Intelligent scheduling method for electric vehicle charging station

Abstract

The embodiment of the invention provides an intelligent scheduling method for an electric vehicle charging station, and belongs to the technical field of data processing, and the method specifically comprises the steps: constructing a multi-dimensional time sequence feature set, and generating a training set with a scheduling label; training a deep learning model based on a gating circulation unit by using the training set, and performing rolling prediction on the arrival rate and the service rate of the vehicle by using the trained deep learning model; constructing a multi-state Markov queuing model based on scheduling control input, calculating steady-state distribution and calculating performance indexes according to the steady-state distribution; constructing a charging port scheduling optimization model based on a model prediction control method and the performance indexes, constructing an objective function, and outputting a rolling optimization strategy according to the objective function; and executing a rolling optimization strategy, adjusting the number of the AC / DC charging piles, generating indication information, and feeding back an operation result to the charging port scheduling optimization model to realize closed-loop correction. According to the scheme, the resource utilization rate is improved, the user waiting time is shortened, and the system stability is enhanced.

Description

Technical Field

[0001] This disclosure relates to the field of data processing technology, and in particular to an intelligent scheduling method for electric vehicle charging stations. Background Technology

[0002] Currently, with the rapid development of the new energy vehicle industry and the in-depth implementation of the "dual-carbon" strategy, the proportion of electric vehicles (EVs) in the transportation system continues to rise. As supporting infrastructure, EV charging stations are moving from large-scale deployment to intelligent operation and management. Especially in areas with high traffic flow such as highway service areas and urban transportation hubs, charging stations not only face the challenge of a large number of vehicles gathering in a short period of time during peak hours, but also need to meet users' demands for a fast, convenient, and low-wait service experience.

[0003] Traditional charging station scheduling strategies often adopt fixed or semi-static resource allocation methods, which are difficult to adapt to dynamic and changing actual scenarios. The main problems are: (1) Traffic flow is highly time-varying and demand is unpredictable: Traffic flow in areas such as highways and urban main roads shows significant peak and valley fluctuations and holiday peak characteristics. Charging stations lack effective response mechanisms when facing sudden high demand, which can easily lead to queuing delays and user loss. (2) User behavior is complex and varies significantly: Users show strong irrational behavior characteristics in their decision-making on whether to queue or accept the allocation of charging ports. They are affected by various factors such as waiting time, charging power, and remaining power, and traditional methods are difficult to accurately model. (3) Charging resource allocation is rigid and inefficient: Existing scheduling schemes often formulate static power allocation strategies based on historical average traffic flow, which lacks a dynamic feedback adjustment mechanism for actual operating status and prediction results, resulting in low resource utilization efficiency and even non-optimal operating status with both energy redundancy and port congestion. (4) Lack of intelligent closed loop throughout the process: Most current systems have failed to form a full-process adaptive mechanism of "traffic flow prediction - queue modeling - scheduling optimization - real-time execution - feedback correction", resulting in delayed scheduling decisions and failure to achieve the unity of maximizing operational revenue and optimizing user satisfaction.

[0004] While existing research has attempted to introduce artificial intelligence technology to optimize local aspects, such as traffic prediction based on big data and service management based on game theory, these studies are mostly limited to a single aspect and lack the integration of joint modeling of the overall state of the charging station and intelligent control strategies, making it difficult to achieve overall system optimization.

[0005] It is evident that there is an urgent need for an intelligent scheduling method for electric vehicle charging stations that can improve resource utilization, shorten user waiting time, and enhance system stability. Summary of the Invention

[0006] In view of this, the present disclosure provides an intelligent scheduling method for electric vehicle charging stations, which at least partially solves the problems of poor resource utilization and system stability in the prior art.

[0007] This disclosure provides an intelligent scheduling method for electric vehicle charging stations, including:

[0008] Step 1: Obtain multi-source data of charging stations in highway service areas, construct a multi-dimensional time-series feature set, and generate a training set with scheduling labels;

[0009] Step 2: Train a deep learning model based on gated recurrent units using the training set, and use the trained deep learning model to make rolling predictions on vehicle arrival rate and service rate to form scheduling control input.

[0010] Step 3: Construct a multi-state Markov queuing model based on the scheduling control input, considering the limitation of the number of charging piles and the user waiting characteristics, calculate the steady-state distribution and calculate the performance index accordingly.

