Electric vehicle charging load space-time prediction method based on inhomogeneous Markov space-time network

Abstract

The invention relates to an electric vehicle charging load space-time prediction method based on a non-homogeneous Markov space-time network, and belongs to the field of electric vehicle charging load prediction. The method comprises two parts of urban area spatial-temporal feature construction and spatial-temporal prediction model construction. The urban area spatio-temporal feature construction module is used for extracting and constructing spatial distribution and time change features of electric vehicle charging loads so as to capture spatio-temporal relevance in an urban area; the space-time prediction model utilizes a method of combining a non-homogeneous Markov chain and a neural network to model the space-time relevance of the charging load of the electric vehicle, and the space-time relevance is used as a high-precision prediction core method. According to the method, efficient prediction of the charging load of the electric vehicle under the urban spatial scale can be realized under the condition of complex spatio-temporal data. Through experimental comparison, the prediction method provided by the invention can effectively improve the prediction precision, has good stability, and can adapt to load changes of different application scenes.

Description

Technical Field

[0001] The invention belongs to the field of electric vehicle charging load prediction, and relates to a method for spatiotemporal prediction of electric vehicle charging load based on a non-homogeneous Markov spatiotemporal network. Background Art

[0002] With the rapid growth of the number of electric vehicles (EVs), EV charging load prediction has become a key technology in smart grid, energy management and charging infrastructure construction. Accurate prediction of charging load not only helps grid load scheduling and rational allocation of energy resources, but also effectively avoids power system overload and power waste. However, the prediction of charging load faces many challenges, mainly reflected in the spatiotemporal complexity and dynamic variability of data. The charging behavior of electric vehicles is affected by many factors, including geographical location, time, weather conditions, traffic flow, and user behavior habits. These factors make the charging load highly uncertain and volatile, which brings great challenges to traditional load forecasting methods.

[0003] Most of the existing charging load forecasting methods are based on time series analysis, regression models or traditional machine learning algorithms, but these methods usually fail to fully consider spatiotemporal characteristics. The distribution of charging load varies significantly in different time and space, while traditional methods often ignore the complexity of spatial data and have poor adaptability to dynamic changes. For example, the fluctuation of charging demand in different regions varies greatly, and the load in urban areas is affected by factors such as traffic flow, distribution of charging stations and user demand, making it difficult for traditional forecasting models to accurately reflect actual charging demand. In addition, many existing methods rely on static models and cannot flexibly handle the nonlinear characteristics of electric vehicle charging load over time, and cannot effectively cope with data changes in real-time forecasting.

[0004] How to design a high-precision load forecasting method that can comprehensively consider the complexity and dynamic changes of spatiotemporal data has become a difficult problem that needs to be solved urgently in the field of electric vehicle charging. Summary of the invention

[0005] In view of this, the purpose of the present invention is to provide a spatiotemporal prediction method for electric vehicle charging load based on a non-homogeneous Markov spatiotemporal network, to enhance the correlation modeling capability of spatiotemporal data, and to effectively improve the accuracy and stability of charging load prediction.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] A method for spatiotemporal prediction of electric vehicle charging load based on a non-homogeneous Markov spatiotemporal network, the method comprising the following steps:

[0008] S1: Construct the spatiotemporal characteristics of urban areas, divide urban spatial areas through gridding, and then extract the functional spatial distribution of residential areas, work areas, and commercial areas. Establish the urban road topology based on real roads, and establish the spatiotemporal characteristic model of the city based on the daily travel mode statistics of electric vehicle data;

[0009] S2: Using the urban spatiotemporal characteristics and based on the transfer value function, a non-homogeneous Markov spatiotemporal state transfer matrix is ​​constructed to form a non-homogeneous Markov spatiotemporal network and establish the urban spatiotemporal transfer feature connection;

[0010] S3: Based on the Monte Carlo simulation process of random energy particles, an electric vehicle charging behavior model is established to output the spatiotemporal distribution prediction of the city-wide electric vehicle charging load;

[0011] S4: Combine local prediction data and real data to establish a loss function, calculate the reverse error propagation of the non-homogeneous Markov network, use the gradient descent method to optimize the transfer matrix parameters, and improve the model prediction performance.

