Multi-target path planning method for ship surface motion of multi-heterogeneous unmanned equipment

Abstract

The invention discloses a multi-target path planning method for ship surface motion of multi-heterogeneous unmanned equipment. The method comprises the following steps: establishing a multi-agent traffic model based on a common evolution principle; a certain area weighted network diagram based on traffic flow, user distribution and power grid load is constructed, nodes in the network represent positions of potential charging stations, side weights reflect path time or cost, the number of potential places of the charging stations in the weighted network diagram and the capacity of the potential places are determined based on a multi-objective evolutionary algorithm, and the charging stations in the weighted network diagram are calculated. Site selection and capacity configuration of the charging station are optimized through simulation selection, crossing and mutation operation; based on the multi-objective evolutionary algorithm, optimizing the site selection condition of the charging station through non-dominated sorting and congestion degree distance, and obtaining a plurality of Pareto optimal solutions; according to the method, the result obtained through the multi-target method is analyzed and researched, the optimal solution of the two multi-target performance indexes is solved through the NSGA-II algorithm, and an important means is provided for solving the problem of the evolutionary algorithm in the multi-target electric vehicle charging station.

Description

Technical Field

[0001] The present invention belongs to the field of aviation support technology and intelligent unmanned systems, and proposes a multi-objective path planning method for the surface motion of multiple heterogeneous unmanned equipment for intelligent aviation support. Background Art

[0002] Recent studies have shown that the proper planning of the site selection and layout of electric vehicle charging stations is crucial to improving the usage rate and user satisfaction of electric vehicles. In the study of site selection issues, there are case studies of different cities and regions, considering the demand for electric vehicle charging infrastructure under different environmental, policy and market conditions. For example, Bouguerra and BharLayeb studied the site selection of charging stations in Tunisia, while Awasthi et al. conducted an optimization planning in Allahabad, India. Although these studies have provided useful insights into the site selection of charging stations, there are still many challenges and unresolved issues.

[0003] The problem of electric vehicle charging station location selection mainly involves how to optimize the number and location of charging stations under given conditions to minimize the total travel time for electric vehicles to reach the charging stations and the construction cost of the charging stations. However, this problem is a multi-objective optimization problem, because increasing the number of charging stations may shorten the travel time, but it will also increase the overall cost of infrastructure construction. In addition, the lack of charging infrastructure and the complexity of the spatiotemporal patterns between electric vehicle mobility make it particularly difficult to determine the optimal charging station size and location. Therefore, this study aims to solve these complex optimization problems through a multi-objective evolutionary algorithm (MOEA) to provide a more efficient solution.

[0004] At present, the traditional methods have the following shortcomings: Traditional methods mainly use single-objective optimization models, which often only consider one optimization goal, such as minimizing the construction cost of charging stations or maximizing service coverage. This method cannot effectively deal with the trade-offs between multiple objectives in reality (such as service efficiency, cost, and environmental impact). For example, although increasing the number of charging stations can reduce users' travel time, it will also lead to increased construction and maintenance costs. Therefore, single-objective models often cannot fully reflect the complexity of the site selection of electric vehicle charging stations.

[0005] Traditional optimization models are usually based on static demand forecasts and lack the ability to adapt to dynamic changes. For example, the utilization rate of electric vehicles, user behavior, and traffic flow will change over time, and traditional models often fail to consider these changes when setting parameters. This may result in the selected charging station location not meeting user needs in actual applications, thereby reducing the utilization rate of electric vehicles.

[0006] Many traditional methods often rely on some simplified assumptions when building models, such as the user's travel mode is fixed, the traffic flow is uniform, etc. Although such assumptions help simplify the model, they may lead to distorted results in practical applications. For example, actual user travel behavior is often affected by many factors (such as weather, time, specific events, etc.), and traditional methods may not be able to reflect these complex real-world situations. Summary of the invention

[0007] In view of the problems existing in the prior art, the present invention discloses a multi-target path planning method for multi-heterogeneous unmanned equipment surface motion, which specifically includes the following steps:

[0008] A multi-agent traffic model based on the principle of co-evolution is established to simulate a city's traffic network. Based on the route allocation iteration cycle and route allocation, each agent considers dynamically adjustable parameter information to design its own route to the charging station, while competing with all other agents on the traffic infrastructure for time slots; the multi-agent traffic model is an interactive framework including multiple vehicles with autonomous capabilities. In the traffic micro-simulation, each agent represents an electric vehicle with autonomous decision-making capabilities;

[0009] Construct a weighted network graph of a region based on traffic flow, user distribution and grid load. In this network, nodes represent potential charging station locations, and edge weights reflect path time or cost. Determine the number of potential locations and their capacity of charging stations in the weighted network graph based on a multi-objective evolutionary algorithm, and optimize the location and capacity configuration of charging stations through simulated selection, crossover and mutation operations.