[0011] Step 4: Construct a charging port scheduling optimization model based on model predictive control methods and performance indicators to improve resource utilization and service quality. Construct an objective function and output a rolling optimization strategy for dynamic configuration of AC / DC charging piles accordingly.

[0012] Step 5: Execute the rolling optimization strategy, adjust the number of AC / DC charging piles and generate indication information, and feed the running results back to the charging port scheduling optimization model to achieve closed-loop correction.

[0013] According to a specific implementation of an embodiment of this disclosure, step 1 specifically includes:

[0014] Step 1.1: Acquire historical traffic flow data, charging behavior data, environmental data, and power grid basic data of highway service area charging stations through sensors. Among them, historical traffic flow data includes vehicle arrival time, vehicle type, and whether it is a new energy vehicle collected by service area entrance cameras and ETC system; charging behavior data includes charging type, charging duration, initial SOC, and end SOC recorded by charging pile sensors; environmental data includes rainfall and visibility collected by weather stations; traffic platform obtains congestion index and traffic accident information of three upstream service areas; and power grid basic data includes peak / flat / valley time period division of the power system's time-of-use pricing.

[0015] Step 1.2: Fill in the missing values ​​for traffic flow and SOC using the average of adjacent time periods. The criteria identify and remove outliers, divide a day into 6 time periods based on peak / flat / valley data in the power grid basic data, construct time period labels, calculate the lag correlation between the upstream service area congestion index and the arrival rate of this service area, and construct congestion correlation features.

[0016] Step 1.3: Extract the time-series features corresponding to historical traffic flow data, charging behavior data, environmental data, and power grid basic data, and combine them with congestion-related features and time period labels to form a training set.

[0017] According to a specific implementation of an embodiment of this disclosure, step 2 specifically includes:

[0018] Step 2.1: Input the training set into the deep learning model based on gated recurrent units using the sliding window method, and calculate the loss function based on the predicted and true values. ; in, For predicted values, For the true value, For the current time, To predict the step size, For time period labels, The weighting coefficients for time period l are: For square norm operations;

[0019] Step 2.2: Set the learning rate and stopping conditions, and train the deep learning model using the Adam optimizer;

[0020] Step 2.3: Use the trained deep learning model to make rolling predictions of the AC arrival rate, DC arrival rate, AC service rate, and DC service rate of vehicles to form the scheduling control input.

[0021] According to a specific implementation of an embodiment of this disclosure, step 3 specifically includes:

[0022] Step 3.1: The Markov queuing model divides a day into multiple environmental states, defines the system state vector, and sets the state transition rules.

[0023] Step 3.2, define the equations that satisfy the global equilibrium. Through iterative convergence Steady-state distribution is obtained ;

[0024] Step 3.3: Calculate performance metrics based on the steady-state distribution. These metrics include average waiting time, throughput, and user churn rate. The expression for average waiting time is: ; ; in, The total number of vehicles waiting to be charged. Busy with AC charging stations Busy with DC charging stations This is the time period label corresponding to the current environmental state. The system is in a state The steady-state probability distribution, The arrival rate of AC vehicles within time period l. The arrival rate of DC vehicles within time period l;

[0025] The expression for throughput is: ; ; in, The service rate of AC charging piles within time period l. The service rate of DC charging piles within time period l;

[0026] The expression for user departure rate is: ; in, These are the fitting parameters.

[0027] According to a specific implementation of an embodiment of this disclosure, step 4 specifically includes:

[0028] Step 4.1: Construct the objective function and constraints corresponding to the charging port scheduling optimization model based on the model predictive control method and performance indicators;

[0029] Step 4.2: Based on the objective function and constraints, perform rolling optimization to obtain the rolling optimization strategy for the dynamic configuration of AC / DC charging piles.

[0030] According to a specific implementation of this disclosure, the expression of the objective function is as follows: ; in, For time period Weighting coefficients;

[0031] The constraints include the total number of charging piles, waiting time, and system stability. The expression for the total number of charging piles constraint is as follows: ; in, , Indicates the total number of charging stations;

[0032] The expression for the waiting time constraint is: ; in, For time period Maximum tolerable waiting time;

[0033] The expression for the system stability constraint is: ; ; in, The arrival rate of AC vehicles within time period l. The arrival rate of DC vehicles within time period l, and These represent the single pile power of AC and DC respectively during time period l. This represents the upper limit of the power grid load during time period l.

