Cooperative scheduling and dynamic regulation and control method for electric vehicle charging station based on space-time demand gradient

Abstract

The invention discloses an electric vehicle charging station cooperative scheduling and dynamic regulation and control method based on a space-time demand gradient. The method comprises the following steps: step 1) obtaining space-time characteristics of a charging station; 2) based on historical operation characteristics, classifying the charging stations according to the utilization rate; 3) based on the time-space characteristics of the charging stations, predicting the future short-term coincidence rate of each type of charging stations to obtain the future short-term coincidence rate of the charging stations; if the future short-term coincidence rate of the charging station is greater than a preset threshold value, entering a step 4), otherwise, returning to the step 1) after T time; 4) calculating a space transfer identifier; 5) calculating the total demand transfer amount of each charging station by using a double-pressure module, and 6) solving the MPC-based multi-target rolling optimization model in a rolling manner in each decision period, and generating and executing an electricity price adjustment strategy and a vehicle guiding strategy. The time-space equilibrium distribution of the charging coincidence rate is realized, the congestion is effectively relieved, and the comprehensive benefit of the system is improved.

Description

Technical Field

[0001] This invention relates to the field of charging station scheduling, specifically a method for collaborative scheduling and dynamic control of electric vehicle charging stations based on spatiotemporal demand gradients. Background Technology

[0002] With the rapid popularization of electric vehicles, charging behavior exhibits high uncertainty and clustering in both spatiotemporal dimensions, leading to increasingly prominent problems such as regional power grid overload, uneven distribution of charging resources, and extended user waiting times. Existing research on charging station scheduling mainly focuses on dynamic pricing and multi-site collaborative optimization, but it still has significant limitations. Most methods rely on idealized assumptions and are poorly adaptable to the spatiotemporal fluctuations of charging behavior in real-world scenarios. Traditional models often ignore the "simultaneity rate," a key indicator reflecting the real-time busyness of charging stations, resulting in fragmented spatiotemporal control. Existing optimization methods are prone to getting trapped in local optima under large-scale constraints, making it difficult to balance congestion relief, revenue improvement, and user cost equilibrium. Furthermore, due to the lack of quantitative modeling of the propagation and transfer trends of charging demand between stations, it is difficult to achieve forward-looking, closed-loop collaborative scheduling. Therefore, there is an urgent need for a charging station collaborative scheduling method that can deeply integrate spatiotemporal characteristics, possess high-precision predictive capabilities, and support multi-objective rolling optimization. Summary of the Invention

[0003] The purpose of this invention is to provide a method for coordinated scheduling and dynamic control of electric vehicle charging stations based on spatiotemporal demand gradients, comprising the following steps:

[0004] Step 1) Obtain the spatiotemporal characteristics of the charging station;

[0005] Step 2) Based on historical operational characteristics, classify charging stations according to usage rate;

[0006] Step 3) Based on the spatiotemporal characteristics of charging stations, predict the future short-term simulacability rate of each type of charging station to obtain the future short-term simulacability rate of the charging station; if the future short-term simulacability rate of the charging station is greater than the preset threshold, proceed to step 4); otherwise, return to step 1 after time T.

[0007] Step 4) Calculate spatial transfer markers to quantify and identify the direction and intensity of charging demand flow between stations;

[0008] Step 5) Calculate the total demand transfer amount of each charging station using the dual-pressure module, and use the total demand transfer amount as a constraint condition for the multi-objective rolling optimization model to provide important boundary information and directional guidance for optimization;

[0009] Step 6) Construct a multi-objective rolling optimization model based on MPC; solve the multi-objective rolling optimization model based on MPC in each decision cycle to generate and execute the electricity price adjustment strategy and vehicle guidance strategy.

[0010] Furthermore, the spatiotemporal characteristics of charging stations include the average simultaneity rate of neighboring stations and the spatiotemporal demand gradient;

[0011] The average simultaneous charging rate of neighboring stations is used to quantify the overall busyness of the area surrounding the target charging station, reflecting the regional charging demand pressure.

[0012] The spatiotemporal demand gradient characterizes the difference in busyness between two stations and the spatial distance attenuation effect, reflecting the direction and intensity of demand flow.

[0013] Furthermore, the average concurrency rate of neighboring stations is shown below:

[0014] (1)

[0015] in, Let be the average simultaneous charging rate of neighboring stations of charging station i at time t; Let i be the set of neighboring stations with a radius of dkm centered at charging station i. Number of neighboring stations; Let be the actual simulcast rate of neighboring station j at time t.

[0016] Furthermore, the spatiotemporal demand gradient is shown below:

[0017] (2)

[0018] in, Let be the spatiotemporal demand gradient of charging station i at time t; Number of neighboring stations; The difference in simultaneous operation rates between the two stations; Let i be the spatial distance between charging station i and its neighboring station j; For smoothing coefficients; This is a Gaussian distance decay function, used to measure the weakening effect of distance on the intensity of demand transfer; A positive value indicates that surrounding sites are generally busier than this site, and there is a trend of demand flowing into this site i; A negative value indicates that this station is busier than surrounding stations, and there is a tendency for demand to overflow from this station i to neighboring stations; denoted as the standard deviation of the Gaussian distance decay function.

[0019] Furthermore, in step 2), the charging stations are classified using the K-means clustering algorithm based on historical operating characteristics.

[0020] Historical operational characteristics include average simultaneous rate, number of piles, and geographical location.

[0021] Furthermore, in step 3), the LSTM-GCN hybrid prediction model is used to predict the short-term simultaneous rate of each type of charging station.

[0022] The LSTM-GCN hybrid prediction model takes the spatiotemporal characteristics of charging stations as input and the short-term simultaneity rate of charging stations as output; the spatiotemporal characteristics of charging stations include the average occupancy rate of neighboring stations and the spatiotemporal demand gradient.

[0023] The LSTM-GCN hybrid prediction model is trained using historical data.