[0010] Based on the multi-objective evolutionary algorithm, the location of the charging station is optimized by non-dominated sorting and congestion distance to obtain multiple Pareto optimal solutions;

[0011] The optimal charging station layout solution is obtained based on the optimization results.

[0012] Furthermore, before establishing a multi-agent traffic model, the driving path and charging requirements of electric vehicles are first simulated based on the urban road network and the travel mode of electric vehicles; at the beginning of the simulation, each electric vehicle agent is initialized based on the set starting position, destination and battery status, and the battery life, speed and acceleration parameters of the electric vehicle are set according to the actual situation and updated in real time during the simulation. When the electric vehicle agent selects a path at each time step, it will consider the following factors: battery life, traffic conditions and charging requirements.

[0013] Furthermore, when the electric vehicle agent determines that its own power is lower than the set threshold, it will actively look for a charging station for charging, and use the utility function to reflect the relationship between the power and the distance to the target charging station:

[0014] Ucharge =α·(E threshold -E current )-β·d chargingstation (1)

[0015] Among them, E threshold is the power threshold, d chargingstation is the distance between the charging station and the current location, α and β are weight coefficients, reflecting the impact of power shortage and distance respectively;

[0016] The Pearson correlation coefficient is used to measure the similarity between the power threshold and the distance to the charging station:

[0017]

[0018] where x i and yi represent the data points of two variables respectively. and is the mean of the data points, r is the correlation coefficient, and charging choices are made based on the correlation coefficient.

[0019] Furthermore, the dynamically adjustable parameter information includes the location of the charging station, the capacity of the charging station, the battery charge threshold, the waiting time and charging time, and the traffic flow and travel mode.

[0020] Furthermore, the multi-objective evolutionary algorithm includes a non-dominated sorting genetic algorithm and a decomposition-based multi-objective evolutionary algorithm, which respectively use the non-dominated sorting genetic algorithm and the decomposition-based multi-objective evolutionary algorithm to calculate the comprehensive index inverse generation distance IGD and the normalized hypervolume index H n (S,u,r)

[0021]

[0022] in d i The calculation is the reference point p on the Pareto approximate front P and the nearest solution x i The Euclidean distance between

[0023]

[0024] H(S,r) refers to the hypervolume index of S, whose reference vector is r, H n The value range of quality index is [0,1], H n The larger the value of , the higher the approximation between the non-dominant set S and the true hT, and the more reasonable the distribution. n The indicator better measures the extreme part of the Pareto front approximation found by each MOEA.

[0025] Due to the adoption of the above technical scheme, the present invention provides a multi-objective path planning method for the movement of multi-heterogeneous unmanned equipment decks. The method first constructs a multi-objective optimization problem by combining the framework and considering the maximum number of charging stations and a specific number of electric vehicles. Secondly, it approximates the problem through two evolutionary algorithms based on multi-objective optimization (NSGA-Ⅱ and MOEA / D-gen) to optimize the travel time and the number of charging stations. Finally, the results obtained by the multi-objective method are analyzed and studied, and the optimal solution of the two multi-objective performance indicators is solved by the NSGA-Ⅱ algorithm. The present invention provides an important means for solving the problem of evolutionary algorithms in multi-objective electric vehicle charging stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0027] Figure 1 Flow chart of the method of the present invention

[0028] Figure 2 Schematic diagram of the joint Pareto front (JPF) approximation set of the present invention

[0029] Figure 3 Schematic diagram of the convergence set of the normalized index (IGD) and DETAILED DESCRIPTION

[0030] In order to make the technical solutions and advantages of the present invention more clear, the technical solutions in the embodiments of the present invention are clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention:

[0031] like Figure 1 A multi-target path planning method for multi-heterogeneous unmanned equipment surface motion is shown, which specifically includes the following steps:

[0032] S1: Establish a multi-agent traffic model based on the principle of co-evolution, which is used to simulate a city's traffic network. Based on the route allocation iterative cycle and route allocation, each agent considers dynamically adjustable parameter information to design its own route to the charging station, while competing with all other agents on the traffic infrastructure for time slots; the multi-agent traffic model is an interactive framework including multiple vehicles with autonomous capabilities. In the traffic micro-simulation, each agent represents an electric vehicle with autonomous decision-making capabilities.