[0034] According to a specific implementation of an embodiment of this disclosure, step 4.2 specifically includes:

[0035] Step 4.2.1, data collection Real-time status at any moment ;

[0036] Step 4.2.2: Call the deep learning model for prediction. ;

[0037] Step 4.2.3: Substitute the prediction results into the multi-state Markov queuing model to calculate the prediction time domain. ;

[0038] Step 4.2.4: Solve the MPC optimization problem to obtain the optimal strategy for the current cycle. ;

[0039] Step 4.2.5: Execute the optimal strategy for the current cycle, and repeat steps 4.2.1 to 4.2.4 within a preset time period to obtain the rolling optimization strategy for the dynamic configuration of AC / DC charging piles.

[0040] According to a specific implementation of an embodiment of this disclosure, step 5 specifically includes:

[0041] Step 5.1: Implement the rolling optimization strategy and adjust the number of AC / DC piles through the control system. , and, when At the same time, push notifications will be sent with waiting time information and charging station selection suggestions;

[0042] Step 5.2: Calculate the deviation between the predicted value and the actual value, and correct the predicted value for the next period accordingly.

[0043] The intelligent scheduling scheme for electric vehicle charging stations in this embodiment includes: Step 1, acquiring multi-source data of charging stations in highway service areas, constructing a multi-dimensional time-series feature set, and generating a training set with scheduling labels; Step 2, using the training set to train a deep learning model based on gated recurrent units, and using the trained deep learning model to perform rolling predictions on vehicle arrival rate and service rate, forming a scheduling control input; Step 3, constructing a multi-state Markov queuing model based on the scheduling control input, considering the limitation of the number of charging piles and user waiting characteristics, calculating the steady-state distribution, and calculating performance indicators accordingly; Step 4, constructing a charging port scheduling optimization model based on the model predictive control method and performance indicators, constructing an objective function to improve resource utilization and service quality, and outputting a rolling optimization strategy for dynamic configuration of AC / DC charging piles; Step 5, executing the rolling optimization strategy, adjusting the number of AC / DC charging piles and generating indication information, while feeding back the running results to the charging port scheduling optimization model to achieve closed-loop correction.

[0044] The beneficial effects of this disclosure are as follows: By collecting historical vehicle flow data, charging behavior data, and environmental data from charging stations, a multi-dimensional time-series feature set is constructed, and a training dataset with scheduling labels is generated. A deep learning model based on gated recurrent units (GRUs) is used to perform rolling predictions of key variables such as arrival rate and service rate for different types of vehicles to form scheduling control inputs. A multi-state queuing model is established based on Markov arrival process (MAP). Combining the prediction results, queuing performance indicators such as average waiting time, throughput, and user departure rate for future periods are calculated in real time. Then, a charging port scheduling optimization model is constructed based on model predictive control (MPC). Considering constraints such as the total number of charging piles and the upper limit of waiting time, the AC / DC converter is solved with the goal of improving resource utilization and service quality. The rolling optimization strategy for dynamic configuration of charging piles is implemented in real time based on the current system status and predictive feedback. This strategy adjusts the configuration of charging pile types and generates indication information. At the same time, the actual operating results are fed back to the model to achieve predictive correction and adaptive update of the strategy, forming a complete closed-loop control of "prediction-modeling-optimization-execution-feedback". By integrating multiple technologies, dynamic and coordinated scheduling of charging resources is achieved, which effectively improves the system's adaptability to complex traffic environments, reduces user waiting time, and improves resource utilization efficiency. Attached Figure Description

[0045] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1A flowchart illustrating an intelligent scheduling method for electric vehicle charging stations provided in this embodiment of the present disclosure;

[0047] Figure 2 A system architecture diagram of an intelligent scheduling method for electric vehicle charging stations provided in this disclosure embodiment;

[0048] Figure 3 A data layer flowchart provided for embodiments of this disclosure;

[0049] Figure 4 A prediction layer flowchart provided for embodiments of this disclosure;

[0050] Figure 5 A flowchart of the parsing layer is provided for an embodiment of this disclosure;

[0051] Figure 6 An optimization layer flowchart provided for embodiments of this disclosure;

[0052] Figure 7 This is an execution layer flowchart provided for an embodiment of the present disclosure. Detailed Implementation

[0053] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0054] The following specific examples illustrate the implementation of this disclosure. Those skilled in the art can easily understand other advantages and effects of this disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. This disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this disclosure. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0055] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this disclosure, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using structures and / or functionalities other than one or more of the aspects set forth herein.

