New energy automobile charging pile intelligent scheduling management method and system

Abstract

The application discloses a new energy automobile charging pile intelligent scheduling management method and system, and relates to the technical field of charging scheduling.The method collects real-time state of charging piles and vehicle request data, constructs a dynamic scheduling graph containing charging pile nodes, vehicle nodes and weighted edges; performs hierarchical walk sampling on the dynamic scheduling graph, extracts multi-scale local topological features of each charging pile node; combines the topological features with real-time power and fault state to calculate comprehensive scheduling priority; and generates a scheduling instruction sequence according to the priority order.The system includes data acquisition, dynamic graphing, topological sampling, priority calculation and scheduling generation modules.The scheme breaks through the information limitation of traditional static scheduling by dynamic graph modeling and multi-scale topological feature fusion, and real-time adapts to the spatio-temporal changes of charging pile groups and vehicle requests, improves the global rationality and response efficiency of scheduling decisions, balances the load and reduces the waiting time.

Description

Technical Field

[0001] This invention relates to the field of new energy vehicle charging scheduling technology, specifically to a method and system for intelligent scheduling and management of new energy vehicle charging piles. Background Technology

[0002] Current new energy vehicle charging scheduling management faces the challenge of matching the uneven distribution of charging pile resources with the dynamic changes in vehicle charging requests. Traditional charging scheduling methods are usually based on static rules or the simple nearest-distance principle, directly guiding vehicles to the nearest available charging pile in a straight line. This approach only considers a single spatial factor and cannot perceive the real-time queuing status, available power, and vehicle request distribution across the entire charging service network. When a large number of vehicles simultaneously request charging within a short period, the distance-first scheduling strategy can cause some charging piles to become overcrowded, resulting in a sharp increase in queuing times, while other charging piles with sufficient available power and shorter queues remain idle or underutilized. This leads to a decrease in the overall service efficiency of the charging pile network and an increase in vehicle waiting times.

[0003] Existing charging dispatch systems lack effective structural representation methods for the charging service network comprised of charging pile clusters and vehicles waiting to be charged. Dispatch decisions typically assess the availability of individual charging piles in isolation, ignoring the complex connections and topological dependencies between charging piles and vehicles caused by factors such as spatial distance and waiting time. The local structural characteristics of each node in the dynamically changing service network cannot be quantitatively extracted, making it difficult for the dispatch system to capture the positional importance and connection patterns of each charging pile in the network from a global perspective, and thus unable to perform accurate priority ranking based on deep information about the network topology.

[0004] When introducing graph structures to model charging service networks, it is necessary to address the problem of how to efficiently and discriminatively extract node topological features. Conventional graph node feature extraction methods sample at a single scale, failing to capture multi-granularity structural information of nodes across different neighborhoods, resulting in insufficient feature representation capabilities. Furthermore, existing scheduling priority evaluation lacks a computational mechanism that effectively couples graph topological features with the real-time operating status of charging piles. Topological features are disconnected from key practical constraints such as power capacity and fault status during the decision-making process, making it difficult to ensure the comprehensive optimization of scheduling instructions under multiple objectives such as safety, efficiency, and load balancing. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent scheduling and management method and system for new energy vehicle charging piles. By constructing a dynamic scheduling graph and performing hierarchical walk sampling to extract multi-scale local topological feature vectors of charging pile nodes, and combining a priority regressor with real-time power and fault status to calculate comprehensive scheduling priority, this invention solves the problems of existing scheduling methods being unable to perceive the global topology of the charging service network and the difficulty in integrating multi-source real-time status for dynamic priority decision-making, thereby achieving efficient and balanced scheduling between charging pile groups and vehicles waiting to be charged.

[0006] To achieve the above objectives, the present invention provides the following technical solution: The present invention provides an intelligent scheduling and management method for new energy vehicle charging piles, which collects real-time status data of each charging pile in the charging pile group and request data of each vehicle waiting to be charged; the real-time status data includes the current operating voltage, current operating current, internal temperature, remaining available power, current queue length and average charging time of each charging pile, and the request data includes the current location coordinates, remaining power, expected charging amount and urgency level tag of each vehicle waiting to be charged.