[0033] Furthermore, the multi-agent traffic model is used as a tool for traffic micro-simulation to simulate the mobility of electric vehicles. The multi-agent traffic model is a framework that views the system as multiple "agents" (i.e., individuals or vehicles) interacting with each other with autonomous capabilities. In traffic micro-simulation, each agent represents an electric vehicle (EV), which has the ability to make autonomous decisions and can interact with other agents (such as other vehicles, traffic lights, etc.). In this way, the behavior and decision-making process of EVs in a traffic environment can be accurately simulated. In the site selection scheme for electric vehicle charging piles, different regions or locations evaluate and compete for limited resources (such as land, funds, charging demand, etc.) within a predetermined time range. By reasonably dividing time slices and coordinating the competition for site selection resources among stakeholders, a quantitative score is assigned to the performance of each plan to measure the quality of the plan and compare it after execution.

[0034] Each EV agent is initialized at the beginning of the simulation based on a specific starting position, destination, and battery status. The EV's battery life, speed, acceleration, and other parameters are set based on actual conditions and updated in real time during the simulation. The EV agent considers the following factors when selecting a path at each time step:

[0035] Battery life: When the battery is low, the EV will choose a shorter route or a route that passes through a charging station.

[0036] Traffic conditions: If there is traffic congestion ahead, the EV may choose another route to save energy.

[0037] Charging demand: When the battery level is below a certain threshold, the EV will first look for a charging station to charge, ensuring that the battery level is sufficient to reach the destination. To this end, the utility function and Pearson correlation coefficient of a virtual individual (an agent created in a simulation environment and capable of autonomous execution) are used to calculate similarity. The utility function can reflect the relationship between the battery level and the distance to the target charging station. For example:

[0038] U charge =α·(E threshold -E current )-β·d chargingstation (5)

[0039] Among them, E threshold is the power threshold, d chargingstation is the distance between the charging station and the current location, α and β are weight coefficients, reflecting the impact of power shortage and distance respectively. The Pearson correlation coefficient is used to measure the linear correlation between two variables. In the charging station selection problem, the similarity between the current power and the target charging station can be calculated, for example:

[0040]

[0041] Among them, x i and yi represent data points of two variables (e.g., power and distance), and is the mean of these data points, and r is the correlation coefficient. This coefficient reflects the similarity between charging demand and charging station availability, helping EVs make the most appropriate charging choice.

[0042] Furthermore, the dynamically adjustable parameters include the following:

[0043] Charging station location: Adjust the location of charging stations based on the needs of each virtual individual to reduce travel distance and waiting time, thereby improving activity utility.

[0044] Charging station capacity: Adjust the capacity of charging stations to ensure sufficient resources to meet demand and avoid reduced utility due to overcrowding of charging stations.

[0045] Battery power threshold: Adjust the power threshold for electric vehicles to start looking for charging stations so that charging can begin at the appropriate moment to avoid problems caused by being too close to exhaustion.

[0046] Waiting time and charging time: Adjust the service efficiency of the charging station according to real-time traffic conditions and charging needs, and optimize waiting and charging times.

[0047] Traffic flow and travel patterns: Dynamically adjust the operation strategy of charging stations according to traffic flow and driving patterns to ensure that EVs are charged at the most suitable time and location.

[0048] S2: Construct a weighted network graph of a certain area based on traffic flow, user distribution and grid load. In this network, nodes represent potential charging station locations, and edge weights reflect path time or cost. The number of potential locations and their capacities of charging stations in the weighted network graph are determined based on a multi-objective evolutionary algorithm. The location and capacity configuration of charging stations are optimized through simulated selection, crossover and mutation operations.

[0049] Furthermore, by setting possible charging station locations and their capacities (charging locations are usually represented as nodes of a network graph), MOEA searches for the optimal solution during the simulation process to reduce the number, capacity, and travel time (excluding charging and waiting time) of infrastructure installations, and to construct a weighted network graph based on traffic flow, user distribution, and grid load, where nodes represent potential charging station locations and edge weights reflect path time or cost, reducing the number, capacity, and travel time of charging piles, transportation supporting facilities, power support facilities, and information and communication infrastructure;

[0050] Probability-based crossover and mutation operators are adopted, and a penalty function constraint processing method is introduced to deal with solution infeasibility and optimize the layout and quality of service (QoS) of the charging network.