[0056] It should also be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this disclosure. The illustrations only show the components related to this disclosure and are not drawn according to the number, shape and size of the components in actual implementation. In actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0057] Furthermore, specific details are provided in the following description to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the described aspects can be practiced without these specific details.

[0058] This disclosure provides an intelligent scheduling method for electric vehicle charging stations, which can be applied to the vehicle charging scheduling process in traffic management scenarios.

[0059] See Figure 1 This is a flowchart illustrating an intelligent scheduling method for electric vehicle charging stations provided in an embodiment of this disclosure. Figure 1 As shown, the method mainly includes the following steps:

[0060] Step 1: Obtain multi-source data of charging stations in highway service areas, construct a multi-dimensional time-series feature set, and generate a training set with scheduling labels;

[0061] In specific implementation, such as Figure 2 The diagram shown is a system architecture diagram corresponding to the method disclosed herein. Figure 3 The diagram shown illustrates the data layer operation flowchart, which covers multi-source data acquisition and preprocessing, including the following steps:

[0062] A. The following data will be collected in real time through sensors and an information platform: Traffic flow data: Service area entrance cameras and ETC systems will collect vehicle arrival time, vehicle type (private car / truck), and whether it is a new energy vehicle, with a sampling frequency of 1 time / minute; Charging behavior data: Charging pile sensors will record charging type (AC / DC), charging duration, initial SOC, and end SOC, with a sampling frequency of 10 times / second; Environmental data: Weather stations will collect rainfall and visibility, and the traffic platform will obtain congestion index (0-10) and traffic accident information from three upstream service areas; Power grid basic data: Peak / flat / valley time period division of the power system's time-of-use electricity price (updated once / hour).

[0063] B. Data preprocessing, including:

[0064] (1) The missing values ​​of traffic flow and SOC are filled using the average value of adjacent time periods. The formula is as follows: ,in, For missing values, , These are valid values ​​for adjacent time periods.

[0065] (2) Through The criteria identify and remove outliers;

[0066] (3) Construct "time period labels": Divide one day into 6 time periods;

[0067] (4) Construct “congestion correlation characteristics”: Calculate the lag correlation between the congestion index of the upstream service area and the arrival rate of the service area (lag time 15 minutes).

[0068] Step 2: Train a deep learning model based on gated recurrent units using the training set, and use the trained deep learning model to make rolling predictions on vehicle arrival rate and service rate to form scheduling control input;

[0069] In practical implementation, to predict arrival rate and service rate based on the GRU model, such as... Figure 4 The diagram shown illustrates the operation flowchart of the prediction layer, which includes the following steps:

[0070] A. Model input and structural design, including:

[0071] (1) Input features (pure time-series features): Dimensions ,in For the sample size, (Historical window length, i.e., the past 6 hours, with 1 time step every 30 minutes). (8-dimensional features, including historical arrival rate, historical service rate, time tags, etc.);

[0072] (2) GRU model structure: 2 hidden layers (64 and 32 hidden units respectively), historical information is controlled by "reset gate" and "update gate", and the reset gate formula is: ; in, To reset the gate output, , For weights and biases, The state was hidden in the previous moment. For the current input, It is the sigmoid activation function.

[0073] B. Model training and optimization, including:

[0074] (1) The sliding window method is adopted, and each sample contains "12 historical inputs + 2 future outputs";

[0075] (2) Through the formula Normalize the input features to interval

[0076] (3) Weighted mean square error (WMSE) is used, with the weights for peak hours (6:00-10:00, 17:00-20:00) as follows: Off-peak hours The loss function is: ; in, For predicted values, For the true value, For the current time, To predict the step size, For time period labels, The weighting coefficients for time period l are: This refers to square norm operations.

[0077] (4) Adam optimizer (initial learning rate 5e-1, decays by 50% every 20 rounds), early stopping mechanism (stops if the validation set loss does not decrease for 5 consecutive rounds).

[0078] C. Prediction Results Output. The model outputs the prediction results for the next 30 minutes. and 60 minutes The parameters include:

[0079] AC arrival rate ; DC arrival rate ; AC service rate ; DC service rate .