[0007] A dynamic scheduling graph is constructed based on the real-time status data of each charging pile and the request data of each vehicle waiting to be charged. The dynamic scheduling graph includes charging pile nodes, vehicle nodes, and edges connecting the charging pile nodes and vehicle nodes. As a technical solution of this invention, each charging pile is mapped to a charging pile node, and each vehicle waiting to be charged is mapped to a vehicle node. For each vehicle node, the spatial distance and expected waiting time between the vehicle node and each charging pile node are calculated. The expected waiting time is determined based on the current queue length and average charging time of the charging pile node. When the spatial distance is less than a preset distance threshold and the expected waiting time is less than a preset time threshold, an edge is established between the vehicle node and the charging pile node, and an initial weight is assigned to this edge. The initial weight is inversely proportional to both the spatial distance and the expected waiting time. This dynamic graph construction method enables the scheduling graph to accurately reflect the accessibility and timeliness between vehicles and charging piles, improving the rationality of subsequent scheduling decisions.

[0008] Preferably, the initial weights of the edges in the dynamic scheduling graph are dynamically updated over time. After each scheduling instruction is executed, the spatial distance and expected waiting time between the affected vehicle node and the charging pile node are recalculated and the initial weights of the corresponding edges are updated to ensure that the scheduling graph is always consistent with the current actual road conditions and charging pile occupancy status.

[0009] A hierarchical walk sampling method is used to perform hierarchical walk sampling on the dynamic scheduling graph to obtain the local topological feature vector of each charging pile node. As a further improvement of this invention, each charging pile node is used as the starting node, and multiple walk step size levels are set, with each level corresponding to a walk step size range. Within each level, a random walk is performed starting from the starting node according to the initial weight of the edges, and the charging pile nodes and vehicle nodes visited during the walk are recorded. All the visited nodes are arranged in the order of visit to form the walk sequence of that level. The walk sequences of all levels corresponding to the same charging pile node are concatenated to form the multi-level walk path of that charging pile node. The multi-level walk path of each charging pile node is transformed by one-hot encoding to obtain the local topological feature vector of that charging pile node. Hierarchical walk sampling can capture the topological structure at different distance scales, fully extract the local connection patterns and competition relationships around the charging pile nodes, and provide a basis for accurately evaluating the scheduling value.

[0010] Preferably, the specific method for setting multiple walking step length levels is as follows: determine the maximum shortest path length between all charging pile nodes and vehicle nodes in the dynamic scheduling diagram; divide the maximum shortest path length by three and round up to obtain the first step length threshold; multiply the maximum shortest path length by two-thirds and round up to obtain the second step length threshold; use the walking step length range from one to the first step length threshold as the first level, from the first step length threshold plus one to the second step length threshold as the second level, and from the second step length threshold plus one to the maximum shortest path length as the third level.

[0011] Preferably, the specific method of random walk according to the initial weight of the edge is as follows: in each step of the walk, obtain the initial weight of all edges connected to the current node, and calculate the sum of all initial weights; generate a transition probability for each edge that is proportional to the initial weight of the edge, the transition probability being equal to the initial weight of the edge divided by the sum of all initial weights; randomly select an edge to move to an adjacent node according to the transition probability, and if the current node is connected to both a charging pile node and a vehicle node, then it is allowed to move to any type of adjacent node.

[0012] The comprehensive scheduling priority of each charging pile node is calculated based on its local topology feature vector. As an optional technical solution of this invention, the local topology feature vector of each charging pile node is input into a pre-trained priority regressor, which outputs an initial priority score. Simultaneously, the remaining available power and current fault flag bit in the real-time status data of the charging pile node are obtained. If the current fault flag bit indicates a fault, the comprehensive scheduling priority of the charging pile node is set to zero; if no fault exists, the initial priority score of the charging pile node is multiplied by the normalized value of the remaining available power to obtain the comprehensive scheduling priority of the charging pile node. This calculation method integrates the scheduling potential contained in the local topology feature vector with the actual available capacity of the charging pile, and directly blocks faulty charging piles in fault conditions, effectively avoiding guiding vehicles to unavailable charging piles, improving scheduling safety and resource utilization.

[0013] Preferably, the normalized value of the remaining available power is the ratio of the remaining available power of the charging pile node to the maximum remaining available power of all charging pile nodes in the charging pile group.

[0014] Preferably, the training method of the pre-trained priority regressor is as follows: collect the historical local topology feature vector and historical comprehensive scheduling priority label of each charging pile node within a historical time period. The historical comprehensive scheduling priority label is back-calibrated according to the actual usage frequency and average waiting time of the charging pile node in the historical scheduling results. The historical local topology feature vector is used as input, and the historical comprehensive scheduling priority label is used as output. The mean squared error loss function is used to iteratively train the multilayer perceptron network until the loss function value is less than a preset loss threshold, thus obtaining the priority regressor.