[0051] S3: Based on the multi-objective evolutionary algorithm, the location of the charging station is optimized through non-dominated sorting and congestion distance to obtain multiple Pareto optimal solutions.

[0052] The objective function of the multi-objective evolutionary algorithm includes the number of charging stations and the travel time of electric vehicles, and the optimization results are evaluated based on the normalized hypervolume index (HV).

[0053] Objective function 1: Minimize the number of charging stations

[0054] The goal is to reduce the infrastructure construction cost by reducing the number of charging stations, but at the same time ensure that there are enough charging stations to cover the entire service area. Let the number of charging stations be N nation ,

[0055] The objective function is:

[0056]

[0057] Among them, x i is a binary variable indicating whether the i-th potential location is selected as a charging station, and M is the number of candidate locations for charging stations.

[0058] Objective function 2: Minimize the travel time of electric vehicles

[0059] This objective function aims to reduce the travel time of electric vehicle users from their current location to the nearest charging station. Assuming that the travel time of electric vehicles is determined by factors such as driving distance and traffic conditions, it can be expressed by the following formula:

[0060]

[0061] Among them, d ij is the distance from the current location of the electric vehicle to the charging station j, t ij is the travel time of the route, and K is the number of electric vehicle users.

[0062] The optimal configuration of a city's charging infrastructure was analyzed by using methods such as demand forecasting, geographic information system (GIS) analysis, queuing theory and queuing model, network flow optimization, game theory and market mechanism.

[0063]

[0064] where c i is a non-dominated solution x and a reference point z refAs a hypercube composed of diagonal lines, its hypervolume volume (c i )for:

[0065]

[0066] Formula (10) shows that the inferior production point z ref The non-dominated solutions x do not participate in the HV operation.

[0067] The comprehensive indicator inverse generation distance (IGD) is used to measure the distance between a given approximate solution set and the true Pareto frontier.

[0068]

[0069] in d i The calculation is the reference point p on the Pareto approximate front P and the nearest solution x i The Euclidean distance between .

[0070] set up is a vector set, u∈R m is an ideal vector, and r∈R m is a reference vector, satisfying u≤s≤r for all s∈S. The normalized hypervolume index H of the set S is n (S,u,r) is the weakly dominated part associated with S and the target space region bounded by vectors u and r. Formally, the normalized hypervolume metric is defined as:

[0071]

[0072] Where H(S,r) refers to the hypervolume index of S, and its reference vector is r. n The value range of the quality index is [0,1]. n The larger the value of , the higher the approximation between the non-dominant set S and the true hT, and the more reasonable the distribution. n The indicator better measures the extreme part of the Pareto front approximation found by each MOEA.

[0073] When calculating HV, a reference point is usually selected (for example, the maximum value of all objective functions), and then the "volume" between the optimized solution set and the reference point is calculated. This "volume" represents the target space area occupied by the solution set. In the problem of site selection for electric vehicle charging piles, HV can be used to compare the advantages and disadvantages of different site selection schemes. If the HV value of a certain scheme is large, it means that the scheme has achieved a good balance in multiple objectives (such as the number of charging piles, driving time, etc.) and covers more potential optimal solution areas.

[0074] The IGD value is obtained by calculating the Euclidean distance between the reference point and each solution on the Pareto approximate front. For each reference point, the Euclidean distance between it and the solution closest to it is calculated, and then the average distance of all reference points is calculated. In the optimization of the location of electric vehicle charging piles, IGD can be used to evaluate the closeness of the current solution set to the theoretical optimal solution set. A lower IGD value means that the optimization solution is closer to the ideal Pareto frontier, indicating that the charging pile location solution has better met the needs of multi-objective optimization.

[0075] For the computational performance comparison of Pareto frontier algorithms, two optimization algorithms, NSGA-II and MOEA / D-gen, are compared, based on Pareto frontier and decomposition methods, respectively. Based on the decomposition-based multi-objective evolutionary framework, the PF approximation problem can be decomposed into scalar optimization subproblems by using the Chebyshev method, and the objective function of the jth subproblem is

[0076]

[0077] here In one run, MOEA / D-gen optimizes all N objectives simultaneously.

[0078] Table 1 Average H of NSGA-Ⅱ and MOEA / D-gen n and IGD value

[0079]

[0080]

[0081] Table 1 shows the average H of the two algorithms NSGA-Ⅱ and MOEA / D-gen n and IGD value.

[0082] Compared with the solutions found by Ⅱ, MOEA / D-gen is more advantageous in finding non-dominant solutions and can provide better H n and IGD value.

