOptimal parameter omega as conditional random field model*After the training of the conditional random field model is finished, entering the step 4); otherwise, making n equal to n +1, and then returning to the step 3-2-2);
4) estimating the space-time price elasticity of the charging demand by using the conditional random field model trained in the step 3); the method comprises the following specific steps:
4-1) estimating the probability distribution of the charging demand by utilizing an annular belief propagation algorithm for a cyclic graph in the conditional random field; the method comprises the following specific steps:
4-1-1) initialization information
The initial iteration number n is 1, wherein
Representing that in the nth iteration, when the charging requirement of the node j is selected from the discrete set x, the value is djInformation passed from node i to node j;
4-1-2) all discrete values d corresponding to the charging requirements of all edges (i, j) E and nodes jjE is x; calculating information passed by node i to node j
Where N (i) represents a set of nodes connected to node i, N (i) \ { j } represents a node connected to node i and does not contain a set of nodes j, ωi，jψi，j(di，dj) Representing a correlation between charging demands;
4-1-3) normalizing different information of the same node j
4-1-4) determination: if the convergence condition is reached
Or when the maximum iteration times is reached, 4-1-5) operation is carried out; otherwise, making the iteration number n equal to n +1, and then returning to the step 4-1-2);
4-1-5) calculating the beliefs according to the converged information, wherein the calculation method comprises the following steps:
4-1-6) normalizing the beliefs to:
4-1-7) the expectation of the implicit random variable to be solved is:
4-2) locally linearizing at a given charge price to obtain the price elasticity of the charge demand;
t of ith area1Time interval charging price to jth area t2Price elasticity for impact of time interval charging demand
Represents; in the ith area t with a set length Deltap1Time-interval-given charging price
Get two charging prices in its vicinity
And keeping the prices of the remaining regions unchanged, thereby obtaining two price vectors containing all periods of all regions
Respectively using the annular belief propagation algorithm of the step 4-1) for the two price vectors to calculate the jth area t respectively corresponding to the two charging price vectors2Desired charging demand for time period
And
the price elasticity expression for the charging demand is obtained as follows:
the invention has the characteristics and beneficial effects that:
1. the invention relates to a data-driven charge price elasticity estimation method, which utilizes the charge historical data of a charge service provider in a period of time, and the obtained price elasticity result is more real and effective.
2. The method mainly reflects the transfer of the charging requirement of the electric automobile in time and space, and simultaneously models the correlation of the charging requirement in time and space by using a conditional random field model. The invention can flexibly adjust the charging requirement correlation to be considered by adjusting the topological graph structure according to local conditions to obtain the required price elasticity required by the user.
3. The invention can be applied to the field of electric vehicle charging, and according to the price elasticity result of the charging demand obtained by the invention, a power grid company can more effectively master the response of an electric vehicle user to the charging price, thereby reasonably formulating the time-of-use and regional electricity price, guiding the reasonable distribution of the charging load at different time and different places, further balancing the utilization rate of the power grid capacity in time and space, and reducing the pressure of the power grid capacity.
Drawings
FIG. 1 is an overall flow diagram of the method of the present invention.
Fig. 2 is a topological diagram of charging demand and charging price in space and time in an embodiment of the invention.
Detailed Description
The invention provides a space-time price elasticity estimation method for a charging demand of a data-driven electric vehicle, and the invention is further described in detail below by combining the attached drawings and specific embodiments.
The invention relates to a method for estimating the space-time price elasticity of the charging demand of an electric vehicle user under the condition of the known historical charging record of a charging service provider. The invention is suitable for charging service providers with more built charging stations and needs to accumulate certain operation data.
The invention provides a space-time price elasticity estimation method for a data-driven electric vehicle charging demand, which comprises the steps of firstly obtaining charging historical data of each region in research time, carrying out normalization processing on the historical data, obtaining the charging demand and the charging price of each region after each time period normalization every day, and forming a training set; then establishing a conditional random field model, wherein the charging price is used as an explicit random variable, the charging demand is used as an implicit random variable, and the topological structure of the conditional random field model and the characteristic function of the conditional random field model are determined; training the conditional random field model by using a training set to obtain a trained conditional random field model; and calculating the influence of the charging price of any time period of any region on the charging demand of any time period of each region by using the trained conditional random field model to obtain the space-time price elasticity.