[0024] Furthermore, the spatial transfer identifier is shown below:

[0025] (3)

[0026] In the formula, This is a sign function, with a value range of {-1, 0, 1}, used to identify the direction of the transition; A positive value indicates that demand is overflowing from i to j; A negative value indicates that demand flows from j to i; Let i be the spatial distance between charging station i and its neighboring station j; , For charging stations j and i, the simultaneous rate; This is a distance scale parameter used to control the intensity of distance attenuation.

[0027] Furthermore, the dual-pressure module includes a price pressure module and a space pressure module;

[0028] The price pressure module takes the spatiotemporal characteristics and adjacency matrix of charging stations as input, extracts the demand shift potential caused by electricity price differences through graph convolutional layers, and then outputs... Price space embedding matrix;

[0029] The output of the (l+1)th graph convolutional layer of the price pressure module is as follows:

[0030] (4)

[0031] In the formula, This is the normalized adjacency matrix; Price pressure embedding matrix; ; For learnable weight matrix

[0032] The spatial pressure module takes the spatiotemporal characteristics and adjacency matrix of the charging station as input, and extracts the demand shift potential caused by congestion through graph convolutional layers, thereby outputting... Congestion space embedding matrix;

[0033] The output of the (l+1)th graph convolutional layer of the spatial pressure module is shown below:

[0034] (5)

[0035] In the formula, NOP stands for Average Concurrency Rate of Neighboring Stations. This is a learnable weight matrix.

[0036] Furthermore, the dual-pressure module calculates the transfer amount caused by the electricity price difference based on the output of the price pressure module;

[0037] The dual-pressure module calculates the amount of transfer caused by congestion based on the output of the spatial pressure module.

[0038] The dual-pressure module weights the transfer volume caused by electricity price differences and the transfer volume caused by congestion to obtain the total demand transfer volume of the charging station.

[0039] The transfer amounts resulting from electricity price differences are shown below:

[0040] (6)

[0041] In the formula, This is the price elasticity coefficient; It is the price difference of electricity between stations; The average 24-hour electricity price for the entire region; The simultaneity rate of station i at time t; This is the distance attenuation factor;

[0042] The amount of movement caused by congestion is shown below:

[0043] (7)

[0044] In the formula, = indicates the intensity of user scrambling when the neighboring station is busy; The total number of charging piles at the station; For site concurrent users;

[0045] The total demand transfer volume for charging stations is shown below:

[0046] (8)

[0047] In the formula, >0 indicates a net inflow of demand, meaning the station can receive vehicles transferred from neighboring stations; <0 indicates a net outflow of demand, requiring users to be guided to migrate to neighboring sites.

[0048] Furthermore, the multi-objective rolling optimization model based on MPC aims to minimize congestion variance, maximize total revenue, and balance user costs, and incorporates the total demand transfer volume. As constraints, the upper and lower limits of the capacity of the receiving station and the transmitting station, their transfer directions, and the physical boundaries of the transfer amount are determined.

[0049] The objective function of the multi-objective rolling optimization model based on MPC is min As shown below:

[0050] min (9)

[0051] in, Used to quantify the simultaneity rate and the simultaneity rate space balance; Severe congestion penalties: Used to manage oversaturated sites; For spatial imbalance penalty terms: Used to capture neighborhood spillover effects; Total revenue item: ; This is a regularization term used to ensure the constraint on vehicle transfer ratios. Consistency with spatial transfer identifiers; , , , , These are the weighting coefficients; This represents the occupancy rate of all sites. Indicates variance; This represents the real-time occupancy rate of site i. The number of charging piles at site i; The average simultaneous rate of neighboring stations; Total charging service volume for station i; Dynamic charging pricing;

[0052] The constraints of the MPC-based multi-objective rolling optimization model are: .

[0053] The technical effectiveness of this invention is undeniable. By proposing a collaborative scheduling and dynamic control method for electric vehicle charging stations based on spatiotemporal demand gradients, this invention effectively solves the problems of insufficient capture of spatiotemporal dynamic characteristics of charging demand, fragmented spatiotemporal control, and difficulties in multi-objective optimization in existing technologies. By constructing a feature system of average simultaneity rate of neighboring stations and spatiotemporal demand gradients, combined with a hybrid prediction model of K-means clustering and LSTM-GCN, high-precision prediction of short-term simultaneity rate of charging stations is achieved. Furthermore, by designing spatial transfer identifiers and dual-pressure modules, a multi-objective rolling optimization model based on MPC is established, and combined with dynamic pricing and spatial guidance strategies, spatiotemporal balanced allocation of charging simultaneity rate, effective congestion mitigation, and improved overall system benefits are achieved. The overall strategy possesses foresight, closed-loop characteristics, and strong adaptability, significantly improving the operational efficiency and user experience of regional charging networks, and is suitable for intelligent scheduling and refined management of large-scale urban charging facilities. Attached Figure Description

[0054] Figure 1 This is a framework diagram of a charging station dynamic control system.

[0055] Figure 2 Flowchart of the short-time simultaneous rate prediction framework;

[0056] Figure 3 Flowchart of the dynamic control strategy for orderly charging. Detailed Implementation

[0057] The present invention will be further described below with reference to embodiments, but it should not be construed that the scope of the present invention is limited to the following embodiments. Various substitutions and modifications made based on ordinary technical knowledge and common practices in the art without departing from the above-described technical concept of the present invention should be included within the scope of protection of the present invention.

[0058] Example 1:

[0059] See Figures 1 to 3 A collaborative scheduling and dynamic control method for electric vehicle charging stations based on spatiotemporal demand gradients includes the following steps:

[0060] Step 1) Obtain the spatiotemporal characteristics of the charging station;

[0061] Step 2) Based on historical operating characteristics, classify charging stations according to usage rate to enhance the model's ability to characterize the operating patterns of different types of stations, so that subsequent prediction models can extract more representative spatiotemporal features for each type of station.