[0083] S4: Obtain the optimal charging station layout plan based on the optimization results.

[0084] Further, in step 4, if Figure 2 and Figure 3As shown in the figure, a joint Pareto front (JPF) approximation method is used to analyze the layout of electric vehicle charging piles, and optimize the balance between the number of charging stations and travel time by selecting non-dominant solutions. According to the behavior of the selected objectives, the convergence and diversity are evaluated by the normalized index (IGD) to ensure that the reference set properly represents the true Pareto frontier. The discharge depth of the power battery has a great influence on its cycle life. Generally speaking, when the discharge depth of the power battery reaches 50% to 70%, the recharging of the power battery is most effective in extending its service life. Therefore, in order to maximize the cycle life of the power battery, the reasonable driving range of the electric vehicle is defined as the mileage that the electric vehicle can travel when the power battery pack is discharged from the optimal discharge depth to the maximum discharge depth. Reasonable cruising range of electric vehicles under uniform driving conditions

[0085] (14) Where: P EV is the rated power of the electric vehicle engine; η EV is the total power of the electric vehicle's mechanical and electrical systems, that is, the total efficiency of converting electrical energy into mechanical energy; EV It is the uniform driving speed specified in the national standard related to the range of electric vehicles. In order to ensure the driving ability of electric vehicles and meet the daily travel needs of users, the distance between two adjacent charging stations cannot be too far. In addition, in order to avoid waste of resources, the distribution of charging stations should be avoided as much as possible while meeting the charging needs of users, so that the distance between two adjacent charging stations cannot be too close. Therefore, the service radius of the charging station The actual distance D between two adjacent charging stations EVCS Should meet:

[0086]

[0087] Where: D EVCS =k*l EVCS , where l EVCS is the length of the power supply line between two adjacent charging stations, and k* is the tortuosity coefficient (the conversion coefficient of the power supply line length into the actual distance). According to the actual distance between two adjacent charging stations and the service radius of each charging station, the given preliminary site selection is screened to determine a reasonable charging station site planning scheme. At the same time, the charging service area of ​​the charging station is divided according to the Voronoi method widely used in geographic information systems, so as to guide car owners to choose the appropriate charging station for charging according to the battery status.

[0088] Furthermore, in step 3, the performance comparison of two multi-objective evolutionary algorithms, NSGA-II and MOEA / D-gen, was implemented, indicating that NSGA-II performs better in terms of convergence and the number of non-dominated solutions. Statistical methods are used to analyze the algorithm performance to ensure the significant differences in the IGD and hypervolume indicators of different algorithms, providing decision support for the layout of electric vehicle charging piles. Through simulation experiments, it is proved that the reasonable layout of charging stations can effectively reduce the travel time of electric vehicles and improve the overall service quality. An evaluation mechanism based on reference vectors and non-dominated solutions is proposed to cope with the dynamic changes in electric vehicle charging demand and improve the adaptability and effectiveness of the model.

[0089] The power allocated to the energy storage is compared with the maximum power allowed by the electrical system: If the conditional constraints are met, the capacity at the end of the estimated time step is calculated as follows:

[0090] E strg,attempt =E strg,prev -ΔP c,attempt ·Δt(16)

[0091] Where E strg,attempt is the remaining energy stored in the battery at the end of the time step, E strg,prev Indicates the battery energy storage at the beginning, E strg,level represents the minimum battery energy level, and Δt is the duration of the time step itself;

[0092] If the power value allocated to the energy storage is higher than the minimum allowed level, the average battery temperature T c,strg As shown in formula (17):

[0093]

[0094] If the energy demand is higher than the available charge capacity in the energy storage, the energy storage capacity level is set at the minimum value and the power output of the energy storage is calculated, the power contribution of the energy storage P c,strg As shown in formula (18):

[0095]

[0096] The average water temperature in the battery is set to the maximum allowable value; in order to meet the needs of industrial user loads, the backup generator contribution P is calculated c,chiller , as shown in formula (19):

[0097] P c,chiller =ΔP c,attempt -P c,strg (19)

[0098] If the power balance in Eq. (19) requests the energy storage to be higher than the maximum value allowed, then its value is set to the maximum value and the energy storage level is verified again according to the strategies in Eq. (16)–Eq. (19);

[0099] To evaluate the power balance, calculate according to formula (20):

[0100] P e,grid =P e,load +P e,chiller -P e,CHP -P e,PV -P e,wind -P e,L2 (20)

