Unmanned vehicle charging station automatic matching method based on vehicle network interaction

Abstract

The invention belongs to the technical field of unmanned vehicle charging, and relates to an unmanned vehicle charging station automatic matching method based on vehicle network interaction, which comprises the steps that an unmanned vehicle is in communication connection with a charging station, and the charging station is connected with a power grid; for the unmanned vehicle being charged, the charging station reads the remaining charging time; when a certain unmanned vehicle needs to be charged, charging stations which are matched in charging interface, close in position and short in charging queuing time are automatically recommended; when the load of the power grid is large, the charging station obtains a reverse charging instruction and automatically searches for the unmanned vehicle which is matched with the discharging interface, close in position, sufficient in electric quantity and in an idle state in a future period of time, so that the unmanned vehicle goes to the charging station to discharge to the power grid. According to the invention, by solving the problems of the existing charging station, the charging efficiency and the power grid operation stability are improved, the operation cost is reduced, the development of vehicle-network interaction is promoted, and the system has important practical application values and social benefits.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned vehicle charging, and relates to an automatic matching method for unmanned vehicle charging stations, in particular to an automatic matching method for unmanned vehicle charging stations based on vehicle-network interaction. Background Art

[0002] With the rapid development of driverless technology, driverless vehicles are increasingly being used in urban transportation and logistics. However, the long-term operation of driverless vehicles requires sufficient power support, which puts forward new requirements for the construction and management of charging stations. Traditional charging stations often have problems such as mismatched charging interfaces, remote locations, and long charging queues, which seriously affect the operational efficiency of driverless vehicles. At the same time, with the continuous increase in the number of electric vehicles, the power grid is also facing huge peak-shaving pressure. How to efficiently utilize the energy storage capacity of driverless vehicles while ensuring the stable operation of the power grid has become an urgent problem to be solved.

[0003] The problems with the existing technology are: 1. The charging interface does not match. Traditional charging stations often have charging interfaces that do not match those of unmanned vehicles, resulting in low charging efficiency. 2. The charging stations are located in remote areas, which increases the driving distance and operating costs of unmanned vehicles. 3. The charging queue time is long. The charging equipment at the charging stations is limited, and people often have to queue up to charge, which affects the operating efficiency of unmanned vehicles. 4. The grid has high peak-shaving pressure: The disorderly charging of electric vehicles has brought great peak-shaving pressure to the grid, affecting its stable operation.

[0004] Therefore, it is necessary to provide an automatic matching method for unmanned vehicle charging stations based on vehicle-grid interaction to solve the above problems. Summary of the invention

[0005] The purpose of the present invention is to provide an automatic matching method for unmanned vehicle charging stations based on vehicle-grid interaction to solve the problems raised in the above background technology.

[0006] The present invention solves the technical problem by the following technical solutions:

[0007] An automatic matching method for unmanned vehicle charging stations based on vehicle-grid interaction, characterized in that: the unmanned vehicle is communicatively connected to the charging station, and the charging station is connected to the power grid;

[0008] For driverless vehicles that are being charged, the charging station reads the remaining charging time;

[0009] When an unmanned vehicle needs to be charged, it will automatically recommend a charging station with a matching charging port, a nearby location, and a short charging queue time;

[0010] When the grid load is heavy, the charging station receives reverse charging instructions and automatically searches for driverless vehicles with matching discharge interfaces, close locations, sufficient power, and that will be idle for a period of time in the future, allowing the driverless vehicles to go to the charging station to discharge to the grid.

[0011] Moreover, the evaluation method of the charging interface matching degree is as follows: an interface compatibility score Cij is introduced, where Cij represents the matching degree of the interface between the i-th driverless car and the j-th charging station.

[0012] Cij=Nmax,i / Nc,ij

[0013] Among them, Nc,ij represents the number of charging interface types compatible with the i-th vehicle and the j-th charging station, and Nmax,i represents the number of all charging interface types supported by the i-th vehicle.

[0014] Moreover, the evaluation method of the location proximity is: using the Euclidean distance Dij to measure the physical distance between the i-th driverless car and the j-th charging station:

[0015]

[0016] Among them, (xi,yi) and (xj,yj) represent the coordinates of the i-th vehicle and the j-th charging station respectively.