[0015] A scheduling instruction sequence is generated based on the comprehensive scheduling priority of each charging pile node. As a specific scheduling strategy of this invention, all charging pile nodes are sorted from high to low according to their comprehensive scheduling priority, resulting in a charging pile priority queue; each vehicle waiting to be charged is sorted from high to low according to the urgency level in its request data, resulting in a vehicle emergency queue; one vehicle is sequentially taken from the vehicle emergency queue, and the charging pile node with the highest comprehensive scheduling priority is selected from the charging pile priority queue, generating a scheduling instruction to dispatch the vehicle to that charging pile node; the matched charging pile node is removed from the charging pile priority queue, and the process continues for the next vehicle, until either the vehicle emergency queue or the charging pile priority queue is empty. All scheduling instructions are then combined into a scheduling instruction sequence according to their generation order. This priority-based dynamic matching mechanism prioritizes the charging needs of emergency vehicles and ensures that high-quality charging pile resources with higher comprehensive priorities are utilized first, thus shortening user waiting time and balancing the workload of the charging pile cluster.

[0016] As a further optimization, when the spatial distance between a vehicle to be charged taken from the vehicle emergency queue and the charging pile node with the highest overall scheduling priority selected from the charging pile priority queue is greater than a preset distance threshold, the charging pile node is skipped and the next charging pile node with the highest overall scheduling priority is selected for matching, thereby avoiding invalid scheduling and driving energy consumption caused by excessive spatial distance and improving the feasibility of executing scheduling instructions.

[0017] As one implementation of the urgency level label in this invention, the remaining battery power and navigation distance from the current location to the nearest charging station group are obtained for each vehicle to be charged. The vehicle's remaining range is calculated based on the remaining battery power, and the ratio of the navigation distance to the remaining range is used as a battery risk coefficient. The battery risk coefficients are sorted from highest to lowest, and the top 30% of vehicles are marked as high urgency, the middle 40% as medium urgency, and the bottom 30% as low urgency. The urgency level label represents either high, medium, or low urgency. This urgency level quantification method objectively reflects the risk level of insufficient vehicle range, providing a reliable basis for scheduling priority allocation.

[0018] This invention provides an intelligent scheduling and management system for new energy vehicle charging piles, comprising: The data acquisition module is used to collect real-time status data of each charging pile in the charging pile group and request data of each vehicle waiting to be charged. The dynamic graph construction module is used to construct a dynamic scheduling graph based on the real-time status data of each charging pile and the request data of each vehicle to be charged. The dynamic scheduling graph includes charging pile nodes, vehicle nodes and edges connecting charging pile nodes and vehicle nodes. The topology sampling module is used to perform hierarchical walk sampling on the dynamic scheduling graph to obtain the local topology feature vector of each charging pile node; The priority calculation module is used to calculate the comprehensive scheduling priority of each charging pile node based on the local topology feature vector of each charging pile node. The scheduling generation module is used to generate a sequence of scheduling instructions based on the comprehensive scheduling priority of each charging pile node.

[0019] This invention, through dynamic mapping, hierarchical topology sampling, and priority regression calculation, can perceive the spatiotemporal coupling relationship between charging pile groups and vehicles waiting to be charged in real time, comprehensively evaluate the scheduling value of each charging pile, and generate a scientific and reasonable scheduling instruction sequence by combining fault status and power margin. While improving the utilization rate of charging piles, it significantly reduces the average waiting time of users and enhances the overall operating efficiency and intelligence level of the charging service network.

[0020] The technical effects and advantages provided by the present invention in the above technical solution are as follows: By constructing a dynamic scheduling graph, charging piles and vehicles are mapped as nodes. Weighted edges are established based on spatial distance and expected waiting time, and the weights are dynamically updated over time, allowing the scheduling graph to reflect the network topology and time-varying state in real time. This dynamic association modeling breaks through the limitations of static rules, enabling scheduling decisions to simultaneously perceive spatial reachability and time expectations, reducing resource mismatch caused by information lag, and making the load distribution of charging pile groups more in line with actual needs.

[0021] Hierarchical walking sampling extracts multi-scale topological features of charging pile nodes from different long-level hierarchies, capturing neighborhood connection details at close range, identifying regional competition patterns at mid-range, and quantifying network location attributes at long range. Multi-level feature fusion enhances the expressive power of node structural differences. Compared with single-scale sampling, it can more accurately distinguish the scheduling value of different charging piles in the network, providing a more granular structural basis for priority calculation and reducing decision bias.