[0062] Step 3) Based on the spatiotemporal characteristics of charging stations, predict the future short-term simulacability rate of each type of charging station to obtain the future short-term simulacability rate of the charging station; if the future short-term simulacability rate of the charging station is greater than the preset threshold, proceed to step 4); otherwise, return to step 1 after time T.

[0063] Step 4) Calculate spatial transfer markers to quantify and identify the direction and intensity of charging demand flow between stations;

[0064] Step 5) Calculate the total demand transfer amount of each charging station using the dual-pressure module, and use the total demand transfer amount as a constraint condition for the multi-objective rolling optimization model to provide important boundary information and directional guidance for optimization;

[0065] Step 6) Construct a multi-objective rolling optimization model based on MPC; solve the multi-objective rolling optimization model based on MPC in each decision cycle to generate and execute the electricity price adjustment strategy and vehicle guidance strategy.

[0066] Example 2:

[0067] A method for collaborative scheduling and dynamic control of electric vehicle charging stations based on spatiotemporal demand gradient, with the same technical content as in Example 1, further wherein the spatiotemporal characteristics of the charging station include the average simultaneity rate of neighboring stations and the spatiotemporal demand gradient.

[0068] The average simultaneous charging rate of neighboring stations is used to quantify the overall busyness of the area surrounding the target charging station, reflecting the regional charging demand pressure.

[0069] The spatiotemporal demand gradient characterizes the difference in busyness between two stations and the spatial distance attenuation effect, reflecting the direction and intensity of demand flow.

[0070] Example 3:

[0071] A method for collaborative scheduling and dynamic control of electric vehicle charging stations based on spatiotemporal demand gradients, with the same technical content as any one of embodiments 1-2, further wherein the average simultaneity rate of neighboring stations is as follows:

[0072] (1)

[0073] in, Let be the average simultaneous charging rate of neighboring stations of charging station i at time t; Let i be the set of neighboring stations with a radius of dkm centered at charging station i. Number of neighboring stations; Let be the actual simulcast rate of neighboring station j at time t.

[0074] Example 4:

[0075] A method for coordinated scheduling and dynamic control of electric vehicle charging stations based on spatiotemporal demand gradient, with the same technical content as any one of embodiments 1-3, further wherein the spatiotemporal demand gradient is as follows:

[0076] (2)

[0077] in, Let be the spatiotemporal demand gradient of charging station i at time t; Number of neighboring stations; The difference in simultaneous operation rates between the two stations; Let i be the spatial distance between charging station i and its neighboring station j; For smoothing coefficients; This is a Gaussian distance decay function, used to measure the weakening effect of distance on the intensity of demand transfer; A positive value indicates that surrounding sites are generally busier than this site, and there is a trend of demand flowing into this site i; A negative value indicates that this station is busier than surrounding stations, and there is a tendency for demand to overflow from this station i to neighboring stations; denoted as the standard deviation of the Gaussian distance decay function.

[0078] Example 5:

[0079] A method for coordinated scheduling and dynamic control of electric vehicle charging stations based on spatiotemporal demand gradients, with the same technical content as any one of embodiments 1-4, further wherein, in step 2), the charging stations are classified by K-means clustering algorithm based on historical operating characteristics.

[0080] Historical operational characteristics include average simultaneous rate, number of piles, and geographical location.

[0081] Example 6:

[0082] A method for coordinated scheduling and dynamic control of electric vehicle charging stations based on spatiotemporal demand gradient, with the same technical content as any one of embodiments 1-5, further, in step 3), the LSTM-GCN hybrid prediction model is used to predict the short-term (generally 24h) simultaneity rate of each type of charging station in the future.

[0083] The LSTM-GCN hybrid prediction model takes the spatiotemporal characteristics of charging stations as input and the short-term simultaneity rate of charging stations as output; the spatiotemporal characteristics of charging stations include the average occupancy rate of neighboring stations and the spatiotemporal demand gradient.

[0084] The LSTM-GCN hybrid prediction model is trained using historical data.

[0085] Example 7:

[0086] A method for collaborative scheduling and dynamic control of electric vehicle charging stations based on spatiotemporal demand gradients, with the same technical content as any one of embodiments 1-6, further wherein the spatial transfer identifier is as follows:

[0087] (3)

[0088] In the formula, This is a sign function, with a value range of {-1, 0, 1}, used to identify the direction of the transition; A positive value indicates that demand is overflowing from i to j; A negative value indicates that demand flows from j to i; Let i be the spatial distance between charging station i and its neighboring station j; , For charging stations j and i, the simultaneous rate; This is a distance scale parameter used to control the intensity of distance attenuation.

[0089] Example 8:

[0090] A method for coordinated scheduling and dynamic control of electric vehicle charging stations based on spatiotemporal demand gradient, with the same technical content as any one of embodiments 1-7, further wherein the dual pressure module includes a price pressure module and a spatial pressure module;

[0091] The price pressure module takes the spatiotemporal characteristics and adjacency matrix of charging stations as input, extracts the demand shift potential caused by electricity price differences through graph convolutional layers, and then outputs... Price space embedding matrix;

[0092] The output of the (l+1)th graph convolutional layer of the price pressure module is as follows:

[0093] (4)

[0094] In the formula, This is the normalized adjacency matrix; Price pressure embedding matrix; ; For learnable weight matrix

[0095] The spatial pressure module takes the spatiotemporal characteristics and adjacency matrix of the charging station as input, and extracts the demand shift potential caused by congestion through graph convolutional layers, thereby outputting... Congestion space embedding matrix;

[0096] The output of the (l+1)th graph convolutional layer of the spatial pressure module is shown below:

[0097] (5)

[0098] In the formula, NOP stands for Average Concurrency Rate of Neighboring Stations. This is a learnable weight matrix.