[0017] Moreover, the charging queue time prediction method is: based on historical data, a machine learning algorithm is used to predict the queue time Qj(t) of the jth charging station at the future time t, specifically:

[0018] S1, historical queue data collection and preprocessing:

[0019] Data collection: Collect historical queue time data of the jth charging station, including timestamp and corresponding queue time and other relevant data;

[0020] Data cleaning: Processing missing values: For missing data, interpolation, filling or deletion methods are used to process them;

[0021] Handling outliers: Outliers that significantly deviate from the normal range need to be identified and handled to avoid negative impact on model training;

[0022] Data transformation: transform the data into a form suitable for time series analysis as needed;

[0023] S2, time series analysis model selection and establishment: According to the characteristics of the data and the prediction target, the neural network model LSTM is selected for prediction;

[0024] Feature selection and extraction: Extract features related to queue time from historical data. These features will be used as input variables for training the model.

[0025] Model building: Use the selected algorithms and features to build a time series analysis model. For the neural network model, it is necessary to design the network structure, determine the number of nodes in the input layer, hidden layer, and output layer, and select a suitable activation function;

[0026] S3, training and optimization of time series analysis models: divide historical data into training sets and test sets for model training and verification; use training set data to train the model and adjust model parameters to minimize prediction errors; optimize the model through cross-validation, grid search and other methods to improve the model's prediction performance;

[0027] S4, prediction and evaluation based on time series analysis model:

[0028] Prediction: Use the trained model to predict the queue time Qj(t) at the future time t;

[0029] Evaluation: Use the test set data to evaluate the forecasting performance of the time series analysis model.

[0030] Moreover, the other relevant data include the charging capacity, operating status, and vehicle arrival rate of the charging station.

[0031] Moreover, when the grid load is large, the method for the charging station to obtain the reverse charging instruction is: setting a grid load threshold Lthreshold, and when the grid load exceeds this threshold, triggering the reverse charging instruction;

[0032] The discharge capability score Eik is introduced to represent the potential of the i-th driverless car to discharge to the grid in the k-th time period:

[0033] Eik=Bi×Pd,i×Δtk

[0034] Where Bi represents the battery capacity of the i-th vehicle, Pd,i represents its discharge power, and Δtk represents the length of the k-th time period.

[0035] In addition, it also includes a multi-objective optimization step, combining the charging interface matching degree, location proximity, charging queue time and grid load conditions to build a multi-objective optimization model to find the optimal charging or discharging strategy. The method is as follows:

[0036] S1, building a multi-objective optimization model;

[0037] Define the objective function:

[0038] Discharge interface matching degree: The objective function f1 represents the matching degree between the discharge interface of the vehicle to be charged and the candidate unmanned vehicle. The higher the matching degree, the smaller the f1 value.

[0039] Position proximity: The objective function f2 represents the physical distance between the vehicle to be charged and the candidate driverless vehicle. The closer the distance, the smaller the f2 value.

[0040] Battery sufficiency: The objective function f3 represents the remaining battery power of the candidate driverless vehicle. The more sufficient the battery power is, the smaller the f3 value is.

[0041] Future idle state: The objective function f4 represents the idleness of the candidate driverless vehicle in the future. The longer the idle time, the smaller the f4 value.

[0042] Combining the above four objective functions, a multi-objective optimization model can be constructed:

[0043] min F(x)=[f1(x),f2(x),-f3(x),-f4(x)]

[0044] Define constraints: the discharge interface of the candidate driverless vehicle must be compatible with the interface of the vehicle to be charged, the location of the candidate driverless vehicle must be within a certain range of the vehicle to be charged, the remaining power of the candidate driverless vehicle must meet the needs of the vehicle to be charged, and the idle time of the candidate driverless vehicle in the future must be long enough;

[0045] S2, solved using particle swarm optimization algorithm

[0046] Initialize the particle swarm: randomly generate a group of particles, each particle represents a combination of candidate driverless vehicles; initialize the position and speed of each particle, the position represents the combination of candidate driverless vehicles, and the speed represents the moving direction and size of the particle in the solution space;

[0047] Evaluate fitness: Calculate the fitness value of each particle based on the objective function and constraints; Use the Pareto dominance principle to evaluate the quality of particles and form a Pareto frontier;