[0022] By combining topological features with real-time physical state to calculate comprehensive scheduling priorities, the historical usage patterns inherent in the graph structure are coupled with current power margins and fault states. Fault flags directly mask unavailable nodes, and power normalization factors dynamically adjust priority weights, ensuring that scheduling recommendations simultaneously conform to long-term operational patterns and real-time safety constraints. This avoids directing vehicles to inefficient or dangerous nodes, improving the reliability and rationality of resource allocation. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0024] Figure 1 This is a flowchart of an intelligent scheduling and management method for new energy vehicle charging piles. Figure 2 This is a flowchart of the dynamic scheduling graph construction process by the dynamic graph construction module; Figure 3 This is a flowchart of the multi-level topology sampling process for charging pile nodes; Figure 4 It is a flowchart of the scheduling instruction generation process based on spatial distance matching; Figure 5 This is a schematic diagram of the structure of an intelligent scheduling and management system for new energy vehicle charging piles. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] See Figure 1 This invention provides an intelligent scheduling and management method for new energy vehicle charging piles, specifically including: Collect real-time status data of each charging pile in the charging pile group and request data of each vehicle waiting to be charged. A dynamic scheduling graph is constructed based on the real-time status data of each charging pile and the request data of each vehicle to be charged. The dynamic scheduling graph includes charging pile nodes, vehicle nodes, and edges connecting charging pile nodes and vehicle nodes. A hierarchical walk sampling is performed on the dynamic scheduling graph to obtain the local topological feature vector of each charging pile node; The comprehensive scheduling priority of each charging pile node is calculated based on the local topology feature vector of each charging pile node. A sequence of scheduling instructions is generated based on the comprehensive scheduling priority of each charging pile node.

[0027] In specific implementation, please refer to Figure 2The dynamic mapping module receives real-time status data of the charging pile group and request data from vehicles waiting to be charged from the data acquisition module. The dynamic mapping module maps each charging pile in the charging pile group to a charging pile node and each vehicle waiting to be charged to a vehicle node. The attributes of a charging pile node include its current queue length and average charging time, while the attributes of a vehicle node include its current location coordinates.

[0028] The dynamic mapping module iterates through all charging pile nodes in the dynamic scheduling graph for each vehicle node, calculating the spatial distance and expected waiting time between the vehicle node and each charging pile node. The spatial distance is calculated using the Euclidean distance between the current coordinates of the vehicle node and the installation coordinates of the corresponding charging pile. The expected waiting time is calculated by obtaining the current queue length and average charging time of each charging pile node, multiplying the current queue length by the average charging time to obtain the expected waiting time. The dynamic mapping module has preset distance and time thresholds; in this embodiment, the preset distance threshold is set to 5 kilometers, and the preset time threshold is set to 30 minutes.

[0029] When the calculated spatial distance is less than a preset distance threshold, and the calculated expected waiting time is less than a preset time threshold, the dynamic mapping module establishes an edge between the vehicle node and the charging pile node, and assigns an initial weight to this edge. The formula for determining the initial weight is: ; in, This represents the initial weight of the edge connecting the i-th vehicle node and the j-th charging pile node; This represents the spatial distance between the i-th vehicle node and the j-th charging pile node, in meters. This represents the expected waiting time for the i-th vehicle node corresponding to the j-th charging pile node, in minutes. The spatial distance adjustment factor is set to 0.6; The adjustment factor for the expected waiting time is set to 0.4; As a normalization constant, it is set to 100. The method of determination is as follows ,in, It is the current queue length of the j-th charging pile node, representing the number of vehicles waiting to charge in front of the j-th charging pile node; is the average charging time of the j-th charging pile node, representing the average historical charging time of the j-th charging pile node, in minutes per vehicle. Using the current position coordinates of the i-th vehicle node The position coordinates of the j-th charging pile node Calculated, i.e. Spatial distance adjustment factor And the expected value adjustment factor for waiting time The value of makes the influence of spatial distance on the initial weights slightly greater than the influence of the expected waiting time, and the normalization constant. This is used to map initial weights to a selected order of magnitude. The initial weights of the edges are dynamically updated over time. After each scheduling instruction is generated and executed, the dynamic graph construction module re-obtains the latest current queue length and latest average charging time for each charging pile node in the charging pile group. For all edges between vehicle nodes and charging pile nodes that still exist in the dynamic scheduling graph, the spatial distance corresponding to each edge is recalculated. and expected waiting time And the recalculated result and Substitute the initial weights into the initial weight calculation formula to update the initial weights of each edge. For vehicles that have completed charging and left, the dynamic graph construction module deletes the corresponding vehicle node and all edges connected to that vehicle node from the dynamic scheduling graph; for newly appearing vehicles waiting to be charged, the dynamic graph construction module creates a new vehicle node for it, establishes new edges according to the judgment conditions of spatial distance and expected waiting time, and calculates the initial weights.