[0101] P e,load is the requirement for all industrial user loads in the method, P e,chiller The power that must be supplied to the compression chiller, P e,PV and P e,wind The predicted power generation is the industrial user load of photovoltaic power plants and wind turbines, P e,L2 It is the load of non-priority industrial users that can be reduced;

[0102] For micro-CHP as follows:

[0103] The power input P of the electric vehicle in,CHP It can be calculated according to formula (21):

[0104]

[0105] Calculate the overall efficiency η of the combined heat and power generation tot,CHP and the total efficiency PES of the primary energy saving (PES) parameter, as shown in Equations (22)–(25):

[0106]

[0107] η tot,CHP =η e,CHP +η th,CHP (twenty four)

[0108]

[0109] where η th,ref and η e,ref are the Italian reference efficiencies of heating equipment and the grid, η e,CHP Efficiency of electricity generated by CHP, η th,CHP Electricity recovered from combined heat and power generation, P c,waste Cooling power loss, P th,waste Power loss in combined heat and power generation.

[0110] like Figure 1As shown, the present invention establishes a multi-agent traffic simulation model based on the system and method for selecting the location of electric vehicle charging stations. The extended function of the electric vehicle simulator can model electric traffic scenarios. The electric vehicle extended function specifies the charging network by setting the power, location and capacity of the charging station. These settings are determined by the number of ports that can charge the vehicle at the same time. The characteristics of electric vehicles are determined by the battery capacity and the initial charging state. We developed a group of electric vehicles distributed along the city based on the population distribution of each town. The charging value of electric vehicles is low, which forces them to find charging stations and charge during driving to complete the journey. In order to complete the journey and reach the destination without exhausting the stored energy, the electric vehicle will choose the charging station with the smallest path change. If the charging station does not have a charging socket, the electric vehicle must wait until there is a charging socket. The traffic simulator simulates the driving of electric vehicles and generates a large amount of data, including the number of charging cars, power consumption, and charging time of electric vehicles. We determine the number of potential locations for charging stations on the network and their capacity. A subset of these potential charging stations and their capacities constitutes the expected solution developed by MOEA. Once the simulator tests a solution, MOEA uses evolutionary operators to find the location of the charging station. The solution reduces the amount of installed infrastructure, its capacity, and travel time (excluding charging and waiting time). Based on the simulation results, we calculate the quality of service (QoS), which is represented by the ratio of charging electric vehicles to electric vehicles to be charged. The static penalty function strategy uses QoS as a constraint. Since there are P possible charging station locations on the network, a direct representation of a solution is an array of P integers C1, C2, ..., C, ..., C. Each value C represents the capacity of each possible location, where Cj = {0, 1, 2, 3, 4} and j = {1, ..., P}. Cj = 0 means that the jth location has not chosen to deploy a station on it, otherwise Cj>0, where the Cj value refers to the capacity of the station to be installed at the jth location. Cj values ​​of 1, 2, 3, and 4 represent 5, 10, 15, and 20 ports per station, respectively. Figure 3 This representation is illustrated by an example. In this example, P = 10 possible positions are considered, each of which is assigned a value from the set {1, 2, 3, 4}. We use a single-point crossover with probability pc as the recombination operator. In this regard, a crossover point is randomly selected for both parents. All information before the crossover point is copied from the data of to the child. To complete the solution, all data beyond this point is copied from the other parent. is an example of this operator.

[0111] As a mutation operator, we implement an integer mutation with probability pm for each gene. This operator randomly selects a new gene value from {0,1,2,3,4} to change the current gene value of Cj.

[0112] In the proposed framework, the constraint handling technique adopted is based on penalty function. In our study, the penalty function is defined using the degree of constraint violation. As defined, the degree of constraint violation of a solution measures the infeasibility of a solution x∈R.

[0113] f1(x)=t(x)+λ1Φ(x) (26)

[0114] f2(x)=ns t (x)+γ2Φ(x) (27)

[0115] Among them, γ1 and γ2 are two positive scalars.

[0116] Due to the different scales of the objective and constraints, in this study, we use γ1 = 1 × 104 and γ2 = 100. Therefore, the unconstrained objective function of the problem is shown in Equations (19) and (20).

[0117] Most Pareto-based MOEAs employ a (μ+λ) selection mechanism. Therefore, they generate a population of offspring and then select the next generation from the parents and offspring. This is different from indicator- and decomposition-based MOEAs, as they are both steady-state algorithms and the single solution generated must be evaluated while employing a survival selection mechanism, and paralysis is not allowed. On the other hand, Pareto- and decomposition-based methods distinguish solutions directly based on the evaluation results of the objective function, while indicator-based methods must further calculate the indicator values ​​on which they are based. The calculation of these indicators is sometimes simple, because in addition to the number of objective functions, they must also consider additional information (such as reference vectors or reference sets). Therefore, the computational cost of this type of MOEA is usually high.