[0048] Update individual optimal and global optimal: For each particle, compare the fitness value of its current position with the individual optimal position. If the current position is better, update the individual optimal position; select a global optimal position from the Pareto frontier as the learning object for all particles; in order to maintain the diversity of solutions, select the point with the largest crowding distance in the Pareto frontier as the global optimal position;

[0049] Update particle speed and position: Update the speed and position of each particle according to the update formula of the particle swarm optimization algorithm; Iteration and convergence: Repeat the above steps until the preset number of iterations is reached or the convergence condition is met. During the iteration process, the Pareto frontier is continuously updated to record the optimal solution set currently found.

[0050] Output results: Select one or more solutions from the final Pareto frontier as the optimal solution, which represents the optimal unmanned vehicle combination plan; according to actual needs, select the solution with the highest comprehensive benefit as the final solution.

[0051] The advantages and beneficial effects of the present invention are:

[0052] The automatic matching method of unmanned vehicle charging stations based on vehicle-network interaction reduces the driving distance and waiting time of unmanned vehicles and improves charging efficiency by automatically matching charging interfaces with nearby charging stations.

[0053] When the grid load is heavy, using the energy storage capacity of unmanned vehicles to discharge into the grid can help alleviate the peak load pressure of the grid and improve the stability and operating efficiency of the grid.

[0054] By optimizing the selection of charging stations and the arrangement of charging times, the operating costs of driverless vehicles are reduced and their economic benefits are improved.

[0055] The present invention realizes two-way interaction between vehicles and power grids, providing a new development direction for future intelligent transportation systems and energy management systems. The automatic matching method for unmanned vehicle charging stations based on vehicle-grid interaction improves charging efficiency and grid operation stability, reduces operating costs, and promotes the development of vehicle-grid interaction, which has important practical application value and social benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 A flow chart of the present invention using a time series analysis model to predict the queuing time of a charging station at a future time;

[0057] Figure 2 A flow chart of solving a multi-objective optimization model using particle swarm optimization in the present invention;

[0058] Figure 3 A flow chart for constructing a multi-objective optimization model for the present invention;

[0059] Figure 4 This is a flow chart of the method for solving the problem using the particle swarm optimization algorithm. DETAILED DESCRIPTION

[0060] The present invention is further described in detail below through specific examples. The following examples are only illustrative and not restrictive, and the protection scope of the present invention cannot be limited thereto.

[0061] An automatic matching method for driverless vehicle charging stations based on vehicle-grid interaction, the innovation of which is that the driverless vehicle communicates with the charging station, and the charging station is connected to the power grid. For the driverless vehicle being charged, the charging station can read the remaining charging time. When a driverless vehicle needs to be charged, it automatically recommends a charging station with a matching charging interface, a nearby location, and a short charging queue time. When the power grid load is large, the charging station obtains a reverse charging instruction and automatically searches for driverless vehicles with matching discharge interfaces, a nearby location, sufficient power, and idleness for a period of time in the future, so that the driverless vehicle goes to the charging station to discharge to the power grid:

[0062] The “future period of time” is 12 hours, 6 hours or 3 hours, which is a fixed value set in advance.

[0063] Data evaluation:

[0064] Charging interface matching evaluation: The interface compatibility score Cij is introduced to indicate the matching degree between the i-th driverless car and the j-th charging station. The calculation formula is:

[0065] Cij=Nmax,i / Nc,ij

[0066] Among them, Nc,ij represents the number of charging interface types compatible with the i-th vehicle and the j-th charging station, and Nmax,i represents the number of all charging interface types supported by the i-th vehicle.

[0067] Location proximity evaluation: Use the Euclidean distance Dij to measure the physical distance between the i-th driverless car and the j-th charging station. Calculation formula:

[0068]

[0069] Among them, (xi,yi) and (xj,yj) represent the coordinates of the i-th vehicle and the j-th charging station respectively.

[0070] Charging queue time prediction: Based on historical data, use machine learning algorithms (such as time series analysis or neural networks) to predict the queue time Qj(t) of the jth charging station at the future time t.