[0012] Furthermore, in S1, the construction of spatiotemporal features of urban areas includes three parts:

[0013] 1) Modeling the spatial distribution of urban residential areas, work areas, and commercial areas. Using ArcGIS software, we used the land use planning map of the study area for georeferencing and divided the entire urban area into grids to obtain the distribution of various functional areas in the city in different grids. These data were stored in the matrix

[0014] 2) Urban road network topology, using online map websites to construct a topological road network through the topological connection relationship between road geographic data to represent the connection between adjacent grid units; matrix D t The topological road network used to represent the entire simulation area is defined as follows:

[0015]

[0016] Where x represents the horizontal coordinate of the grid position; y represents the vertical coordinate of the grid position; D c (z),z∈[1,8] represents the connection direction of the grid position, D c (z), z = 1, 2, ..., 8 represent the index of the connection direction, including east, west, south, north, southeast, southwest, northeast and northwest;

[0017] The road topology modeling is performed on N grid cells in the study area, and the Floyd shortest path search algorithm is used to calculate the shortest road distance between any two spaces to obtain the transfer distance matrix between any two spaces.

[0018]

[0019] 3) The probability distribution of daily travel purposes is statistically analyzed. The electric vehicle user group has different travel totals at different times in a 24-hour day. According to the different travel purposes, the travel at each time is divided into five types: travel from residential area to work area H2W, travel from residential area to commercial area H2C, travel between work area and commercial area W&C, travel from work area to residential area W2H, and travel from commercial area to residential area C2H.

[0020] Furthermore, in S2, the city's spatiotemporal characteristics are used to construct a non-homogeneous Markov spatiotemporal state transfer matrix based on the transfer value function, forming a non-homogeneous Markov spatiotemporal network, and establishing a city's spatiotemporal transfer feature connection. The transfer value function is defined as:

[0021]

[0022] Among them, h i ,w i ,c i represents the h, w, c attributes of space i, h j ,w j ,c j represents the h, w, c attributes of space j, I t Represents the probability of different travel purposes corresponding to time t;

[0023] The non-homogeneous Markov transition matrix is ​​defined as:

[0024]

[0025] Furthermore, in S3, the random process energy particle is defined as:

[0026] e={τ,γ,υ|τ∈{1,2,...,L},γ∈[0,1],υ∈{0,1}}

[0027] Among them, τ represents the urban grid space where the particle is located, γ represents the energy of the particle, and its value range is [0,1], and υ represents the motion state of the particle. If υ=1, it means that the particle is in motion, and if υ=0, it means that the particle is in a stopped state.

[0028] Furthermore, in S4, the loss function is defined as:

[0029]

[0030] The error back propagation formula for the non-homogeneous Markov space-time network is defined as:

[0031]

[0032] The gradient descent method is defined as:

[0033]

[0034] The beneficial effects of the present invention are:

[0035] (1) By constructing an urban spatiotemporal characteristic model and a non-homogeneous Markov spatiotemporal network, the dynamic changes and non-homogeneous correlations of charging loads in different time and space can be effectively captured, thus improving the adaptability and expressiveness of the prediction model for complex data.

[0036] (2) Based on the Monte Carlo simulation process of random energy particles, a charging behavior model of electric vehicles is established, which outputs the precise spatiotemporal distribution of charging load in the urban area from a global perspective, significantly improving the prediction accuracy and reliability of the model.

[0037] (3) By establishing a loss function and combining it with the gradient descent method to optimize the transfer matrix parameters, the prediction error can be effectively reduced, the stability and robustness of the model can be enhanced, and the actual needs of large-scale urban charging load prediction can be met.

[0038] (4) The present invention can adapt to the complex spatial structure and changes in charging demand in different cities, provide a scientific basis for smart grid optimization scheduling, charging station layout planning and power resource management, and has broad application prospects.

[0039] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0041] Figure 1 This is the overall framework diagram of the non-homogeneous Markov space-time network model;

[0042] Figure 2 Modeling spatial distribution maps for urban functional areas;

[0043] Figure 3 It is the topological structure diagram of urban roads;

[0044] Figure 4 is a non-homogeneous Markov space-time network graph. DETAILED DESCRIPTION

[0045] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and the features in the embodiments can be combined with each other without conflict.