[0080] Step 3: Construct a multi-state Markov queuing model based on the scheduling control input, considering the limitation of the number of charging piles and the user waiting characteristics, calculate the steady-state distribution and calculate the performance index accordingly.

[0081] In practical implementation, to construct a multi-state Markov queuing model, such as Figure 5 The diagram shown is a flowchart of the parsing layer operation, which includes the following steps:

[0082] A. Definitions of environmental and system states, including:

[0083] (1) Divided into 6 time periods according to peak / flat / valley electricity prices. The duration follows the parameter. The exponential distribution;

[0084] (2) System state vector: ,in: Total number of vehicles waiting to charge ; Busy with AC charging stations ; Busy with DC charging stations ; Current environmental status .

[0085] B. Queued vehicle state transition rules, including: (1) AC vehicle arrives Transfer rate ; (2) DC vehicle arrives Transfer rate ; (3) All charging stations are busy Transfer rate ; (4) AC service completed Transfer rate ; (5) DC service completed Transfer rate ; (6) Environmental state transition Transfer rate .

[0086] C. Steady-state distribution and performance index calculation, including:

[0087] (1) Solving for steady-state probability: satisfying the global equilibrium equation Through iterative convergence get .

[0088] (2) Performance indicators:

[0089] Average waiting time: ; ; Throughput: ; ; User churn rate: .

[0090] Step 4: Construct a charging port scheduling optimization model based on model predictive control methods and performance indicators to improve resource utilization and service quality. Construct an objective function and output a rolling optimization strategy for dynamic configuration of AC / DC charging piles accordingly.

[0091] In practical implementation, a scheduling strategy is generated for MPC optimization, such as... Figure 6 The diagram shown is a flowchart of the optimization layer's operation, which includes the following steps:

[0092] A. Definition of MPC parameters and optimization variables, including:

[0093] (1) Prediction time domain: 2 time steps (60 minutes),

[0094] (2) Control time domain: 1 time step (30 minutes);

[0095] (3) Optimize variables: ,in Number of AC-to-DC charging stations , (Initial AC pile count)

[0096] B. Objective function and constraints, including:

[0097] (1) The objective function aims to maximize resource utilization and service quality. ; in, is the weighting coefficient for time period l.

[0098] (2) Constraints include the total number of charging piles: (Total number of piles), of which Waiting time constraint: ( For time period Maximum tolerable waiting time); stability constraints: ; (Power grid load constraints).

[0099] (3) The rolling optimization process includes: (a) Real-time status is collected continuously. (b) Calling the GRU model for prediction (c) Substitute into the queuing model to calculate the prediction time domain. (d) Solve the MPC optimization problem to obtain the optimal solution. (e) Execution and in Repeat steps (a) through (d) every few minutes.

[0100] Step 5: Execute the rolling optimization strategy, adjust the number of AC / DC charging piles and generate indication information, and feed the running results back to the charging port scheduling optimization model to achieve closed-loop correction.

[0101] In practice, to execute optimization strategies and establish feedback correction mechanisms, such as... Figure 7 The diagram shown is a flowchart of the execution layer, which includes the following steps:

[0102] Step 5.1 Implementation of the optimization strategy, including adjusting the number of AC / DC piles through the control system. ; and when At that time, a notification will be pushed to the charging pile screen and navigation APP indicating "shorter waiting time for DC charging pile".

[0103] Step 5.2 is the feedback correction mechanism. The first step is prediction error correction, which is achieved by calculating the deviation. ,like The predicted value for the next period will be revised to Meanwhile, considering contingency plans, if emergency mode is triggered, [the system will...] The upper limit was increased by 1 and the MPC prediction time domain was shortened to 30 minutes, improving response speed;

[0104] Compared with the prior art, the advantages of the method disclosed herein are as follows:

[0105] First, the method disclosed in this paper fully utilizes the technical characteristics of deep learning and dynamic modeling to construct a collaborative mechanism of "GRU prediction-Markov queuing analysis" to achieve precise control over the dynamic state of charging stations. This is because the GRU model efficiently captures the temporal patterns of traffic flow (such as periodicity and trend continuity during peak hours) through a gating mechanism, making it suitable for rolling predictions of AC / DC vehicle arrival rates and service rates, providing forward-looking data for scheduling. The Markov queuing model, on the other hand, can depict the dynamic transitions of system states in real time (such as vehicle arrival, service completion, and environmental changes), accurately outputting key indicators such as average waiting time and throughput. The combination of these two methods not only solves the problem of insufficient adaptation of traditional static prediction to time-varying traffic flow but also overcomes the limitation of idealized queuing assumptions being disconnected from real-world scenarios, significantly improving the system's ability to perceive and analyze dynamic traffic flow.