[0071] Specifically: Historical queuing data collection and preprocessing: Data collection: Collect historical queuing time data of the jth charging station, including timestamps and corresponding queuing times. Other relevant data may also need to be collected, such as the charging capacity, operating status, vehicle arrival rate, etc. of the charging station, in order to more comprehensively understand the factors affecting the queuing time. Data cleaning: Processing missing values: For missing data, interpolation, filling, or deletion can be used to process them. Processing outliers: For outliers that obviously deviate from the normal range, they need to be identified and processed to avoid negative impact on model training. Data conversion: Convert the data into a form suitable for time series analysis as needed, such as time series graphs, difference series, etc. Time series analysis model selection and establishment: Select a suitable time series analysis model based on the characteristics of the data and the prediction target. Commonly used models include ARIMA models, exponential smoothing models, neural network models, etc. Considering that the queuing time may be affected by multiple factors, and there may be complex nonlinear relationships between these factors, a neural network model (such as LSTM) can be selected for prediction. Feature selection and extraction: Extract features related to queuing time from historical data, such as time (hours, weeks, months, etc.), charging capacity of charging stations, vehicle arrival rate, etc. These features will be used as input variables for training models. Model establishment: Use the selected algorithm and features to establish a time series analysis model. For neural network models, it is necessary to design the network structure, determine the number of nodes in the input layer, hidden layer and output layer, and select appropriate activation functions. Training and optimization of time series analysis models: Divide historical data into training sets and test sets for model training and verification. Use the training set data to train the model and minimize the prediction error by adjusting the model parameters. Optimize the model through cross-validation, grid search and other methods to improve the prediction performance of the model. You can try different algorithms, network structures, feature combinations, etc. to find the best model configuration. Prediction and evaluation based on the time series analysis model. Prediction: Use the trained model to predict the queue time Qj(t) at the future time t. Evaluation: Use the test set data to evaluate the prediction performance of the model. Commonly used evaluation indicators include mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), etc.

[0072] The following is a detailed explanation using LSTM as an example:

[0073] Introduction to LSTM algorithm: LSTM (Long Short-Term Memory Network) is a special recurrent neural network (RNN) that can learn long-term dependencies and overcome the gradient vanishing or gradient exploding problems that traditional RNNs are prone to when processing long sequences. LSTM network structure: The LSTM network consists of an input layer, a hidden layer (containing LSTM units), and an output layer. The input layer is responsible for receiving input data and converting it into a format suitable for model processing. The hidden layer contains multiple LSTM units, each of which is able to learn and memorize long-term dependencies in the input data. The output layer is responsible for converting the output of the hidden layer into the final prediction result.

[0074] LSTM model training: During the training process, the LSTM model updates the weights and bias parameters in the network through the back-propagation algorithm to minimize the prediction error. Optimization algorithms (such as Adam, RMSprop, etc.) can be used to speed up the training process and improve the convergence performance of the model.

[0075] LSTM model prediction: In the prediction stage, the LSTM model passes input data (such as historical queuing time, charging amount, vehicle arrival rate, etc.) to the input layer. The input layer converts the data into a format suitable for LSTM unit processing and passes it to the hidden layer. The LSTM unit in the hidden layer calculates new state information and output information based on the input data and previous state information. Finally, the output layer converts the output of the hidden layer into the final prediction result (i.e., the queuing time Qj(t) at the future time t). In summary, using a time series analysis model based on historical data to predict the queuing time Qj(t) of the jth charging station at the future time t requires multiple stages such as data preparation and preprocessing, model selection and establishment, model training and optimization, prediction and evaluation. At the same time, choosing a suitable algorithm (such as LSTM) and understanding its network structure, training process, and prediction mechanism are the keys to achieving accurate predictions.

[0076] Grid load and reverse charging decision: Set the grid load threshold Lthreshold. When the grid load exceeds this threshold, the reverse charging instruction is triggered. Introduce the discharge capacity score Eik, which represents the potential of the i-th driverless car to discharge to the grid in the k-th time period. Calculation formula:

[0077] Eik=Bi*Pd,i*Δtk

[0078] Where Bi represents the battery capacity of the i-th vehicle, Pd,i represents its discharge power, and Δtk represents the length of the k-th time period.