[0046] Among them, the drawings are only used for illustrative explanations, and they only represent schematic diagrams rather than actual pictures, and should not be understood as limitations on the present invention. In order to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of actual products. For those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0047] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right", "front", "rear", etc. indicate the orientation or position relationship, they are based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, the terms describing the position relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting the present invention. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0048] The electric vehicle charging load prediction method based on the non-homogeneous Markov spatiotemporal network of the present invention is as follows: Figure 1 As shown, the method mainly includes the following four steps:

[0049] The first part is to construct the spatiotemporal characteristics of urban areas. The urban spatial areas are divided by gridding, and then the functional spatial distribution of residential areas, work areas, and commercial areas is extracted. The urban road topology is established based on real roads. The spatiotemporal characteristic model of the city is established based on the daily travel mode statistics of electric vehicle data.

[0050] In the second part, using the urban spatiotemporal characteristics and based on the transfer value function, a non-homogeneous Markov spatiotemporal state transfer matrix is ​​constructed to form a non-homogeneous Markov spatiotemporal network and establish the urban spatiotemporal transfer feature connection;

[0051] In the third part, based on the Monte Carlo simulation process of random energy particles, an electric vehicle charging behavior model is established to output the spatiotemporal distribution prediction of the city's electric vehicle charging load;

[0052] In the fourth part, we combine the local prediction data and the real data to establish the loss function, calculate the reverse error propagation of the non-homogeneous Markov network, and use the gradient descent method to optimize the transfer matrix parameters to improve the model prediction performance.

[0053] In a specific embodiment, we take the main urban area of ​​a city as an example to model the urban spatiotemporal characteristics, establish a non-homogeneous Markov spatiotemporal network, conduct spatiotemporal prediction of electric vehicle charging load at the urban spatial scale, and conduct model parameter optimization experiments. The steps are as follows:

[0054] Step 1: Construct the spatiotemporal characteristics of urban areas and establish a spatiotemporal characteristic model of cities;

[0055] ArcGIS software was used to georeference the land use planning map of the study area and to divide the entire urban area into grids to obtain the distribution of various functional areas in the city in different grids, such as Figure 2 These data are stored in the matrix

[0056] like Figure 3 As shown in the figure, a topological road network is constructed by using the topological connection relationship between road geographic data using an online map website to represent the connection between adjacent grid cells. t The topological road network used to represent the entire simulation area is defined as follows:

[0057]

[0058] Where x represents the horizontal coordinate of the grid position; y represents the vertical coordinate of the grid position; D c (z),z∈[1,8] represents the connection direction of the grid position, D c (z) (z=1, 2, ..., 8) represent the index of the connection direction (east, west, south, north, southeast, southwest, northeast, northwest).

[0059] Step 2: Using the urban spatiotemporal characteristics and based on the transfer value function, a non-homogeneous Markov spatiotemporal state transfer matrix is ​​constructed to form a non-homogeneous Markov spatiotemporal network and establish the urban spatiotemporal transfer feature connection;

[0060] Taking space as the research unit, consider the number of electric vehicles p in a space G(i), i = 0, 1, ..., L at a certain time t. i (t) and the probability distribution of electric vehicles going to other spaces

[0061] The following definition is given: At any time and its corresponding next time t, t+1∈T, the number of electric vehicles P(t) and P(t+1) in the space is defined as follows:

[0062] P T (t)=(p1(t),p2(t),…,p L (t))

[0063] P T (t+1)=(p1(t+1),p2(t+1),…,p L (t+1))

[0064] According to the properties of non-homogeneous Markov chains, we have:

[0065]

[0066] At time t, the probability of an electric car in space i transferring to space j is The transfer value of space j to space i The decisions are:

[0067]

[0068] The value function The distribution of travel itineraries for different travel purposes at time t and the travel distance between space i and space j are related to:

[0069]

[0070] Among them, h i ,w i ,c i represents the h, w, c attributes of space i, h j ,w j ,c j represents the h, w, c attributes of space j, I t Represents the probability of different travel purposes corresponding to time t.