[0099] Example 9:

[0100] A method for coordinated scheduling and dynamic control of electric vehicle charging stations based on spatiotemporal demand gradient, with the same technical content as any one of embodiments 1-8, further wherein the dual pressure module calculates the transfer amount caused by the price difference based on the output of the price pressure module;

[0101] The dual-pressure module calculates the amount of transfer caused by congestion based on the output of the spatial pressure module.

[0102] The dual-pressure module weights the transfer volume caused by electricity price differences and the transfer volume caused by congestion to obtain the total demand transfer volume of the charging station.

[0103] The transfer amounts resulting from electricity price differences are shown below:

[0104] (6)

[0105] In the formula, This is the price elasticity coefficient; It is the price difference of electricity between stations; The average 24-hour electricity price for the entire region; The simultaneity rate of station i at time t; This is the distance attenuation factor;

[0106] The amount of movement caused by congestion is shown below:

[0107] (7)

[0108] In the formula, = indicates the intensity of user scrambling when the neighboring station is busy; The total number of charging piles at the station; For site concurrent users;

[0109] The total demand transfer volume for charging stations is shown below:

[0110] (8)

[0111] In the formula, >0 indicates a net inflow of demand, meaning the station can receive vehicles transferred from neighboring stations; <0 indicates a net outflow of demand, requiring users to be guided to migrate to neighboring sites.

[0112] Example 10:

[0113] A collaborative scheduling and dynamic control method for electric vehicle charging stations based on spatiotemporal demand gradients, with technical content identical to any one of embodiments 1-9, further comprising a multi-objective rolling optimization model based on MPC, with the optimization objectives of minimizing congestion variance, maximizing total revenue, and balancing user costs, and incorporating total demand transfer... As constraints, the upper and lower limits of the capacity of the receiving station and the transmitting station, their transfer directions, and the physical boundaries of the transfer amount are determined.

[0114] The objective function of the multi-objective rolling optimization model based on MPC is min As shown below:

[0115] min (9)

[0116] in, Used to quantify the simultaneity rate and the simultaneity rate space balance; Severe congestion penalties: Used to manage oversaturated sites; For spatial imbalance penalty terms: Used to capture neighborhood spillover effects; Total revenue item: ; This is a regularization term used to ensure the constraint on vehicle transfer ratios. Consistency with spatial transfer identifiers; , , , , These are the weighting coefficients; This represents the occupancy rate of all sites. Indicates variance; This represents the real-time occupancy rate of site i. The number of charging piles at site i; The average simultaneous rate of neighboring stations; Total charging service volume for station i; Dynamic charging pricing;

[0117] The constraints of the MPC-based multi-objective rolling optimization model are: .

[0118] Example 11:

[0119] A method for coordinated scheduling and dynamic control of electric vehicle charging stations based on spatiotemporal demand gradients, with technical content identical to any one of Examples 1-10, and the LSTM-GCN hybrid prediction model structure is as follows:

[0120] The GCN module is used to extract spatial features of charging stations. GCN aggregates neighboring station information through graph convolution operations, enhancing the feature representation of each station. Specifically, a two-layer GCN is used, with the calculation formula for each layer as follows:

[0121] =A+I is the adjacency matrix with added self-loops (A is the original adjacency matrix, and I is the identity matrix). yes The degree matrix, i.e. It is the node feature matrix of the l-th layer. These are the features of the input nodes; σ is the trainable weight matrix of the l-th layer; σ is the activation function (such as ReLU). Each charging station outputs a spatially augmented feature vector through the GCN module.

[0122] LSTM module: Used to process time series data. The spatial augmentation features output from the G module are used as input to the LSTM layer to capture temporal dynamics. The core computation formula of the LSTM unit is as follows:

[0123]

[0124]

[0125]

[0126]

[0127]

[0128]

[0129] in, It is the input at time step t (i.e., the spatial augmentation feature output by GCN).

[0130] It is the hidden state of the previous time step; It represents the cell state at the previous time step;

[0131] , These represent the forget gate, input gate, and output gate, respectively. It is a candidate cell state;

[0132] It represents the cell state at the current time step; It is the hidden state of the current time step. , , , and , , , These are trainable parameters.

[0133] Example 12:

[0134] A method for coordinated scheduling and dynamic control of electric vehicle charging stations based on spatiotemporal demand gradient, with the same technical content as any one of embodiments 1-11, further wherein the LSTM-GCN hybrid prediction model is obtained by training historical data;

[0135] During training, the LSTM-GCN hybrid prediction model uses the mean squared error (MSE) as the loss function, which is calculated as follows:

[0136]

[0137] During model training, the Adam optimizer is used to minimize the loss function and update the model parameters; the parameter update formula for the Adam optimizer is as follows:

[0138] First, calculate the gradient:

[0139]

[0140] in, For model parameters, For parameters The loss function value at that point.

[0141] Then, update the first-order moment estimate and the second-order moment estimate:

[0142]

[0143]

[0144] Next, the first-order moment estimate and second-order moment estimate of the bias correction are calculated:

[0145]

[0146]

[0147] Finally, update the model parameters:

[0148]

[0149] Where t is the time step (number of iterations). For learning rate, This is the exponential decay rate (usually taken as 0.9 and 0.999). It is the numerical stability constant. and Specifically, the first-order moment estimate and the second-order moment estimate of the gradient. and It is the estimate after bias correction.

[0150] Through the above steps, the model parameters are updated until the loss function converges.

[0151] Example 13:

[0152] A collaborative scheduling and dynamic control method for electric vehicle charging stations based on spatiotemporal demand gradients is presented below:

[0153] A spatiotemporal characteristic system for charging stations is constructed. The Neighborhood Occupancy Percentage (NOP) is designed to quantify the overall busyness of the area surrounding the target charging station, reflecting the regional charging demand pressure. Its calculation formula is as follows:

[0154]

[0155] in, Let be the average simultaneous charging rate of neighboring stations of charging station i at time t; This is the set of neighboring stations centered at charging station i with a radius of 5km. Number of neighboring stations; Let be the actual simulcast rate of neighboring station j at time t.