The overall flow of the method is shown in fig. 1, and comprises the following steps:
1) and dividing research areas, acquiring charging historical data of each area in research time, and performing normalization processing on the historical data to obtain the normalized charging demand and charging price of each area every day and each time period and form a training set.
2) Establishing a conditional random field model, comprising: selecting a topological structure of the conditional random field model, and selecting a characteristic function of the conditional random field model.
3) Training the conditional random field model established in the step 2) by using the training set obtained in the step 1) to obtain a trained conditional random field model.
4) And estimating the space-time price elasticity of the charging demand by using the conditional random field model trained in the step 3).
In the method, the step 1) comprises the following specific steps:
1-1) dividing the region (usually a city) into a plurality of regions with the same size, dividing the region by using a grid or honeycomb network, usually dividing the region into a plurality of square regions with the side length of 3-5 km, and recording the set of the divided regions as
The set collectively contains k regions.
The time of one day is divided into T periods, and the whole day is divided into three periods in this embodiment in a period division manner similar to the beijing industrial and commercial electricity price, wherein the peak period: [10h, 15h ]]∪[16h，17h]∪[18h，21h]And at the ordinary time stage: [7h, 10h ]]∪[15h，16h]∪[17h，18h]∪[21h，23h]And a valley period: [23h, 0h]∪[0h，7h]The time interval set divided into is
Where H represents a peak period, M represents a flat period, and L represents a valley period.
1-2) obtaining the charging records of all charging stations in each area divided in the step 1) in the study time (usually, the latest period of time is selected), and obtaining the daily average charging quantity of each area in the study time
Number of charging stations N in each areai(one) of the steps. And the total daily time-interval per time-interval for each region over a selected time-interval (typically three time-intervals, peak-to-valley) during the study time is obtainedAmount of charge
Daily per time period weighted electricity charge price in study time
Weighted degree electric charge price averaged per time period day in research time
kWh), average daily charge per period of time in the study
Wherein the subscript
Indicating the ith area, superscript
Denotes the t-th period and τ denotes the τ -th day.
1-3) for each region obtained in step 1-2)
And
respectively carrying out normalization to obtain a normalized total charging quantity (namely normalized charging demand) of each region in each time period per day and a normalized weighted electric charging price (namely normalized charging price) of each region, wherein the normalization formulas are respectively as follows:
1-4) forming the normalized charging price and charging demand into a training set
Where the small brackets represent the charging demand and charging price pair for one area in one period, and the middle brackets represent the charging demand and charging price pair data for all areas in all periods on day τ.
The step 2) comprises the following specific steps:
2-1) establishing a conditional random field model under the charging problem. In the conditional random field model, the charging price is used as an explicit random variable, and the charging demand is used as an implicit random variable. One period of a region corresponds to one charge price, one charge demand. All display random variables (namely the charging requirements of each area in each period) are formed into a node set V in a graph structure, each display random variable has a corresponding implicit random variable (charging price), and the correlation relationship between the random variables is represented by an edge connecting two nodes, and the method comprises the following steps: the charging price and the corresponding charging requirement correlation relationship and the correlation relationship between the charging requirements; the edge set of the connection between the charging demands is represented by an edge set E, and the edge set E is further divided into E ═ Ek+EtIncluding a set of spatially topologically connected edges EkAnd a set of edges E topologically connected in timetIn which EkThe spatial correlation representing the charging demand of all regions in the same time period, in this embodiment
Three periods, i.e. EkThe three-dimensional space topology consists of three layers of same space topologies in three time periods; etShows the correlation of the charging requirements in different periods of time in the same region, in this embodiment there are k regions, i.e. EtConsisting of k identical time topologies. The graph formed by node set and edge set is G ═ V, E, and this embodiment is shown in fig. 2. The conditional random field model can infer the charge demand after the charge price is known.
The relationship between charge price and charge demand in the conditional random field model is as follows:
wherein:
wherein the content of the first and second substances,
is a vector of normalized charging requirements, in this embodiment
Is a normalized charge demand vector, in this example
For simplifying writing, the following
Substitution
Representing normalized price and demand.