[0030] In specific implementation, please refer to Figure 3 The topology sampling module reads the dynamic scheduling graph generated by the dynamic graphing module. The dynamic scheduling graph contains multiple charging pile nodes and multiple vehicle nodes. The topology sampling module sets multiple walk step size levels for each charging pile node in the dynamic scheduling graph. Each walk step size level corresponds to a walk step size range, which is used to perform multi-scale sampling of the local connection structure of the charging pile node.

[0031] The topology sampling module, with its multiple walk step levels, is implemented as follows: It determines the maximum shortest path length between all charging pile nodes and all vehicle nodes in the dynamic scheduling graph. The module traverses each charging pile node and each vehicle node in the dynamic scheduling graph, using Dijkstra's algorithm to calculate the shortest path length between each pair of charging pile nodes and vehicle nodes, and records the maximum value among all shortest path lengths as the maximum shortest path length. During the traversal, since the dynamic scheduling graph is a bipartite graph with no direct edges between charging pile nodes and no direct edges between vehicle nodes, the shortest path calculation always starts from a charging pile node and ends at a vehicle node. The maximum shortest path length is denoted as... The topology sampling module will determine the maximum and shortest path lengths. Divide by three and round up to obtain the first step long threshold, denoted as . ; Calculate the maximum and shortest path length Multiply by two-thirds and round up to obtain the second step size threshold, denoted as . The walking step length range is from one to the first step length threshold. As the first level, starting from the first long threshold Add one to the second step threshold As the second level, starting from the second step size threshold Add one to the maximum shortest path length As the third level. When When the value is less than 3, the range of the first-level walking step length is 1 to... The second and third levels are no longer distinguished; when When the value is 3, the first level is from 1 to 1, the second level is from 2 to 2, and the third level is from 3 to 3. The three levels defined in this way correspond to short-range topology exploration, medium-range topology exploration, and long-range topology exploration, respectively.

[0032] For each charging pile node in the dynamic scheduling graph, the topology sampling module uses that charging pile node as the starting node and performs a random walk within each walk step size level. The specific process of the random walk is as follows: within the walk step size range corresponding to a selected walk step size level, a random integer value is selected as the number of steps for this random walk, denoted as . Starting from the initial node, move gradually, for a total of [number] moves. Step. In each step of the walk, the topology sampling module obtains the initial weights of all edges connected to the current node and calculates the sum of all initial weights. Assume the current node is... , with nodes The set of adjacent nodes is Connecting nodes With neighboring nodes The initial weight of the edge is denoted as Then the sum of all initial weights is `k` is an index variable for the mathematical summation symbol, used to iterate through each node in the set of all neighboring nodes of the current node `u` (i.e., `k` represents the number or identifier of the neighboring node). The topology sampling module is for nodes. Each edge generates a transition probability, and the formula for calculating the transition probability is as follows: ;in, Indicates starting from the current node Move to adjacent node The transition probability; Indicates connection to the current node With neighboring nodes The initial weights of the edges are calculated by the dynamic graph construction module based on the expected values ​​of spatial distance and waiting time. Indicates the current node The set of all adjacent nodes, the current node It is a charging pile node or a vehicle node; Indicates the current node The sum of the initial weights of all connected edges. Transition probability. The value of is proportional to the initial weight of the edge; the larger the initial weight of the edge, the higher the probability of moving along that edge.

[0033] The topology sampling module determines the target node for movement using a roulette wheel selection method based on all calculated transition probabilities. This involves randomly generating a random number between 0 and 1, selecting a target node within the corresponding interval based on the cumulative distribution, and then moving to that node. If the current node is connected to both a charging pile node and a vehicle node, the topology sampling module allows random walks to move to adjacent nodes of any type without imposing type restrictions. During the movement, the topology sampling module records the starting node and the nodes reached at each step, forming an ordered node sequence. When the number of movement steps reaches [a certain threshold], [the module continues this process]. After each step, the random walk is stopped, and the recorded ordered node sequence is taken as the walk sequence corresponding to that walk step size level. The topology sampling module performs the above random walk operation once for each walk step size level for the same charging pile node, resulting in walk sequences at three levels. The walk sequences at the three levels are then concatenated in the order of the first level, the second level, and the third level to form a complete ordered node sequence, which is the multi-level walk path for that charging pile node.