[0156] The design uses a spatio-temporal demand gradient (∇V) to characterize the busy / busy conditions between two stations.

[0157] The degree of difference and the spatial distance attenuation effect accurately describe the direction and intensity of demand flow, and its calculation formula is as follows:

[0158]

[0159] in, Let be the spatiotemporal demand gradient of charging station i at time t; Number of neighboring stations; The difference in simultaneous operation rates between the two stations; Let i be the spatial distance between charging station i and its neighboring station j; For smoothing coefficients; Using a Gaussian distance decay function, this simulates the weakening effect of distance on the intensity of demand transfer.

[0160] This indicates that surrounding stations are generally busier than this station, and there is a trend of demand flowing into this station i;

[0161] The negative sign indicates that this station is busier than the surrounding stations, and there is a tendency for demand to overflow from this station i to neighboring stations.

[0162] ∇V can represent the differences in busyness between stations and the spatial distance attenuation effect, accurately depicting the direction and intensity of demand flow.

[0163] Finally, the input feature matrix of the spatiotemporal feature system is constructed. The input feature matrix enables the organic integration of temporal features and spatial information, and systematically incorporates the dynamic status of neighboring stations into the temporal feature system, ensuring that the prediction results accurately reflect the dynamic changes in regional demand.

[0164] 2) Short-term simultaneous rate prediction based on clustering and hybrid models. The K-means clustering algorithm is used to group all charging stations in the region. Based on their historical operating characteristics (such as average simultaneous rate, number of charging piles, geographical location, etc.), they are divided into three categories: high, medium and low utilization rate, so as to improve the model's ability to characterize the charging behavior pattern in the region and the pertinence of the prediction.

[0165] For each type of charging station, an LSTM-GCN hybrid prediction model is constructed to achieve high-precision prediction of the simulcast rate in the short term (e.g., 24 hours). The prediction process is as follows: Figure 2 As shown.

[0166] 3) Design spatial transfer markers and dual-pressure modules. The spatial flow of charging demand exhibits "directional uncertainty" and "intensity attenuation," making it difficult for traditional control methods to accurately guide it due to a lack of quantitative indicators. Spatial transfer markers are designed to accurately quantify and identify the direction and intensity of charging demand flow between stations.

[0167]

[0168] In the formula, This is a sign function, taking values ​​in the range {-1, 0, 1}, used to identify the direction of the transition; overall The value range is [-1, 1]; an exponential decay function is introduced to weaken the transfer impact of distant sites.

[0169] A positive value indicates that demand is overflowing from i to j;

[0170] A negative value indicates that demand flows from j to i, and the larger the absolute value, the stronger the transfer trend.

[0171] The design includes price pressure and spatial pressure modules, which are used to quantify the driving effects of electricity price differences and congestion on demand shifts, respectively.

[0172] The price pressure module takes the aforementioned feature matrix and adjacency matrix as input and extracts the demand shift potential caused by electricity price differences through a graph convolutional layer. The spatial pressure module introduces NOP features based on the adjacency matrix and extracts the demand shift potential caused by congestion at neighboring stations through a graph convolutional layer. The formula is as follows:

[0173] Price pressure module:

[0174]

[0175] The price pressure module can quantify the driving effect of electricity price differences on demand shifts. Spatial embedding matrix.

[0176] in, This is the price pressure embedding matrix for the (l+1)th layer, i.e., the next layer; This represents the modified linear unit, a common activation function in neural networks. This is the normalized adjacency matrix; Price pressure embedding matrix, for the first layer l=0, ; For learnable weight matrix This module can capture cross-regional demand shifts caused by electricity price fluctuations. When electricity prices rise in a certain region, This will spread the signal of reduced demand to adjacent areas.

[0177] Space pressure module:

[0178]

[0179] The spatial pressure module can quantify the driving effect of congestion on demand shifting. Spatial embedding matrix. Wherein, This is the spatial pressure embedding matrix for the (l+1)th layer, i.e., the next layer; Spatial pressure embedding matrix, for the first layer l=0, NOP stands for Average Concurrency Rate of Neighboring Stations. This is a learnable weight matrix. This module can capture cross-station vehicle transfers triggered by congestion information and guide vehicle flow within the area.

[0180] Based on the output of the dual-pressure module, the demand transfer volume between stations can be calculated in a weighted manner, taking into account the transfer volume driven by price differences and congestion conditions respectively, and introducing a distance decay function to sum up the net transfer volume of each station, thereby realizing the spatial redistribution of demand.

[0181] A multi-objective rolling optimization model based on MPC is established. With "minimizing congestion variance, maximizing total revenue, and balancing user costs" as the core objectives, the following multi-objective optimization function is constructed:

[0182] min

[0183] in, Quantify the simultaneity rate and the spatial balance of simultaneity rate;

[0184] Severe congestion penalties: Used to manage oversaturated sites;

[0185] For spatial imbalance penalty terms: It can capture neighborhood spillover effects;

[0186] Total revenue item: By incorporating a negative sign into the objective function, the profit can be maximized.

[0187] Regularization term: Ensures the proportion of constrained vehicle transfers. Consistency with the Spatial Transfer Facility (STF) ensures that subsequent dynamic pricing is synchronized with the vehicle flow strategy.

[0188] By incorporating a Model Predictive Control (MPC) framework, the optimization problem within a finite time domain is solved on a rolling basis in each decision cycle, dynamically generating electricity price adjustment strategies and vehicle guidance strategies. During the optimization process, the constraints are updated and the solution is recalculated based on the latest simultaneity rate prediction results and real-time system state.

[0189] 5) Dynamic pricing and spatial guidance are implemented in tandem. This step transforms the optimization strategy into a real-time control mechanism that combines "price leverage" and "information guidance" through two channels.