[0079] Function expansion:

[0080] Multi-objective optimization: Combine the charging interface matching degree, location proximity, charging queue time and grid load to build a multi-objective optimization model to find the optimal charging or discharging strategy. Use heuristic algorithms such as genetic algorithm and particle swarm optimization to solve this multi-objective optimization problem.

[0081] In this embodiment, the specific steps of using particle swarm optimization to solve the multi-objective optimization model to find the optimal unmanned vehicle combination are as follows:

[0082] S1, building a multi-objective optimization model

[0083] Define the objective function: Discharge interface matching degree: objective function f1, which represents the matching degree Cij between the discharge interface of the vehicle to be charged and the candidate unmanned vehicle. The higher the matching degree, the smaller the f1 value (or the larger, depending on the specific setting, here we take the smaller as an example). Position proximity: objective function f2, which represents the physical distance Dij between the vehicle to be charged and the candidate unmanned vehicle. The closer the distance, the smaller the f2 value. Power adequacy: objective function f3, which represents the remaining power of the candidate unmanned vehicle. The more sufficient the power, the smaller the f3 value (or the larger, here we take the larger as an example, indicating a higher power adequacy) Future idle state: objective function f4, which represents the idleness of the candidate unmanned vehicle in the future. The longer the idle time, the smaller the f4 value (or the larger, here we take the larger as an example, indicating a higher idleness). Combining the above four objective functions, a multi-objective optimization model can be constructed: min F(x) = [f1(x), f2(x), -f3(x), -f4(x)] (Note that since the larger the f3 and f4, the better, negative values ​​are taken here to unify the minimum problem). f3 is the remaining power, f4 is reflected by the number of orders for available vehicles, and the idleness of f4 is inversely proportional to the number of orders.

[0084] Define the constraints: The discharge interface of the candidate driverless vehicle must be compatible with the interface of the vehicle to be charged. The location of the candidate driverless vehicle must be within a certain range of the vehicle to be charged. The remaining power of the candidate driverless vehicle must meet the needs of the vehicle to be charged. The idle time of the candidate driverless vehicle in the future must be long enough.

[0085] S2, solved using particle swarm optimization algorithm

[0086] Initialize the particle swarm: randomly generate a group of particles, each particle represents a combination of candidate driverless vehicles. Initialize the position and velocity of each particle, the position represents the combination of candidate driverless vehicles, and the velocity represents the moving direction and size of the particle in the solution space.

[0087] Evaluate fitness: Calculate the fitness value of each particle based on the objective function and constraints. Use the Pareto dominance principle to evaluate the quality of particles and form a Pareto frontier.

[0088] Update individual optimal and global optimal: For each particle, compare the fitness value of its current position with the individual optimal position. If the current position is better, update the individual optimal position. Select a global optimal position from the Pareto frontier as the learning object for all particles. In order to maintain the diversity of solutions, the point with the largest crowding distance in the Pareto frontier can be selected as the global optimal position.

[0089] Update particle speed and position: Update the speed and position of each particle according to the update formula of the particle swarm optimization algorithm. The particle speed update formula takes into account the influence of inertia, individual optimal position and global optimal position. The particle position update formula is calculated based on the current position and the updated speed.

[0090] Iteration and convergence: Repeat the above steps until the preset number of iterations is reached or the convergence condition is met. During the iteration process, the Pareto frontier is continuously updated to record the optimal solution set currently found.

[0091] Output result: One or more solutions are selected from the final Pareto frontier as the optimal solution, which represents the optimal unmanned vehicle combination plan. According to actual needs, the solution with the highest comprehensive benefit (such as comprehensive consideration of discharge interface matching, location proximity, power sufficiency and future idle state) can be selected as the final solution. Through the above steps, the particle swarm optimization algorithm can be used to solve the multi-objective optimization model and find the optimal unmanned vehicle combination plan. This method not only considers the comprehensive influence of multiple factors, but also can effectively search for the optimal solution in a complex solution space.

[0092] Real-time dynamic adjustment: Dynamically adjust the charging or discharging strategy based on real-time changes in grid load, driverless vehicle status (such as remaining power, location, driving speed, etc.) and charging station status (such as the number of idle interfaces, queue status, etc.).

[0093] Introduce the formula:

[0094] 1. Comprehensive score calculation: A comprehensive score Sij or Sik is calculated for each charging station or discharge scheme to comprehensively consider factors such as interface matching, location proximity, charging queue time (or discharge potential), etc.