[0034] The topology sampling module performs unique encoding and transformation on the multi-level traversal paths of each charging pile node to obtain the local topology feature vector of that charging pile node. Specifically, the topology sampling module pre-assigns a unique integer index number to each charging pile node and each vehicle node in the dynamic scheduling graph, and determines the total number of nodes in the dynamic scheduling graph, denoted as . The one-hot encoding vector of a node is a vector with dimension 1. The vector is such that the element at the index of the node is 1, and the elements at other positions are 0. For each node appearing in the multi-level traversal path, the topology sampling module generates a one-hot encoded vector for that node. The one-hot encoded vectors of all nodes in the multi-level traversal path are then summed element-wise to obtain a vector with dimension . The aggregated vector is used directly as the local topology feature vector of the charging pile node. The value of each dimension in the local topology feature vector represents the total number of times the node with the corresponding index number is visited in the multi-level traversal path.

[0035] In practical implementation, the priority calculation module receives the local topological feature vector of each charging pile node from the topology sampling module, and the real-time status data of each charging pile node from the data acquisition module. Internally, the priority calculation module deploys a pre-trained priority regressor, which employs a multilayer perceptron network structure. The multilayer perceptron network consists of an input layer, a first hidden layer, a second hidden layer, and an output layer, with fully connected layers. The number of neurons in the input layer is equal to the dimension of the local topological feature vector. This represents the total number of charging pile nodes and vehicle nodes in the dynamic scheduling graph. The first hidden layer contains 128 neurons, and the second hidden layer contains 64 neurons. Both the first and second hidden layers use linear rectified functions as their activation functions. The output layer contains one neuron and does not use an activation function; it directly outputs a real value as the initial priority score. During the forward propagation of the multilayer perceptron network, the local topological feature vector is input from the input layer, passes through the first and second hidden layers sequentially, and is finally output by the output layer as the initial priority score.

[0036] For each charging pile node, the priority calculation module uses the node's local topology feature vector as input and feeds it into the priority regressor for forward computation. The priority regressor outputs the initial priority score for that charging pile node. Simultaneously, the priority calculation module extracts the remaining available power and the current fault flag bit from the node's real-time status data. The current fault flag bit is a Boolean value; a true value indicates a fault in the charging pile node, and a false value indicates no fault. The priority calculation module reads the content of the current fault flag bit and determines the overall scheduling priority.

[0037] If the current fault flag is true, the priority calculation module sets the overall scheduling priority of the charging pile node to zero. If the current fault flag is false, the priority calculation module calculates the normalized value of the remaining available power of the charging pile node. The normalized value of the remaining available power is calculated as follows: obtain the remaining available power of all charging pile nodes in the charging pile group, find the maximum value, and take it as the maximum remaining available power; divide the remaining available power of the charging pile node by the maximum remaining available power to obtain the normalized value of the remaining available power. Then, the priority calculation module multiplies the initial priority score of the charging pile node by the normalized value of the remaining available power. The product is the overall scheduling priority of the charging pile node, and the calculation method is expressed by the following formula: ; in, Indicates the first The overall scheduling priority of each charging pile node; This indicates the first priority regressor output. The initial priority score of each charging pile node; Indicates the first The remaining available power of each charging station node, in kilowatts; This represents the maximum remaining available power of all charging pile nodes in the charging pile group, expressed in kilowatts. The method for obtaining the value is as follows: traverse the remaining available power of each charging pile node in the charging pile group, and determine the maximum value by comparing them one by one. By priority regressor for the first The local topological feature vectors of each charging pile node are obtained by forward propagation calculation.