[0190] The dynamic pricing system, acting as an economic lever, consists of three parts: a base time-of-use electricity price, a spatial shift pricing mechanism responding to spatial shift trends, and a congestion premium to mitigate localized congestion. This system adjusts electricity prices at each station in real time based on the STF matrix and simultaneity rate forecasts, guiding demand flow through cost signals.

[0191] The spatial guidance mechanism intervenes directly through information push. Based on the real-time concurrency rate and the status of neighboring stations, the system proactively sends staggered peak reminders and optimal station recommendations to users, thereby achieving precise distribution of demand.

[0192] Both are triggered synchronously and linked in a closed loop under the MPC framework: price signals create macro incentives, and information guidance provides micro paths, jointly shaping users' charging choices and ultimately achieving spatiotemporal equilibrium of simultaneous charging rates and improving system efficiency.

[0193] Example 14:

[0194] A collaborative scheduling and dynamic control method for electric vehicle charging stations based on spatiotemporal demand gradients is presented below:

[0195] 1) Data preprocessing and feature construction: The original charging records are cleaned and preprocessed, including handling outliers, filling in missing data, and normalizing key features to eliminate the influence of units. The actual occupancy rate is calculated based on the number of charging piles (pile_count) at each charging station to eliminate the impact of differences in the number of charging piles on the absolute occupancy value.

[0196]

[0197] Based on this, the average simultaneity rate of neighboring stations is calculated, a set of neighboring stations is constructed with a 5-kilometer radius, and the regional average simultaneity rate level is quantified. A spatiotemporal demand gradient is established to characterize the demand flow trend based on the distance between stations and the difference in simultaneity rate. A spatial transfer indicator is designed to accurately describe the direction and intensity of demand transfer by combining a sign function and the distance decay effect. Finally, a complete multi-dimensional feature matrix is ​​constructed, including latitude and longitude coordinates, number of charging piles, historical charging volume, real-time electricity price, and weather conditions. As shown below:

[0198]

[0199]

[0200]

[0201] :longitude; :latitude; Charging station capacity; Total charging capacity of the charging station; Simultaneity rate; Time-of-use electricity pricing; Average occupancy rate

[0202] 2) Charging station clustering and model training: The K-means clustering algorithm was used to divide all charging stations into three usage groups—high, medium, and low—based on their historical simultaneous usage rate characteristics, geographical distribution, and infrastructure scale. The clustering results showed that stations in core commercial areas were mostly in the high usage rate group, suburban stations were mostly in the medium usage rate group, and stations in remote suburban areas were mostly in the low usage rate group.

[0203] For each type of charging station, a hybrid LSTM-GCN prediction model was constructed and trained. The model input consisted of historical data from the past 96 hours for each type of station, including temporal simultaneity rate, spatial correlation features, and external environmental factors. Through joint optimization, the model could output the predicted simultaneity rate for each station within the next 24 hours. An early stopping strategy was employed during training to prevent overfitting, and the final model demonstrated high prediction accuracy on the test set.

[0204] 3) Dynamic regulation of spatial distribution The module focuses on overloaded sites and... The maximum distance between stations is defined as a radius, establishing a spatial control zone. All stations within this zone are subject to control. Overly busy stations are defined using the following rule: based on the short-term concurrency rate prediction results of the LSTM-GCN model, a trigger threshold is set as the station concurrency rate. If the charging station's usage rate exceeds 70%, it is determined whether the station is in an "overloaded state." When a charging station enters an "overloaded state," the control module is activated, and demand is distributed across stations through physical guidance.

[0205] Next, the transfer volume between stations is calculated based on the dual-pressure module. According to the principle of the dual-pressure module, the transfer volume of the price pressure module and the spatial pressure module are calculated separately. Combined with the weighted summation of the total volume, the transfer results of all stations at a certain moment are obtained, realizing spatial transfer congestion. Specifically:

[0206] Based on the output of the price pressure module, the transfer amount caused by the electricity price difference is calculated using the following formula:

[0207]

[0208] In the formula, = 0.35 is the price elasticity coefficient (the user's sensitivity to price differences); It is the price difference of electricity between stations; The average 24-hour electricity price for the entire region; The concurrency rate of station i at time t; = 0.8 / km is the distance attenuation factor.

[0209] Based on the output of the space pressure module, the amount of transfer caused by congestion is calculated using the following formula:

[0210]

[0211] In the formula, = 0.28 (saturation transfer coefficient) represents the intensity of user transfer when neighboring stations are busy, based on K-menas adaptation (high = 0.35, medium = 0.28, low = 0.2). This ensures that a transfer is triggered only when the neighboring station j is too busy. The amount of data transferred under this constraint shall not exceed the remaining capacity of station i. The total number of charging piles at the station.

[0212] Summarizing the outputs of the above modules, the total demand transfer amount at station i at time t is:

[0213]

[0214] >0, net demand inflow, able to receive vehicles transferred from neighboring stations;

[0215] <0, net demand outflow, users need to be guided to migrate to neighboring sites.

[0216] 4) Dynamic Pricing Control. This invention establishes a composite electricity price model consisting of three parts to accurately guide charging demand. Combining the STF matrix and the prediction results of LSTM-GCN, the electricity price Pi(t) of station i at time t is constructed, consisting of three parts:

[0217]

[0218] The basic time-of-use tariff indicates the time-of-use tariff policy setting for the regional power grid.

[0219] Using the STF matrix as input, the calculation formula is as follows:

[0220]

[0221] This represents the maximum price adjustment. To determine the sensitivity coefficient and control the smoothness of the STF effect, a gradient descent method is introduced for solution. and The goal is to minimize the simultaneous rate imbalance of the overall system and maximize the overall benefit. ,

[0222] like <0 (net demand outflow), tanh output is negative, price reduction attracts users to stay on the site;

[0223] like >0 (net demand inflow), tanh output is positive, price increase diverts users to neighboring stations.