[0095] Calculation formula: Sij = α*Cij + β*1 / Dij + γ*1 / Qj(t) (charging scenario); Sik = α′*Eik + β′*1 / Dik + other factors (discharging scenario), where α, β, γ, α′, β′ are weight coefficients and can be adjusted according to actual conditions.

[0096] 2. Grid load regulation target: Set the grid load regulation target Ltarget and keep the grid load within the target range through the reverse charging strategy.

[0097] Calculation formula:

[0098] ∑Eik*xik≈Ltarget-Lcurrent

[0099] Among them, xik is a decision variable (0 or 1), indicating whether the i-th vehicle participates in reverse charging in the k-th time period, and Lcurrent is the current grid load.

[0100] This technical solution can more accurately match charging stations with driverless vehicles, improve charging efficiency and grid stability, while taking into account user preferences and real-time dynamic changes to enhance user experience. A multi-objective optimization model is constructed by combining multiple factors such as discharge interface matching, location proximity, power sufficiency, and future idle state. Particle swarm optimization is used to solve the multi-objective optimization model to find the optimal combination of driverless vehicles. Specifically, it includes: building a multi-objective optimization model, defining the objective function, and defining constraints. Then the particle swarm optimization algorithm is used to solve: initialize the particle swarm, evaluate fitness, update individual optimality and global optimality, update particle speed and position, iterate and converge, and find the optimal solution set.

[0101] Although the embodiments and drawings of the present invention are disclosed for illustrative purposes, those skilled in the art will appreciate that various substitutions, changes and modifications are possible without departing from the spirit and scope of the present invention and the appended claims. Therefore, the scope of the present invention is not limited to the contents disclosed in the embodiments and drawings.

Claims

1. A method for automatically matching unmanned vehicle charging stations based on vehicle-network interaction, characterized in that: The driverless vehicle is connected to the charging station through communication, and the charging station is connected to the power grid; For driverless vehicles that are being charged, the charging station reads the remaining charging time; When an unmanned vehicle needs to be charged, it will automatically recommend a charging station with a matching charging port, a nearby location, and a short charging queue time; When the grid load is heavy, the charging station receives reverse charging instructions and automatically searches for driverless vehicles with matching discharge interfaces, close locations, sufficient power, and that will be idle for a period of time in the future, allowing the driverless vehicles to go to the charging station to discharge to the grid.

2. The method for automatically matching unmanned vehicle charging stations based on vehicle-grid interaction according to claim 1 is characterized in that: The evaluation method of the charging interface matching degree is as follows: an interface compatibility score Cij is introduced, where Cij represents the matching degree of the interface between the i-th driverless car and the j-th charging station. Cij=Nmax,i / Nc,ij Among them, Nc,ij represents the number of charging interface types compatible with the i-th vehicle and the j-th charging station, and Nmax,i represents the number of all charging interface types supported by the i-th vehicle.

3. The method for automatically matching unmanned vehicle charging stations based on vehicle-grid interaction according to claim 1 is characterized in that: The evaluation method of the location proximity is: using the Euclidean distance Dij to measure the physical distance between the i-th driverless car and the j-th charging station: Among them, (xi,yi) and (xj,yj) represent the coordinates of the i-th vehicle and the j-th charging station respectively.

4. The method for automatically matching unmanned vehicle charging stations based on vehicle-grid interaction according to claim 1 is characterized in that: The method for predicting the charging queue time is: based on historical data, a machine learning algorithm is used to predict the queue time Qj(t) of the jth charging station at the future time t, specifically: S1, historical queue data collection and preprocessing: Data collection: Collect historical queue time data of the jth charging station, including timestamp and corresponding queue time and other relevant data; Data cleaning: Processing missing values: For missing data, interpolation, filling or deletion methods are used to process them; Handling outliers: Outliers that significantly deviate from the normal range need to be identified and handled to avoid negative impact on model training; Data transformation: transform the data into a form suitable for time series analysis as needed; S2, time series analysis model selection and establishment: According to the characteristics of the data and the prediction target, the neural network model LSTM is selected for prediction; Feature selection and extraction: Extract features related to queue time from historical data. These features will be used as input variables for training the model. Model building: Use the selected algorithms and features to build a time series analysis model. For the neural network model, it is necessary to design the network structure, determine the number of nodes in the input layer, hidden layer, and output layer, and select a suitable activation function; S3, training and optimization of time series analysis models: divide historical data into training sets and test sets for model training and verification; use training set data to train the model and adjust model parameters to minimize prediction errors; optimize the model through cross-validation, grid search and other methods to improve the model's prediction performance; S4, prediction and evaluation based on time series analysis model: Prediction: Use the trained model to predict the queue time Qj(t) at the future time t; Evaluation: Use the test set data to evaluate the forecasting performance of the time series analysis model.