Is an unknown parameter of the conditional random field model, where it，jtRespectively refer to a node in the graph corresponding to the t-th period of the ith area and a node in the graph corresponding to the t-th period of the jth area,
refer to the ith region1，t2The time intervals correspond to nodes in the graph. OmegaiI ∈ V is a self-elastic parameter for quantifying the charging demand d of the ith nodeiAnd its corresponding charge price ρiCorrelation between (note here d)iI in the topology map, and
i in (i) refers to the ith area), the characteristic function matched with the i area is psii(di，ρi)；
Is a parameter of space elasticity, used for quantifying the charging requirement between the ith area and the jth area in the tth period
The correlation between the characteristic functions matched with the correlation is
Is a parameter of spatial elasticity for quantifying t in the i-th region1，t2Demand for charging between time periods
The correlation between the characteristic functions matched with the correlation is
In the present embodiment, it is preferred that,
chi is for fillingElectric demand
A set of discrete values. | V | is the number of elements in the edge set. So d ∈ χ|V|The vector d is shown where the | V | sub-element in χ is selected to form the charging demand. The partition function Z (ρ) is used to normalize the probability.
And
the three characteristic functions are used for describing the correlation between the charging price and the charging demand, the correlation between the spatial charging demand and the correlation between the temporal charging demand.
2-2) selecting the correlation of the charging requirement on the space to be researched. Establishment of a charging demand correlation relationship on a space: firstly, an interested area needs to be selected, which is generally an area with a large number of charging stations and a large daily average charge (10 areas are selected in this embodiment); then connecting each interested area and the adjacent area of the interested area, wherein all the connected edges represent the path of the charge demand transfer and form an edge set Ek；
2-3) selecting the charging demand correlation relation in the time period needing to be researched. Establishment of a temporal charging demand correlation: taking each time interval of each region, connecting the charging demand nodes of each time interval of the region together in pairs, and forming an edge set E by all connected edgest。
The final topology of the present embodiment is shown in fig. 2. The topology is divided into three layers of peaks, levels and valleys according to time periods, wherein white circles represent charging requirements (implicit random variables), black circles represent charging prices (explicit random variables), and connecting lines represent correlation relations among the random variables. Therefore, in fig. 2, there is a correlation between the charge price and the charge demand as well as a correlation between the charge demands.
2-4) selecting three characteristic functions of the conditional random field model
And
the three characteristic functions respectively describe the correlation between the charging price and the charging demand, the spatial correlation between the charging demand and the temporal correlation between the charging demand. The following can be selected:
ψi(di，ρi)＝diρi，i∈V
the step 3) comprises the following specific steps:
3-1) maximizing the likelihood function: solving unknown parameters omega to make training set
The probability of occurrence is the greatest.
The log-likelihood function of ω is:
ln L(d|ρ；ω)＝ln Pω(d|ρ)
the gradient of the log-likelihood function is, according to three characteristic functions:
where E (-) is the expectation of the probability distribution.
3-2) training method: training the conditional random field model by adopting a random gradient ascent algorithm with forgetting to obtain a trained conditional random field model; in each iteration, charging price and charging demand data of one day in the training set are randomly selected, and gradient rising is carried out by using the data of the day. The absolute value of the gradient of the abnormal data is relatively large, the selection probability of the abnormal data is gradually reduced by using the gradient value in the training process, and the algorithm content is as shown in algorithm one:
in this embodiment, the parameters in algorithm one are set as follows: the initialization parameters are initialized by standard normal distribution, the learning rate is 0.05, the forgetting coefficient is 5, and the training days are 1000.
3-2-1) initialization parameter ω0Assuming that the iteration number N is 0, a forgetting coefficient α, a learning rate γ, the number of days of the training set N, and a data selection probability for the ith day N are set
3-2-2) in training set
Using the daily data probability piSelecting charging demand and charging price vector data (d) of day ii，ρi)。
3-2-3) calculating the gradient of the maximum likelihood function of the nth iteration
When the expectation of the charging requirement is obtained, the algorithm of the steps 4-1-1) -4-1-7) is needed.