[0106] Second, the method disclosed herein employs Model Predictive Control (MPC) to achieve dynamic collaborative optimization of resources, controlling computational overhead while ensuring scheduling accuracy. On one hand, MPC uses AC / DC charging pile configuration as the core optimization variable, combining prediction results with real-time queuing status for rolling optimization, focusing only on decision execution during the current control period, significantly reducing computational complexity compared to global optimization across all time periods. On the other hand, compared to traditional static resource allocation strategies (such as fixed charging pile type allocation), MPC can dynamically adjust resource allocation according to traffic flow changes, achieving a balance between service quality and resource utilization, significantly improving intelligence and adaptability.

[0107] Third, the method disclosed herein constructs a complete closed-loop control system of "prediction-modeling-optimization-execution-feedback," providing a systematic solution for charging station scheduling in complex scenarios. This system can not only collaboratively solve single problems such as traffic flow prediction, queue modeling, and dynamic resource allocation, but also continuously iterate and optimize through feedback correction mechanisms (such as prediction error adjustment and emergency response to sudden scenarios), exhibiting good scalability. For example, when more levels of scheduling needs are required (such as deeper optimization combined with grid load), corresponding modules can be added within the existing closed-loop framework without reconstructing the entire system, thus broadening its applicability.

[0108] The intelligent scheduling method for electric vehicle charging stations provided in this embodiment collects historical vehicle flow data, charging behavior data, and environmental data from charging stations to construct a multi-dimensional time-series feature set and generate a training dataset with scheduling labels. It then uses a deep learning model based on gated recurrent units (GRUs) to perform rolling predictions of key variables such as arrival rate and service rate for different types of vehicles to form scheduling control inputs. A multi-state queuing model is established based on Markov arrival process (MAP), and the prediction results are combined to calculate queuing performance indicators such as average waiting time, throughput, and user departure rate for future periods in real time. Finally, a charging port scheduling optimization model is constructed based on model predictive control (MPC). Considering constraints such as the total number of charging piles and the upper limit of waiting time, the AC / DC converter is solved with the goal of improving resource utilization and service quality. The rolling optimization strategy for dynamic configuration of charging piles is implemented in real time based on the current system status and predictive feedback. This strategy adjusts the configuration of charging pile types and generates indication information. At the same time, the actual operating results are fed back to the model to achieve predictive correction and adaptive update of the strategy, forming a complete closed-loop control of "prediction-modeling-optimization-execution-feedback". By integrating multiple technologies, dynamic and coordinated scheduling of charging resources is achieved, which effectively improves the system's adaptability to complex traffic environments, reduces user waiting time, and improves resource utilization efficiency.

[0109] The method of this disclosure will be further described below with reference to a specific embodiment, such as... Figure 2 The system framework diagram shown below contains the following specific technical solutions:

[0110] Step 1. Data Layer: Multi-source Data Acquisition and Feature Construction

[0111] A. Collect traffic flow data (vehicle arrival time, new energy type, vehicle type, etc.), charging behavior data (charging type, duration, initial and final SOC, etc.), environmental data (weather, upstream traffic congestion index, accident information, etc.) and basic power grid information from highway service area charging stations;

[0112] B. Preprocess the collected data, including filling in missing values ​​using the average of adjacent time periods and cleaning outliers.

[0113] C. Perform feature engineering to construct time period labels, congestion correlation features (the lagging correlation between upstream congestion and the arrival rate in this region), and user behavior features (average SOC in different time periods, etc.) to form a structured dataset.

[0114] Step 2. Prediction Layer: Arrival and Service Parameter Prediction Based on GRU

[0115] A. Construct a gated cyclic unit (GRU) model, using historical time-series data as input, to perform rolling predictions of arrival and service rates for AC / DC vehicles;

[0116] B. Use the sliding window method to generate training samples, and train the model using a weighted loss function (giving higher weight to errors during peak periods) and an early stopping mechanism to optimize model parameters;

[0117] C. The model outputs the arrival rate and service rate of AC / DC for the next two time steps (each step is 30 minutes), which are used as input parameters for subsequent queuing models.