5. The method for automatic matching of unmanned vehicle charging stations based on vehicle-grid interaction according to claim 4 is characterized in that: The other relevant data include the charging capacity, operating status, and vehicle arrival rate of the charging station.

6. The method for automatic matching of unmanned vehicle charging stations based on vehicle-grid interaction according to claim 1, characterized in that: When the grid load is large, the method for the charging station to obtain the reverse charging instruction is: setting a grid load threshold Lthreshold, and when the grid load exceeds this threshold, triggering the reverse charging instruction; The discharge capability score Eik is introduced to represent the potential of the i-th driverless car to discharge to the grid in the k-th time period: Eik=Bi×Pd,i×Δtk Where Bi represents the battery capacity of the i-th vehicle, Pd,i represents its discharge power, and Δtk represents the length of the k-th time period.

7. The method for automatically matching unmanned vehicle charging stations based on vehicle-grid interaction according to claim 1, characterized in that: It also includes a multi-objective optimization step, combining the charging interface matching degree, location proximity, charging queue time and grid load conditions to build a multi-objective optimization model to find the optimal charging or discharging strategy. The method is as follows: S1, building a multi-objective optimization model; Define the objective function: Discharge interface matching degree: The objective function f1 represents the matching degree between the discharge interface of the vehicle to be charged and the candidate unmanned vehicle. The higher the matching degree, the smaller the f1 value. Position proximity: The objective function f2 represents the physical distance between the vehicle to be charged and the candidate driverless vehicle. The closer the distance, the smaller the f2 value. Battery sufficiency: The objective function f3 represents the remaining battery power of the candidate driverless vehicle. The more sufficient the battery power is, the smaller the f3 value is. Future idle state: The objective function f4 represents the idleness of the candidate driverless vehicle in the future. The longer the idle time, the smaller the f4 value. Combining the above four objective functions, a multi-objective optimization model can be constructed: min F(x)=[f1(x),f2(x),-f3(x),-f4(x)] Define constraints: the discharge interface of the candidate driverless vehicle must be compatible with the interface of the vehicle to be charged, the location of the candidate driverless vehicle must be within a certain range of the vehicle to be charged, the remaining power of the candidate driverless vehicle must meet the needs of the vehicle to be charged, and the idle time of the candidate driverless vehicle in the future must be long enough; S2, solved using particle swarm optimization algorithm Initialize the particle swarm: randomly generate a group of particles, each particle represents a combination of candidate driverless vehicles; initialize the position and speed of each particle, the position represents the combination of candidate driverless vehicles, and the speed represents the moving direction and size of the particle in the solution space; Evaluate fitness: Calculate the fitness value of each particle based on the objective function and constraints; Use the Pareto dominance principle to evaluate the quality of particles and form a Pareto frontier; Update individual optimal and global optimal: For each particle, compare the fitness value of its current position with the individual optimal position. If the current position is better, update the individual optimal position; select a global optimal position from the Pareto frontier as the learning object for all particles; in order to maintain the diversity of solutions, select the point with the largest crowding distance in the Pareto frontier as the global optimal position; Update particle speed and position: Update the speed and position of each particle according to the update formula of the particle swarm optimization algorithm; Iteration and convergence: Repeat the above steps until the preset number of iterations is reached or the convergence condition is met. During the iteration process, the Pareto frontier is continuously updated to record the optimal solution set currently found; Output: One or more solutions are selected from the final Pareto frontier as the optimal solutions, which represent the optimal unmanned vehicle combination solutions; According to actual needs, the solution with the highest comprehensive benefit is selected as the final solution.

Images