3-2-4) updating unknown parameter omegan+1＝ωn+γΔωn；
3-2-5) calculate average loss of day i data
And judging that: if the data of the day has already been recordedThe selection times are m times, then
Where δ (i) represents the set of corresponding iterations n at the time the day i data was selected.
3-2-6) the probability of changing the data chosen for all days, taking day i as an example:
3-2-7) determining n: if n reaches the maximum number of iterations (set to 5000 in this embodiment), the current ω is outputnOptimal parameter omega as conditional random field model*After the training of the conditional random field model is finished, entering the step 4); otherwise, let n be n +1, and then return to step 3-2-2).
The step 4) comprises the following specific steps:
4-1) a ring belief propagation algorithm for a cyclic graph in a conditional random field: the inference of the conditional random field model is at the optimal parameter ω*Next, using the charge prices d of all the regions, a probability distribution of the charge demand ρ is inferred. The last selected graph is typically a circled graph, as shown in fig. 2. To infer the probability distribution of the charging demand, its circular belief propagation algorithm is as follows:
in the present embodiment, the parameter settings in algorithm 2 are as follows: the initialization information is all 1, and the discrete set of charging requirements is χ { -1, -0.5, 0, 0.5, 1}, and ε { -0.0001. For writing convenience, the time division and the space division are not distinguished any more, and the correlation between the charging requirements is unified
And
writing omegai，jψi，j(di，dj)。
4-1-1) initialization information
The initial iteration number n is 1, wherein
Representing that in the nth iteration, when the charging requirement of the node j is selected from the discrete set χ, the value is djInformation passed from node i to node j.
4-1-2) all discrete values d corresponding to the charging requirements of all edges (i, j) E and nodes jjE.x. Calculating information passed by node i to node j
Where N (i) represents a set of nodes connected to node i, and N (i) \ { j } represents a node connected to node i and does not contain a set of nodes j.
4-1-3) normalizing different information of the same node j
4-1-4) determination: if the convergence condition is reached
Or when the maximum iteration times (the value is 100 in the embodiment) is reached, the operation is skipped to 4-1-5); otherwise, making the iteration number n equal to n +1, and then returning to the step 4-1-2);
4-1-5) calculating the beliefs according to the converged information, wherein the calculation method comprises the following steps:
4-1-6) normalizing the beliefs to:
4-1-7) the expectation of the implicit random variable to be solved is:
4-2) local linearization at a given charge price yields the price elasticity of the charge demand:
t of ith area1Time interval charging price to jth area t2Price elasticity for impact of time interval charging demand
Denotes (i and j may be the same, t1，t2Or the same) in a smaller step Δ ρ (0.05 bins given in this example) in the ith region t1Time-interval-given charging price
(1 yuan is given in this example) and two charge prices in the vicinity thereof are taken
(1.05, 0.95 dollars for this example) and keep the remaining zone prices unchanged, thereby resulting in two price vectors that contain all periods of all zones
Using a cyclic belief propagation algorithm of 4-1-1) -4-1-7) for the two price vectors, respectively, it is possible to deduce the corresponding jth region t under the two charge price vectors2Desired charging demand for time period
This makes it possible to obtain the price flexibility of the charging demand, as follows:
the price elastic matrix solved by the invention can be used for generating an optimal operation strategy and a time-of-use electricity price strategy of a power grid, and related patents can be found in CN202011014332.7 and CN 201910317334.4.
Although the present invention has been described in detail with reference to the above embodiments, it should be understood by those skilled in the art that: modifications and equivalents may be made to the specific embodiments of the invention without departing from the spirit and scope of the invention, which should be construed to be covered by the claims.
Claims (2)
1. The method is characterized in that charging historical data of each region in research time are obtained, normalization processing is carried out on the historical data, the charging demand and the charging price of each region after each time period normalization every day are obtained, and a training set is formed; then establishing a conditional random field model, wherein the charging price is taken as an explicit random variable, the charging demand is taken as an implicit random variable, and the topological structure of the conditional random field model and the characteristic function of the conditional random field model are selected; training the conditional random field model by using a training set to obtain a trained conditional random field model; and calculating the influence of the charging price of any time period of any region on the charging demand of any time period of each region by using the trained conditional random field model to obtain the space-time price elasticity.