[0071] Step 3: Based on the Monte Carlo simulation process of random energy particles, establish an electric vehicle charging behavior model and output the spatiotemporal distribution prediction of the city's electric vehicle charging load;

[0072] The random process particles simulate the behavior of electric vehicles based on the following three factors: (1) the total maximum battery capacity of the electric vehicle cluster in the study area; (2) the total battery energy consumed by the electric vehicle cluster at time t; (3) the total energy supplemented by the electric vehicle cluster from the power grid at time t. The random process energy particle is defined as

[0073] e={τ,γ,υ|τ∈{1,2,…,L},γ∈[0,1],υ∈{0,1}}

[0074] Among them, τ represents the urban grid space where the particle is located, γ represents the energy of the particle, and its value range is [0,1], and υ represents the motion state of the particle. If υ=1, it means that the particle is in motion, and if υ=0, it means that the particle is in a stopped state.

[0075] Step 4: Combine the local prediction data and real data, establish the loss function, calculate the reverse error propagation of the non-homogeneous Markov network, use the gradient descent method to optimize the transfer matrix parameters, and improve the model prediction performance.

[0076] Assuming that there is actual load data y(t) for a space g in the study area, t = 0, 1, ..., n, where n is the number of samples, the output of the model proposed in this paper for this space is n is the number of samples. Taking mean square error (MSE) as the loss function, according to the definition of loss function, we have:

[0077]

[0078] For space g, the parameters to be optimized are h, w, c, In order to minimize the loss function, find the partial derivative of the loss function with respect to these parameters. Taking parameter h as an example, according to the chain derivation rule, we have:

[0079]

[0080] Using the gradient descent method to minimize the loss function, the parameter update process is:

[0081]

[0082] The iterative process of the model is as follows:

[0083] (1) Run the model, with the initial simulation time t = 1;

[0084] (2) Obtain the predicted output of space g And the true value y, calculate the loss function;

[0085] (3) Update the model parameters h, w, c corresponding to the space g,

[0086] (4) Set the simulation time t = t + 1, and repeat steps (1) to (3) until all time periods are completed, that is, one iteration is completed;

[0087] (5) Repeat steps (1) to (4) until the set iteration cycle is completed or the model converges.

[0088] The proof is complete.

[0089] Figure 4 is a non-homogeneous Markov space-time network graph.

[0090] Verification example:

[0091] In order to verify the effectiveness and superiority of the electric vehicle charging load prediction method based on non-homogeneous Markov spatiotemporal network described in the present invention, a verification example taking the main urban area of ​​a certain city as an example is given here.

[0092] The spatial distribution map of the average daily total electric vehicle charging load in the central urban area of ​​the case city output by the prediction model shows that the predicted electric vehicle charging demand in the case city presents a multi-center and multi-peak spatial distribution feature, which is highly consistent with the city’s geographical characteristics and urban layout.

[0093] On the two data sets, the prediction accuracy of the model on the validation set is compared before and after the model is optimized using real data. It can be seen that before the model is optimized using real data, its prediction results generally match the trend of the real load. After optimization using real load data, its prediction results are closer to the real load and have a better fitting effect on the peaks and troughs of the load.

[0094] Table 1 shows the load forecasting accuracy evaluation indicators of the models before and after optimization in two cases. In the validation set of data set 1, the MAE, RMSE, and R2 of the prediction results of the initial model are 44.28, 54.01, and 0.53, respectively. The MAE, RMSE, and R2 of the prediction results after model-driven optimization are 22.52, 30.43, and 0.85, and the performance is improved by 49.14%, 43.66%, and 60.38%; in data set 2, the MAE, RMSE, and R2 of the prediction results of the initial model are 30.13, 34.89, and 0.57, respectively. The MAE, RMSE, and R2 of the prediction results after model-driven optimization are 13.21, 18.26, and 0.88, respectively, and the accuracy is improved by 56.16%, 47.66%, and 54.39%.