[0224] To address congestion pricing, based on the simultaneous charging station rate predicted by the LSTM-GCN model, a tiered pricing system is used to guide users to charge during off-peak hours, avoiding congestion caused by a concentrated influx of users to a single station during peak periods. The calculation formula is as follows:

[0225]

[0226] Congestion coefficient; Indicates when Regulation is triggered when the rate is >70%, and it increases linearly with the simultaneous rate.

[0227] Spatial transfer pricing uses the STF matrix as input and transforms demand transfer trends into pricing signals through the hyperbolic tangent function. It automatically raises electricity prices when net demand inflows are detected and lowers them when net demand outflows are detected, forming an effective spatial price gradient. The congestion premium is directly linked to the prediction results of the LSTM-GCN model. It is activated immediately when the predicted simultaneity rate exceeds a 70% threshold and linearly increases the premium as the simultaneity rate rises, establishing a price barrier to prevent localized congestion. These two dynamic components, α, β, and γ, control the price adjustment magnitude, response sensitivity, and congestion coefficient, respectively. These parameters are optimized online using gradient descent to ensure that overall revenue is maximized while minimizing system simultaneity rate imbalances, ultimately forming an intelligent pricing system capable of adapting to changes in spatiotemporal simultaneity rates.

[0228] 5) MPC Rolling Optimization and Strategy Generation. In the model predictive control phase, the system employs a 24-hour prediction time domain and a 6-hour optimization time domain, executing a complete rolling optimization process at 1-hour intervals. At the beginning of each cycle, the system uses a pre-trained LSTM-GCN model to predict the simultaneity rate sequence of all charging stations within the next 24 hours based on the latest historical operating data. Subsequently, based on the latest prediction results and real-time monitoring data, the system dynamically updates the spatiotemporal characteristic indicators of each station, including the average simultaneity rate of neighboring stations (NOP), the spatiotemporal demand gradient (∇V), and the spatial transfer identifier (STF) matrix. Based on this, a multi-objective optimization problem is constructed for the future optimization time domain, adjusting the price parameters α, β, and the congestion coefficient γ in real time to respond to changes in simultaneity rate, and determining the core objective function as "minimizing congestion variance, maximizing total revenue, and balancing user costs." By solving this optimization problem, the obtained first-period strategy (including specific electricity prices and recommended transfer schemes for each station) is immediately implemented. Finally, the system rolls the time forward by one period, collects new actual operating data, and repeats the above process, forming a closed-loop control mechanism for continuous online optimization and dynamic adjustment. The overall flowchart of the control strategy is as follows: Figure 3 As shown.

[0229] In summary, this invention proposes an innovative solution for the intelligent scheduling and operation management of electric vehicle charging facilities. By constructing a collaborative scheduling and dynamic control system that integrates spatiotemporal demand gradients, it achieves accurate perception, prediction, and optimized control of the simultaneity rate of regional charging networks. This method innovatively introduces features such as the average simultaneity rate of neighboring stations, spatiotemporal demand gradients, and spatial transfer indicators, combined with a hybrid prediction model of K-means clustering and LSTM-GCN, to establish a high-precision prediction mechanism for the short-term simultaneity rate of charging stations. Furthermore, it designs a demand transfer quantification method based on a dual-pressure module and constructs a multi-objective rolling optimization model within the model predictive control framework, achieving spatiotemporal collaborative optimization of the charging simultaneity rate.

[0230] Traditional charging station scheduling methods often rely on static data analysis or single-dimensional dynamic pricing, resulting in insufficient capture of the spatiotemporal dynamic characteristics of charging demand and difficulty in achieving effective spatiotemporal control. This invention constructs a multi-dimensional spatiotemporal feature system to systematically characterize the spatiotemporal propagation patterns of charging demand; through cluster analysis and hybrid prediction models, it significantly improves the accuracy and adaptability of simultaneous charging station rate prediction in different regions; through the design of spatial transfer markers and dual-pressure modules, it accurately quantifies the flow trend of demand between stations; and finally, through MPC multi-objective rolling optimization, it achieves synergistic optimization of congestion relief, revenue improvement, and user cost balance.

[0231] Furthermore, this invention designs a collaborative execution mechanism for dynamic pricing and spatial guidance. Through the synergistic effect of "price leverage" and "information guidance," it guides user charging behavior through economic signals and provides decision support through precise recommendations, forming a complete closed-loop control system. This mechanism can dynamically adjust the electricity price structure based on real-time concurrent charging rates, including three components: basic time-of-use pricing, spatial transfer pricing, and congestion premium, ensuring the timeliness and effectiveness of the control strategy.

[0232] Furthermore, the method of this invention possesses good scalability and practicality, supporting rapid deployment and application in large-scale urban charging networks. Empirical research in Shenzhen demonstrates that this method can significantly reduce system congestion, while simultaneously increasing peak-hour revenue and improving spatial load balancing, effectively validating its technical advantages in real-world scenarios.

[0233] In summary, through systematic innovations in feature engineering, prediction models, optimization algorithms, and execution mechanisms, this invention not only comprehensively improves the operational efficiency, user experience, and profitability of charging stations, but also provides a complete technical solution for the intelligent scheduling and refined management of urban charging facilities, providing solid technical support for the efficient operation and sustainable development of electric vehicle charging infrastructure.