[0118] Step 3. Analysis Layer: Queuing State Analysis Based on Markov Processes

[0119] A. Construct a multi-state Markov queuing model, dividing a day into 6 environmental states, and defining a system state vector (including the number of waiting vehicles, the number of busy AC / DC charging stations, the current environmental state, etc.).

[0120] B. Set state transition rules, including vehicle arrival transition (state change when AC / DC vehicle arrives), service completion transition (state change after AC / DC charging is completed), and environmental state transition (switching between different time periods).

[0121] C. Solve for the steady-state distribution of the system, calculate key performance indicators such as average waiting time, throughput, and user churn rate, and quantitatively characterize the system's operating state.

[0122] Step 4. Optimization Layer: MPC-based Charging Resource Scheduling Optimization

[0123] A. Construct a model predictive control (MPC) optimization framework, setting the prediction time domain (60 minutes) and control time domain (30 minutes), with the number of AC-to-DC charging piles as the core optimization variable;

[0124] B. Define the optimization objective as improving resource utilization and service quality (such as minimizing waiting time, maximizing throughput, and reducing user departure rate), and set constraints (total number of charging piles, upper limit of waiting time, system stability, etc.).

[0125] C. Execute the rolling optimization process, combine the GRU prediction results with the performance indicators output by the Markov queuing model, and solve for the optimal charging pile configuration strategy for the current time period.

[0126] Step 5. Execution Layer: Scheduling Strategy Execution and Closed-Loop Feedback Correction

[0127] A. Execute the charging pile configuration strategy obtained from MPC optimization, adjust the number of AC / DC charging piles through the control system, and push user guidance information based on the waiting time difference (such as prompting that the DC charging pile has a shorter waiting time).

[0128] B. Establish a feedback correction mechanism to calculate the deviation between the actual arrival rate and the predicted value. If the deviation exceeds the threshold, correct the prediction result for the next period.

[0129] C. In response to sudden scenarios (such as a sudden increase in the congestion index due to an upstream traffic accident), trigger emergency modes (such as increasing the upper limit of the number of AC-to-DC charging piles and shortening the prediction time domain) to ensure that the system adapts to dynamic changes.

[0130] The method disclosed in this embodiment addresses key challenges faced by charging stations in operation, such as drastic traffic flow fluctuations, user waiting sensitivity, and rigid resource allocation. It achieves dynamic and coordinated scheduling of charging pile resources by integrating deep learning time-series prediction, Markov queuing modeling, and model predictive control optimization techniques. This method effectively reduces average user waiting time and system operational risks while significantly improving charging pile resource utilization and service stability, exhibiting good adaptability and environmental robustness. It can be widely applied in typical scenarios such as highway service areas and urban public charging stations, demonstrating high engineering value and promising application prospects.

[0131] It should be understood that the various parts of this disclosure can be implemented in hardware, software, firmware, or a combination thereof.

[0132] The above description is merely a specific embodiment of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this disclosure should be included within the scope of protection of this disclosure. Therefore, the scope of protection of this disclosure should be determined by the scope of the claims.

Claims

1. A method for intelligent scheduling of electric vehicle charging stations, characterized in that, include: Step 1: Obtain multi-source data of charging stations in highway service areas, construct a multi-dimensional time-series feature set, and generate a training set with scheduling labels; Step 2: Train a deep learning model based on gated recurrent units using the training set, and use the trained deep learning model to make rolling predictions on vehicle arrival rate and service rate to form scheduling control input. Step 3: Construct a multi-state Markov queuing model based on the scheduling control input, considering the limitation of the number of charging piles and the user waiting characteristics, calculate the steady-state distribution and calculate the performance index accordingly. Step 4: Construct a charging port scheduling optimization model based on model predictive control methods and performance indicators to improve resource utilization and service quality. Construct an objective function and output a rolling optimization strategy for dynamic configuration of AC / DC charging piles accordingly. Step 5: Execute the rolling optimization strategy, adjust the number of AC / DC charging piles and generate indication information, and feed the running results back to the charging port scheduling optimization model to achieve closed-loop correction.