[0118] This study relied on scenario simulations to understand the values ​​of the two objective functions considered in our multi-objective optimization model. These simulations took a lot of computing time. It is worth mentioning that it takes an average of two minutes to evaluate one person on a 64-core desktop computer.

[0119] 2.6GHz processor and 64GB memory. Therefore, although our experimental study was conducted on a parallel computer, the indicator-based MOEAs experiment lasted a considerable amount of time. Therefore, we chose to exclude the indicator-based MOEAs from the experimental study.

[0120] Nevertheless, we considered two MOEAs, one each of a Pareto-based approach and a decomposition-based approach. For the former, we employed NSGA-II, which, although no longer a state-of-the-art algorithm, has shown high efficiency in dealing with dual-objective problems. MOEA / D-gen generates the entire population of offspring to be evaluated in the second step. Therefore, individuals in the offspring population can be evaluated in parallel. For the latter approach, we employed the generational version of MOEA / D, namely MOEA / D-gen proposed by Zapotecas-Martínez and Menchaca-Méndez (2020). For performance comparison, we used the following parameter settings in our experiments: population size = 50, number of generations = 50, crossover probability p c =0.9, mutation probability p m = 1 / n, where n is the number of decision variables. The simplex method is used to calculate the weight vector of MOEA / D-gen.

[0121] The EVCS optimal planning model proposed in this paper is a complex multi-objective problem, and traditional numerical methods are difficult to solve it quickly and accurately. Therefore, FC-MOEO / AEP is introduced to solve the optimal planning model. The flow chart is as follows: Figure 3 As shown in the figure, Algorithm 3 is a multi-objective balance optimizer; Algorithm 4 is the archive evolution path.

[0122] The flow chart verifies the behavior of the objective chosen for this study, and combined with the coordinate plot between the number of stations and travel time, we can see that as the number of stations increases, the travel time decreases. According to the Person coefficient, the negative correlation between the two is -0.83. The reason for this trade-off is that the more charging stations there are, the shorter the travel distance of the electric vehicle; conversely, the fewer charging stations, the longer the travel time. The JPF approximation is shown for the size of the points classified according to the total capacity of the charging ports. Note that the capacity of the system does not necessarily increase with the number of stations. For example, the solution with six stations (capacity for 105 electric vehicles) differs from the solution with twelve stations (capacity for 110 electric vehicles) by only five ports.

[0123] Below, we will analyze the phenotypes of the three solutions labeled A, B, and C. In contrast, solution B is located at the knee of the JPF in the figure. Charging station locations for A with two charging stations, B with six charging stations, and C with 13 charging stations. Blue triangles indicate active charging stations, and red dots indicate inactive charging stations. The labels under the triangles indicate the station IDs and their capacities. In addition to the number of stations and capacity, we also include QoS as an additional factor in analyzing the solutions. We measure the quality of service as the ratio of charged vehicles to vehicles waiting to be charged. This configuration is currently the best arrangement in terms of the number of stations, capacity, and quality of service. Note that solution A only sets two stations, with a total system capacity of 40 ports, which guarantees a QoS of 2.26. For this particular solution configuration, according to the simulation, the location can gather many electric vehicles because it can be regarded as a hub that aggregates many destinations and is strategically important. Solution B disperses the station locations to the periphery of the city. It should be noted that it includes two stations in solution A.

[0124] Solution B has a QoS of 9.89, which is close to the maximum value. On the other hand, solution C is the most expensive infrastructure. Its service quality is only half of the solution B QoS value in terms of the number of stations and capacity, which is due to the overestimation of the system capacity for this demand. In this solution, there may be only a few stations that concentrate more electric vehicles for charging, and their capacity is not enough in these specific places, nevertheless, in general, the system's service quality is good. More stations and capacity do not mean better service quality, so we consider QoS as a new conflicting goal for further experiments.