2. A method as claimed in claim 1, characterized in that the method comprises the following steps:
1) dividing research areas, acquiring charging historical data of each area in research time, and performing normalization processing on the historical data to obtain the charging demand and the charging price of each area after each time period normalization every day and form a training set; the method comprises the following specific steps:
1-1) dividing the region of interest into a plurality of regions of the same size, said divided regions being grouped into sets of regions
The set comprises k regions in total;
the time of day is divided into T time intervals, and the divided time intervals are combined into
1-2) obtaining the charging history data of all charging stations in each area divided in the step 1) in the research time to obtain the daily average charging quantity of each area in the research time
Number of charging stations N in each areai(ii) a And obtaining the total daily charge per period for each zone over the study time
) The electric charging price of each time interval of each day in the research time
Weighted degree electric charge price averaged per time period day in research time
Average daily charge per time period over the study period
Wherein the subscript
Indicating the ith area, superscript
Denotes the t-th period, τ denotes the τ -th day;
1-3) for each region obtained in step 1-2)
And
respectively carrying out normalization to obtain the normalized total charging quantity of each region in each time period every day, namely the normalized charging demand, and the normalized weighted electric charging price of each region, namely the normalized charging price, wherein the normalization formulas are respectively as follows:
1-4) forming the normalized charging price and charging demand into a training set
Wherein the small brackets represent the charging demand and charging price pairs for an area over a period of time, and the middle brackets represent day τCharging demand and charging price pair data for all regions at all time periods;
2) establishing a conditional random field model, comprising: selecting a topological structure of a conditional random field model, and selecting a characteristic function of the conditional random field model; the method comprises the following specific steps:
2-1) establishing a conditional random field model; in the conditional random field model, taking a charging price as an explicit random variable and taking a charging demand as an implicit random variable; each time interval of each area corresponds to a charging price and a charging demand respectively; all the displayed random variables form a node set V in a graph structure, each displayed random variable has a corresponding implicit random variable, and the correlation between the random variables is represented by an edge connecting two nodes; the set of edges connecting between the charging demands is represented by an edge set E, where E ═ Ek+EtThe edge set E comprises edge sets E which are topologically connected in spacekAnd a set of edges E topologically connected in timetIn which EkA spatial correlation representing the charging demand of all the zones in the same time period, EtRepresenting correlations of charging demands at different periods in the same area; the graph formed by the node set and the edge set is G ═ V, E;
the relationship between charge price and charge demand in the conditional random field model is as follows:
wherein:
wherein the content of the first and second substances,
is a vector of normalized charging requirements;
is the normalized charging demand vector;
is an unknown parameter of the conditional random field model, it，jtRespectively a node in the graph corresponding to the t-th period of the ith area and a node in the graph corresponding to the t-th period of the jth area,
t-th area of i-th area1，t2Nodes in the graph corresponding to the time intervals; omegaiI ∈ V is a self-elastic parameter for quantifying the charging demand d of the ith nodeiAnd its corresponding charge price ρiThe correlation between the two characteristic functions is phii(di，ρi)；
Is a parameter of space elasticity, used for quantifying the charging requirement between the ith area and the jth area in the tth period
The correlation between the characteristic functions matched with the correlation is
Is a parameter of spatial elasticity for quantifying t in the i-th region1，t2Demand for charging between time periods
The correlation between the characteristic functions matched with the correlation is
For charging requirements
A discrete value set of (a); | V | is the number of elements in the edge set;
is shown in
Selecting a | V | sub-element to form a vector d of the charging requirement; z (rho) is a partition function;
2-2) selecting a correlation relation of charging demands on a space to be researched, wherein the method comprises the following steps: firstly, an interested area is selected, then each interested area and an adjacent area of the interested area are connected, all connected edges represent a path of charge demand transfer, and an edge set E is formedk；