2. The method according to claim 1, characterized in that, Step 1 specifically includes: Step 1.1: Acquire historical traffic flow data, charging behavior data, environmental data, and power grid basic data of highway service area charging stations through sensors. Among them, historical traffic flow data includes vehicle arrival time, vehicle type, and whether it is a new energy vehicle collected by service area entrance cameras and ETC system; charging behavior data includes charging type, charging duration, initial SOC, and end SOC recorded by charging pile sensors; environmental data includes rainfall and visibility collected by weather stations; traffic platform obtains congestion index and traffic accident information of three upstream service areas; and power grid basic data includes peak / flat / valley time period division of the power system's time-of-use pricing. Step 1.2: Fill in the missing values ​​for traffic flow and SOC using the average of adjacent time periods. The criteria identify and remove outliers, divide a day into 6 time periods based on peak / flat / valley data in the power grid basic data, construct time period labels, calculate the lag correlation between the upstream service area congestion index and the arrival rate of this service area, and construct congestion correlation features. Step 1.3: Extract the time-series features corresponding to historical traffic flow data, charging behavior data, environmental data, and power grid basic data, and combine them with congestion-related features and time period labels to form a training set.

3. The method according to claim 2, characterized in that, Step 2 specifically includes: Step 2.1: Input the training set into the deep learning model based on gated recurrent units using the sliding window method, and calculate the loss function based on the predicted and true values. ； in, For predicted values, For the true value, For the current time, To predict the step size, For time period labels, The weighting coefficients for time period l are: For square norm operations; Step 2.2: Set the learning rate and stopping conditions, and train the deep learning model using the Adam optimizer; Step 2.3: Use the trained deep learning model to make rolling predictions of the AC arrival rate, DC arrival rate, AC service rate, and DC service rate of vehicles to form the scheduling control input.

4. The method according to claim 3, characterized in that, Step 3 specifically includes: Step 3.1: The Markov queuing model divides a day into multiple environmental states, defines the system state vector, and sets the state transition rules. Step 3.2, define the equations that satisfy the global equilibrium. Through iterative convergence Steady-state distribution is obtained ; Step 3.3: Calculate performance metrics based on the steady-state distribution. These metrics include average waiting time, throughput, and user churn rate. The expression for average waiting time is: ； ； in, The total number of vehicles waiting to be charged. Busy with AC charging stations Busy with DC charging stations This is the time period label corresponding to the current environmental state. The system is in a state The steady-state probability distribution, The arrival rate of AC vehicles within time period l. The arrival rate of DC vehicles within time period l; The expression for throughput is: ； ； in, The service rate of AC charging piles within time period l. The service rate of DC charging piles within time period l; The expression for user departure rate is: ； in, These are the fitting parameters.

5. The method according to claim 4, characterized in that, Step 4 specifically includes: Step 4.1: Construct the objective function and constraints corresponding to the charging port scheduling optimization model based on the model predictive control method and performance indicators; Step 4.2: Based on the objective function and constraints, perform rolling optimization to obtain the rolling optimization strategy for the dynamic configuration of AC / DC charging piles.

6. The method according to claim 5, characterized in that, The expression for the objective function is: ； in, For time period Weighting coefficients; The constraints include the total number of charging piles, waiting time, and system stability. The expression for the total number of charging piles constraint is as follows: ； in, , Indicates the total number of charging stations; The expression for the waiting time constraint is: ； in, For time period Maximum tolerable waiting time; The expression for the system stability constraint is: ； ； in, The arrival rate of AC vehicles within time period l. For the arrival rate of DC vehicles within time period l, and These represent the single-pile power for AC and DC respectively. This represents the upper limit of the power grid load during time period l.

7. The method according to claim 6, characterized in that, Step 4.2 specifically includes: Step 4.2.1, data collection Real-time status at any moment ; Step 4.2.2: Call the deep learning model for prediction. ; Step 4.2.3: Substitute the prediction results into the multi-state Markov queuing model to calculate the prediction time domain. ; Step 4.2.4: Solve the MPC optimization problem to obtain the optimal strategy for the current cycle. ; Step 4.2.5: Execute the optimal strategy for the current cycle, and repeat steps 4.2.1 to 4.2.4 within a preset time period to obtain the rolling optimization strategy for the dynamic configuration of AC / DC charging piles.

8. The method according to claim 7, characterized in that, Step 5 specifically includes: Step 5.1: Implement the rolling optimization strategy and adjust the number of AC / DC piles through the control system. , and, when At the same time, push notifications will be sent with waiting time information and charging station selection suggestions; Step 5.2: Calculate the deviation between the predicted value and the actual value, and correct the predicted value for the next period accordingly.

Images