[0125] The process of simulating the spatiotemporal distribution of charging demand points for electric vehicles on the highway network by using the Monte Carlo method is as follows: the first step is to generate electric vehicles entering the highway network; the second step is to generate the types of electric vehicles. This paper divides the types of electric vehicles into electric logistics vehicles and household electric cars, and generates the battery parameters of electric vehicles and the starting power value of electric vehicles. This paper considers that there are charging stations built by other companies at the entrance of the highway. Electric vehicles with mileage anxiety and low starting power will choose to fill up the power before entering the highway network, so the starting power is set to more than 80%, and the travel time of electric vehicles is determined according to the time law of electric vehicle travel, and the starting point and destination position of electric vehicles are generated by the OD matrix, and the actual driving distance of the electric vehicle is calculated; the third step is to assume that when the remaining power of the electric vehicle reaches its anxious power coefficient, a charging demand is generated, and the battery power is replenished to 100%; the fourth step is that the electric vehicle travels at a uniform speed, and the spatiotemporal coordinates when the charging demand is generated are calculated; the fifth step is to repeat this Monte Carlo process in large numbers to obtain the spatiotemporal coordinate set of charging demand generating points on the highway network.

[0126] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for selecting a site for electric vehicle charging facilities based on a multi-objective evolutionary algorithm, characterized in that include: A multi-agent traffic model based on the principle of co-evolution is established to simulate a city's traffic network. Based on the route allocation iteration cycle and route allocation, each agent considers dynamically adjustable parameter information to design its own route to the charging station, while competing with all other agents on the traffic infrastructure for time slots; the multi-agent traffic model is an interactive framework including multiple vehicles with autonomous capabilities. In the traffic micro-simulation, each agent represents an electric vehicle with autonomous decision-making capabilities; Construct a weighted network graph of a region based on traffic flow, user distribution and grid load. In this network, nodes represent potential charging station locations, and edge weights reflect path time or cost. Determine the number of potential locations and their capacity of charging stations in the weighted network graph based on a multi-objective evolutionary algorithm, and optimize the location and capacity configuration of charging stations through simulated selection, crossover and mutation operations. Based on the multi-objective evolutionary algorithm, the location of the charging station is optimized by non-dominated sorting and congestion distance to obtain multiple Pareto optimal solutions; The optimal charging station layout solution is obtained based on the optimization results.

2. A multi-target path planning method for multi-heterogeneous unmanned equipment surface motion according to claim 1, characterized in that: Before establishing a multi-agent traffic model, the driving path and charging requirements of electric vehicles are simulated based on the urban road network and the travel mode of electric vehicles. At the beginning of the simulation, each electric vehicle agent is initialized based on the set starting position, destination and battery status. The battery life, speed and acceleration parameters of the electric vehicle are set according to the actual situation and updated in real time during the simulation. When the electric vehicle agent selects a path at each time step, it will consider the following factors: battery life, traffic conditions and charging requirements.

3. A multi-target path planning method for multi-heterogeneous unmanned equipment surface motion according to claim 2, characterized in that: When the electric vehicle agent determines that its own power is lower than the set threshold, it will actively look for a charging station for charging, and use the utility function to reflect the relationship between the power and the distance to the target charging station: U charge =α·(E threshold -E current )-b·d chargingstation (1) Among them, E threshold is the power threshold, d chargingstation is the distance between the charging station and the current location, α and β are weight coefficients, reflecting the impact of power shortage and distance respectively; The Pearson correlation coefficient is used to measure the similarity between the power threshold and the distance to the charging station: where x i and yi represent the data points of two variables respectively. and is the mean of the data points, r is the correlation coefficient, and charging choices are made based on the correlation coefficient.

4. A multi-target path planning method for multi-heterogeneous unmanned equipment surface motion according to claim 3, characterized in that: The dynamically adjustable parameter information includes the location of the charging station, the capacity of the charging station, the battery charge threshold, the waiting time and the charging time, and the traffic flow and travel mode.

5. The multi-target path planning method for multi-heterogeneous unmanned equipment surface motion according to claim 3 is characterized by: The multi-objective evolutionary algorithm comprises a non-dominated sorting genetic algorithm and a decomposition-based multi-objective evolutionary algorithm, which respectively use the non-dominated sorting genetic algorithm and the decomposition-based multi-objective evolutionary algorithm to calculate the comprehensive index inverse generation distance IGD and the normalized hypervolume index H n (S,u,r) in d i The calculation is the reference point p on the Pareto approximate front P and the nearest solution x i The Euclidean distance between H(S,r) refers to the hypervolume index of S, whose reference vector is r, H n The value range of quality index is [0,1], H n The larger the value of , the higher the approximation between the non-dominant set S and the true hT, and the more reasonable the distribution. n The indicator better measures the extreme part of the Pareto front approximation found by each MOEA.

Images