2-3) selecting a charging demand correlation relation in a time period needing to be researched, wherein the method comprises the following steps: taking each time interval of each region, connecting the charging demand nodes of each time interval of the region together in pairs, and forming an edge set E by all connected edgest；
2-4) selecting three characteristic functions psi of the conditional random field modeli(di，ρi)，
And
the expressions are respectively as follows:
ψi(di，ρi)＝diρi，i∈V
3) training the conditional random field model established in the step 2) by using the training set obtained in the step 1) to obtain a trained conditional random field model; the method comprises the following specific steps:
3-1) maximizing a likelihood function;
solving unknown parameters omega to make training set
The probability of occurrence is maximal;
the log-likelihood function of ω is:
lnL(d|ρ；ω)＝lnPω(d|ρ)
the gradient of the log-likelihood function is, according to three characteristic functions:
where E (-) is the expectation of the probability distribution;
3-2) training the conditional random field model by adopting a random gradient ascent algorithm with forgetting to obtain a trained conditional random field model; the method comprises the following specific steps:
3-2-1) initialization parameter ω0Assuming that the iteration number N is 0, a forgetting coefficient α, a learning rate γ, the number of days of the training set N, and a data selection probability for the ith day N are set
3-2-2) in training set
Using the daily data probability piSelecting charging demand and charging price vector data (d) of day ii，ρi)；
3-2-3) calculating the gradient of the maximum likelihood function for the nth iteration
3-2-4) update parameter omegan+1＝ωn+γΔωn；
3-2-5) calculate average loss of day i data
And judging that: if the data of the day has been selected m times, then
Wherein δ (i) represents a set of corresponding iteration times n when the data of the ith day is selected;
3-2-6) probability of changing data for all days being selected:
3-2-7) determining n: if n reaches the maximum number of iterations, ω is outputnAs conditional random fieldsOptimal parameter omega of model*After the training of the conditional random field model is finished, entering the step 4); otherwise, making n equal to n +1, and then returning to the step 3-2-2);
4) estimating the space-time price elasticity of the charging demand by using the conditional random field model trained in the step 3); the method comprises the following specific steps:
4-1) estimating the probability distribution of the charging demand by utilizing an annular belief propagation algorithm for a cyclic graph in the conditional random field; the method comprises the following specific steps:
4-1-1) initialization information
The initial iteration number n is 1, wherein
Indicating that in the nth iteration, the charging requirement of node j is from a discrete set
In the selection of value djInformation passed from node i to node j;
4-1-2) all discrete values corresponding to the charging requirements of all edges (i, j) E E and node j
Calculating information passed by node i to node j
Where N (i) represents a set of nodes connected to node i, N (i) \ { j } represents a node connected to node i and does not contain a set of nodes j, ωi，jψi，j(di，dj) Representing a correlation between charging demands;
4-1-3) normalizing different information of the same node j
4-1-4) determination: if the convergence condition is reached
Or when the maximum iteration times is reached, 4-1-5) operation is carried out; otherwise, making the iteration number n equal to n +1, and then returning to the step 4-1-2);
4-1-5) calculating the beliefs according to the converged information, wherein the calculation method comprises the following steps:
4-1-6) normalizing the beliefs to:
4-1-7) the expectation of the implicit random variable to be solved is:
4-2) locally linearizing at a given charge price to obtain the price elasticity of the charge demand;
t of ith area1Time interval charging price to jth area t2Time interval chargingPrice elasticity for impact of electrical demand
Represents; in the ith area t with a set length Deltap1Time-interval-given charging price
Get two charging prices in its vicinity
And keeping the prices of the remaining regions unchanged, thereby obtaining two price vectors containing all periods of all regions
Respectively using the annular belief propagation algorithm of the step 4-1) for the two price vectors to calculate the jth area t respectively corresponding to the two charging price vectors2Desired charging demand for time period
And
the price elasticity expression for the charging demand is obtained as follows:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110328125.7A CN113159832A (en) | 2021-03-26 | 2021-03-26 | Time-space price elasticity estimation method for data-driven electric vehicle charging demand |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110328125.7A CN113159832A (en) | 2021-03-26 | 2021-03-26 | Time-space price elasticity estimation method for data-driven electric vehicle charging demand |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113159832A true CN113159832A (en) | 2021-07-23 |