[0095] Table 1

[0096] Area Model MAE RMSE R2 Example 1 Initial Model 44.28 54.01 0.53 Optimized model 22.52 30.43 0.85 Example 2 Initial Model 30.13 34.89 0.57 Optimized model 13.21 18.26 0.88

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A method for spatiotemporal prediction of electric vehicle charging load based on non-homogeneous Markov spatiotemporal network, characterized by: The method comprises the following steps: S1: Construct the spatiotemporal characteristics of urban areas, divide urban spatial areas through gridding, and then extract the functional spatial distribution of residential areas, work areas, and commercial areas. Establish the urban road topology based on real roads, and establish the spatiotemporal characteristic model of the city based on the daily travel mode statistics of electric vehicle data; S2: Using the urban spatiotemporal characteristics and based on the transfer value function, a non-homogeneous Markov spatiotemporal state transfer matrix is ​​constructed to form a non-homogeneous Markov spatiotemporal network and establish the urban spatiotemporal transfer feature connection; S3: Based on the Monte Carlo simulation process of random energy particles, an electric vehicle charging behavior model is established to output the spatiotemporal distribution prediction of the city-wide electric vehicle charging load; S4: Combine local prediction data and real data to establish a loss function, calculate the reverse error propagation of the non-homogeneous Markov network, use the gradient descent method to optimize the transfer matrix parameters, and improve the model prediction performance.

2. The method for spatiotemporal prediction of electric vehicle charging load based on non-homogeneous Markov spatiotemporal network according to claim 1 is characterized in that: In S1, the construction of spatiotemporal features of urban areas includes three parts: 1) Modeling the spatial distribution of urban residential areas, work areas, and commercial areas. Using ArcGIS software, we used the land use planning map of the study area for georeferencing and divided the entire urban area into grids to obtain the distribution of various functional areas in the city in different grids. These data were stored in the matrix 2) Urban road network topology, using online map websites to construct a topological road network through the topological connection relationship between road geographic data to represent the connection between adjacent grid units; matrix D t The topological road network used to represent the entire simulation area is defined as follows: Where x represents the horizontal coordinate of the grid position; y represents the vertical coordinate of the grid position; D c (z),z∈[1,8] represents the connection direction of the grid position, D c (z), z = 1, 2, ..., 8 represent the index of the connection direction, including east, west, south, north, southeast, southwest, northeast and northwest; The road topology modeling is performed on N grid cells in the study area, and the Floyd shortest path search algorithm is used to calculate the shortest road distance between any two spaces to obtain the transfer distance matrix between any two spaces. 3) The probability distribution of daily travel purposes is statistically analyzed. The electric vehicle user group has different travel totals at different times in a 24-hour day. According to the different travel purposes, the travel at each time is divided into five types: travel from residential area to work area H2W, travel from residential area to commercial area H2C, travel between work area and commercial area W&C, travel from work area to residential area W2H, and travel from commercial area to residential area C2H.

3. The method for spatiotemporal prediction of electric vehicle charging load based on non-homogeneous Markov spatiotemporal network according to claim 1 is characterized in that: In S2, the city's spatiotemporal characteristics are used to construct a non-homogeneous Markov spatiotemporal state transfer matrix based on the transfer value function, forming a non-homogeneous Markov spatiotemporal network, and establishing a city's spatiotemporal transfer feature connection. The transfer value function is defined as: Among them, h i ,w i ,c i represents the h, w, c attributes of space i, h j ,w j ,c j represents the h, w, c attributes of space j, I t Represents the probability of different travel purposes corresponding to time t; The non-homogeneous Markov transition matrix is ​​defined as:

4. The method for spatiotemporal prediction of electric vehicle charging load based on non-homogeneous Markov spatiotemporal network according to claim 1 is characterized in that: In S3, the random process energy particle is defined as: e={τ,γ,υ|τ∈{1,2,…,L},γ∈[0,1],υ∈{0,1}} Among them, τ represents the urban grid space where the particle is located, γ represents the energy of the particle, and its value range is [0,1], and υ represents the motion state of the particle. If υ=1, it means that the particle is in motion, and if υ=0, it means that the particle is in a stopped state.

5. The method for spatiotemporal prediction of electric vehicle charging load based on non-homogeneous Markov spatiotemporal network according to claim 1 is characterized in that: In S4, the loss function is defined as: The error back propagation formula for the non-homogeneous Markov space-time network is defined as: The gradient descent method is defined as:

Images