[0038] The training process of the pre-trained priority regressor is completed offline. During the training phase, historical data is collected over a period of thirty days. For each day within this historical period, a dynamic scheduling graph is constructed and hierarchical walk sampling is performed in the same manner as in online operation, obtaining the historical local topological feature vector for each charging pile node each day. Simultaneously, for each charging pile node, a historical comprehensive scheduling priority label is calculated based on its actual scheduling results within the historical period. The historical comprehensive scheduling priority label is defined as follows: the actual usage frequency and average waiting time of the charging pile node are statistically analyzed for each day within the historical period. The actual usage frequency is defined as the number of charging tasks assigned to the charging pile node each day, and the average waiting time is defined as the average time interval between the request and the start of charging for all vehicles assigned to the charging pile node. For each charging pile node, the statistically obtained daily actual usage frequency is subjected to min-max normalization to obtain the normalized usage frequency; the daily average waiting time is inversely normalized by subtracting the average waiting time after min-max normalization from 1 to obtain the inversely normalized waiting time. The average of the normalized usage frequency and the reverse normalized waiting time is used as the historical comprehensive scheduling priority label for that charging pile node on that day.

[0039] The training process of the multilayer perceptron network is as follows: Historical local topological feature vectors of all charging pile nodes across all dates are collected as input samples, and the corresponding historical integrated scheduling priority labels are used as the expected output. The mean squared error loss function is used to measure the difference between the output of the multilayer perceptron network and the expected output. The mean squared error loss function is calculated as follows: for each training sample, the square of the difference between the initial priority score output by the multilayer perceptron network and the historical integrated scheduling priority label is calculated, and the average of the squared differences of all training samples is calculated. An adaptive moment estimation optimizer is used to update the weight parameters of the multilayer perceptron network. The learning rate of the adaptive moment estimation optimizer is set to 0.001, and the exponential decay rate is... Set to 0.9, The threshold value is set to 0.999. During training, in each iteration, all training samples are input into the multilayer perceptron network for forward propagation, and the loss function value is calculated. The gradient is calculated through backpropagation, and the network weights are updated using the gradient. Iteration continues until the loss function value is less than a preset loss threshold, which is set to 0.001. When the loss function value is less than 0.001, training stops, and all weight parameters and bias parameters of the multilayer perceptron network are saved. This trained multilayer perceptron network is the pre-trained priority regressor.

[0040] In specific implementation, please refer to Figure 4 The scheduling generation module receives the comprehensive scheduling priority of all charging pile nodes from the priority calculation module, and the request data of all vehicles waiting to be charged from the data acquisition module. The scheduling generation module then extracts the urgency level tag for each vehicle from the request data of all vehicles waiting to be charged.

[0041] The scheduling generation module constructs the charging pile priority queue as follows: It iterates through all charging pile nodes in the charging pile group, combining the node identifier of each charging pile node with its corresponding comprehensive scheduling priority into an element, and puts all elements into a list. This list is then sorted from highest to lowest comprehensive scheduling priority, using descending order of comparison, with charging pile nodes having higher comprehensive scheduling priority values ​​preceding those with lower values. If multiple charging pile nodes have the same comprehensive scheduling priority, they are arranged lexicographically according to their node identifiers. The resulting sequence of charging pile nodes after sorting constitutes the charging pile priority queue, with charging pile nodes at the beginning of the queue having higher comprehensive scheduling priority.

[0042] The scheduling generation module constructs the vehicle emergency queue as follows: It obtains the urgency level tag for each vehicle waiting to be charged, with values ​​ranging from high to low urgency. A value of 3 is assigned to high urgency, 2 to medium urgency, and 1 to low urgency. It iterates through all vehicles waiting to be charged, combining each vehicle's identifier with its corresponding urgency level value into an element, and placing all elements into a list. This list is then sorted from highest to lowest urgency level, using descending order of comparison. Vehicles with an urgency level of 3 are listed before vehicles with an urgency level of 2, and so on, from highest to lowest. If multiple vehicles have the same urgency level, they are sorted from lowest to highest based on the remaining battery power in their request data, with vehicles having lower remaining battery power listed first. If the remaining battery power is also the same, they are sorted from highest to lowest based on the desired charging amount. The sequence of vehicles waiting to be charged after sorting is the emergency vehicle queue. Vehicles at the front of the emergency vehicle queue have a higher degree of urgency.

[0043] The scheduling generation module initializes an empty scheduling instruction sequence to store the generated scheduling instructions. The module then enters a cyclic matching process, which continuously performs the following operations: It retrieves the vehicle at the head of the emergency vehicle queue as the current vehicle to be matched; it retrieves the charging pile node with the highest overall scheduling priority at the head of the charging pile priority queue as a candidate charging pile node. The scheduling generation module obtains the spatial distance between the current vehicle to be matched and the candidate charging pile node. This spatial distance is calculated using the Euclidean distance between the current location coordinates of the vehicle to be matched and the installation location coordinates of the corresponding charging pile of the candidate charging pile node. The scheduling generation module compares the calculated spatial distance with a preset distance threshold, which is set to 5 kilometers.