Family
ID=76885623
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110328125.7A Pending CN113159832A (en) | 2021-03-26 | 2021-03-26 | Time-space price elasticity estimation method for data-driven electric vehicle charging demand |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113159832A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113450023A (en) * | 2021-07-28 | 2021-09-28 | 东北电力大学 | Electric automobile ordered charging method under gridding time-of-use electricity price |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109910670A (en) * | 2019-04-19 | 2019-06-21 | 国网能源研究院有限公司 | A kind of power distribution network electric car space-time charge and discharge electricity pricing method |
| CN112202201A (en) * | 2020-09-24 | 2021-01-08 | 上海电机学院 | Joint microgrid operation strategy considering demand response and electric automobile |
-
2021
- 2021-03-26 CN CN202110328125.7A patent/CN113159832A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109910670A (en) * | 2019-04-19 | 2019-06-21 | 国网能源研究院有限公司 | A kind of power distribution network electric car space-time charge and discharge electricity pricing method |
| CN112202201A (en) * | 2020-09-24 | 2021-01-08 | 上海电机学院 | Joint microgrid operation strategy considering demand response and electric automobile |
Non-Patent Citations (1)
| Title |
|---|
| ZECHUN HU等: "Pricing mechanisms design for guiding electric vehicle charging to fill load valley", 《APPLIED ENERGY》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113450023A (en) * | 2021-07-28 | 2021-09-28 | 东北电力大学 | Electric automobile ordered charging method under gridding time-of-use electricity price |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Liang et al. | Mobility-aware charging scheduling for shared on-demand electric vehicle fleet using deep reinforcement learning | |
| Lee et al. | Federated reinforcement learning for energy management of multiple smart homes with distributed energy resources | |
| Shahriar et al. | Machine learning approaches for EV charging behavior: A review | |
| You et al. | Digital twins based day-ahead integrated energy system scheduling under load and renewable energy uncertainties | |
| CN112614009B (en) | Power grid energy management method and system based on deep expectation Q-learning | |
| Martinez-Pabon et al. | Optimizing residential energy management using an autonomous scheduler system | |
| CN110929948A (en) | Fully distributed intelligent power grid economic dispatching method based on deep reinforcement learning | |
| Charbonnier et al. | Scalable multi-agent reinforcement learning for distributed control of residential energy flexibility | |
| Unterluggauer et al. | Short‐term load forecasting at electric vehicle charging sites using a multivariate multi‐step long short‐term memory: A case study from Finland | |
| CN112785029B (en) | Charging station power consumption prediction method based on deep clustering model | |
| CN113159832A (en) | Time-space price elasticity estimation method for data-driven electric vehicle charging demand | |
| CN110264012A (en) | Renewable energy power combination prediction technique and system based on empirical mode decomposition | |
| CN114119159A (en) | Network taxi appointment real-time order matching and idle vehicle scheduling method and system | |
| Kim et al. | Building energy management for demand response using kernel lifelong learning | |
| Wang et al. | Deep reinforcement learning for energy trading and load scheduling in residential peer-to-peer energy trading market | |
| Wang et al. | A reinforcement learning approach for EV charging station dynamic pricing and scheduling control | |
| CN115409431A (en) | Distributed power resource scheduling method based on neural network | |
| Hafezi et al. | Framework for development of the Scheduler for Activities, Locations, and Travel (SALT) model | |
| Shi et al. | Predicting household energy consumption in an aging society | |
| Boulakhbar et al. | Electric vehicles arrival and departure time prediction based on deep learning: the case of Morocco | |
| Li et al. | Online transfer learning-based residential demand response potential forecasting for load aggregator | |
| Hui et al. | Regulation capacity evaluation of large-scale residential air conditioners for improving flexibility of urban power systems | |
| van Tilburg et al. | MARL-iDR: Multi-Agent Reinforcement Learning for Incentive-Based Residential Demand Response | |
| Hassan et al. | Data-driven learning and load ensemble control | |
| CN114611823B (en) | Optimized dispatching method and system for electricity-cold-heat-gas multi-energy-demand typical park |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210723 |
|
| RJ01 | Rejection of invention patent application after publication |