Claims

1. A method for collaborative scheduling and dynamic regulation of electric vehicle charging stations based on spatiotemporal demand gradient, characterized in that, The method comprises the following steps: Step 1) obtaining the space-time characteristics of the charging station; Step 2) classifying the charging stations according to the utilization rate based on the historical operation characteristics; Step 3) predicting the future short-term simultaneous rate of each type of charging station based on the space-time characteristics of the charging station, and obtaining the future short-term simultaneous rate of the charging station; if the future short-term simultaneous rate of the charging station is greater than a preset threshold, step 4) is entered, otherwise, step 1) is returned after T time; Step 4) calculating the space transfer identifier to quantify and identify the flow direction and trend intensity of the charging demand between stations; Step 5) calculating the total demand transfer amount of each charging station by using a double-pressure module, and taking the total demand transfer amount as a constraint condition of a multi-objective rolling optimization model; Step 6) constructing a multi-objective rolling optimization model based on MPC; in each decision cycle, the multi-objective rolling optimization model based on MPC is solved to generate and execute the electricity price adjustment strategy and the vehicle guidance strategy. 2.The method of claim 1, wherein, The space-time characteristics of the charging station include the average simultaneous rate of the adjacent station and the space-time demand gradient; The average simultaneous rate of the adjacent station is used to quantify the overall busy degree of the surrounding area of the target charging station, and reflects the regional charging demand pressure; The space-time demand gradient describes the difference in busy degree between two stations and the spatial distance decay effect, and reflects the flow direction and intensity of the demand. 3.The method of claim 2, wherein, The average simultaneous rate of the adjacent station is as follows: （1） wherein, is the average simultaneous rate of neighboring stations for charging station i at time t; is the set of neighboring stations centered at charging station i with radius dkm, is the number of neighboring stations; is the actual simultaneous rate of neighboring station j at time t. 4.The method of claim 2, wherein, The space-time demand gradient is as follows: （2） where, is the space-time demand gradient of charging station i at time t; is the number of neighboring stations; is the difference in simultaneous rate between two stations; is the spatial distance between charging station i and neighboring station j; is the smoothing coefficient; is the Gaussian distance decay function, which measures the weakening effect of distance on the intensity of demand transfer; is positive, indicating that the overall surrounding stations are busier than the current station, and there is a tendency for demand to flow into the current station i; is negative, indicating that the current station is busier than the surrounding stations, and there is a tendency for demand to overflow from the current station i to neighboring stations; is the standard deviation of the Gaussian distance decay function.

5. The method of claim 1, wherein, In step 2), the charging stations are classified based on the historical operation characteristics by using a K-means clustering algorithm; The historical operation characteristics include the average simultaneous rate, the number of piles, and the geographical location. 6.The method of claim 1, wherein, In step 3), the future short-term simultaneous rate of each type of charging station is predicted by using an LSTM-GCN hybrid prediction model; The LSTM-GCN hybrid prediction model takes the space-time characteristics of the charging station as input and takes the short-term simultaneous rate of the charging station as output; the space-time characteristics of the charging station include the average simultaneous rate of the adjacent station and the space-time demand gradient; The LSTM-GCN hybrid prediction model is obtained by training historical data. 7.The method of claim 1, wherein, The space transfer identifier is as follows: （3） where, is a sign function with value range {-1, 0, 1} to identify the direction of transition; is positive to indicate the demand overflow from i to j; is negative to indicate the demand inflow from j to i; is the spatial distance between charging station i and neighbor station j; , is the simultaneous rate of charging station j, i; is the distance scale parameter to control the strength of distance decay. 8.The method of claim 1, wherein, The double-pressure module includes a price pressure module and a space pressure module; The price pressure module takes the charging station space-time characteristics and the charging station adjacency relationship matrix as input, extracts the demand transfer potential caused by price difference through a graph convolution layer, and outputs a price space embedding matrix; The output of the (l+1)th layer graph convolution layer of the price pressure module is as follows: （4） wherein is a normalized adjacency matrix; is a price pressure embedding matrix; ; is a learnable weight matrix The space pressure module takes the space-time characteristics of the charging station and the adjacency matrix as input, extracts the demand transfer potential caused by congestion state through a graph convolution layer, and outputs a congestion space embedding matrix; The output of the (l+1)th layer graph convolution layer of the space pressure module is as follows: （5） wherein ; NOP is the neighbor average simultaneous rate; is a learnable weight matrix. 9.The method of claim 1, wherein, The double-pressure module calculates the transfer amount caused by the price difference according to the output of the price pressure module; The double-pressure module calculates the transfer amount caused by congestion according to the output of the space pressure module; The double-pressure module weights the transfer amount caused by the price difference and the transfer amount caused by congestion to obtain the total demand transfer amount of the charging station; The transfer amount caused by the price difference is as follows: （6） wherein is the price elasticity coefficient; is the price difference between sites; is the 24h average price of the whole area; represents the simultaneity rate of site i at time t; is the distance decay factor; The transfer amount caused by congestion is as follows: （7） wherein represents the intensity of user diversion when the neighboring station is busy; is the total number of charging piles for station i; is the site simultaneous rate; The total demand transfer amount of the charging station is as follows: （8） In the formula, > 0 indicates a net inflow of demand, vehicles can be accepted from neighboring stations; < 0 indicates a net outflow of demand, users need to be guided to neighboring stations. 10.The method of claim 1, wherein, The MPC-based multi-objective rolling optimization model takes the minimum congestion variance, the maximum total revenue, and the balanced user cost as optimization objectives, and transfers the total demand As constraint conditions, the upper and lower limits of capacities of the receiving station and the sending station, their transfer directions, and physical boundaries of the transfer amount are determined. Objective function of the MPC-based multi-objective rolling optimization model min As follows: min (9) wherein, for quantifying the simultaneity rate and the spatial balance of simultaneity rate; a severe congestion penalty term: for managing and controlling the oversaturation sites; a spatial imbalance penalty term: for capturing the neighborhood spillover effect; a total revenue term: ; a regularization term for ensuring the consistency of the constraint vehicle transfer proportion with the spatial transfer identifier; , , , , a weight coefficient; an occupancy rate of all sites; denotes a variance; a real-time occupancy rate of site i; a number of charging piles of site i; an average simultaneity rate of neighboring sites; a total charging service amount of site i; a dynamic charging price; The constraint conditions of the MPC-based multi-objective rolling optimization model are: .

Images