[0044] When the calculated spatial distance is less than or equal to a preset distance threshold, the scheduling generation module generates a scheduling instruction. This instruction includes the vehicle identifier of the vehicle to be matched and the node identifier of the candidate charging pile node, instructing the vehicle to be scheduled to the charging pile corresponding to the candidate charging pile node for charging. The scheduling generation module appends the generated scheduling instruction to the end of the scheduling instruction sequence. The scheduling generation module then permanently removes the candidate charging pile node from the charging pile priority queue, preventing other vehicles subsequently removed from the queue from being matched with the removed charging pile node.

[0045] When the calculated spatial distance is greater than a preset distance threshold, the scheduling generation module skips candidate charging pile nodes and does not generate scheduling instructions for the current vehicle to be matched and the candidate charging pile nodes. The scheduling generation module removes the skipped candidate charging pile nodes from the charging pile priority queue and places them in a temporary waiting queue for temporary storage. It then continues to remove the next charging pile node at the head of the charging pile priority queue as a new candidate charging pile node, recalculates the spatial distance between the current vehicle to be matched and the new candidate charging pile node, and compares the recalculated spatial distance with the preset distance threshold again. If the spatial distance of the new candidate charging pile node is still greater than the preset distance threshold, the new candidate charging pile node is moved to the temporary waiting queue, and the next charging pile node is removed from the charging pile priority queue for further attempts. If all charging pile nodes in the charging pile priority queue are empty and no charging pile node with a spatial distance less than or equal to the preset distance threshold is found, the matching attempt for the current vehicle to be matched is stopped, and the current vehicle to be matched is not allocated a charging pile in this round of scheduling. The scheduling generation module then puts all the charging pile nodes in the temporary waiting queue back into the charging pile priority queue in descending order of their original comprehensive scheduling priority.

[0046] Once a candidate charging pile node with a spatial distance less than or equal to a preset distance threshold is successfully found and matched in the charging pile priority queue, the scheduling generation module returns the remaining unmatched charging pile nodes from the temporary waiting queue to the head of the charging pile priority queue in descending order of their original comprehensive scheduling priority. The scheduling generation module then continues to retrieve the next vehicle to be charged from the vehicle emergency queue for matching.

[0047] The cyclic matching process terminates when either the vehicle emergency queue or the charging pile priority queue is empty. An empty vehicle emergency queue indicates that all vehicles awaiting charging have been processed, while an empty charging pile priority queue indicates that all available charging pile nodes in the charging pile group have been allocated. After the cyclic matching process terminates, the scheduling generation module outputs the generated scheduling instruction sequence to the external scheduling execution system. Each scheduling instruction in the sequence contains the vehicle identifier of the vehicle awaiting charging, the node identifier of the target charging pile, and the generation sequence number of the scheduling instruction.

[0048] In practice, the data acquisition module establishes a wired communication connection with the charging pile controller of each charging pile in the charging pile group, and a wireless communication connection with the on-board terminal of each vehicle waiting to be charged. The data acquisition module periodically collects real-time status data from each charging pile and request data from each vehicle waiting to be charged through these communication connections.

[0049] The data acquisition module collects real-time status data for each charging pile as follows: A status query command is sent to the charging pile controller of each charging pile in the charging pile group. The charging pile controller responds to the status query command and returns a real-time status data packet. The real-time status data packet contains six data fields: current operating voltage, current operating current, internal temperature, remaining available power, current queue length, and average charging time. The current operating voltage is a floating-point number in volts, representing the output voltage value of the charging pile at the time of data collection. The current operating current is a floating-point number in amperes, representing the output current value of the charging pile at the time of data collection. The internal temperature is a floating-point number in degrees Celsius, representing the temperature value of the internal power module of the charging pile. The remaining available power is a floating-point number in kilowatts, representing the maximum charging power that the charging pile can provide at the current time, calculated by the charging pile controller based on the difference between the rated power of the charging pile and the currently output power. The current queue length is an integer, representing the number of vehicles waiting in the queue at the charging station corresponding to this charging pile at the time of data collection. The current queue length is obtained by the charging pile controller through vehicle detection sensors at the entrance of the charging station. The average charging time is a floating-point number, expressed in minutes per vehicle, representing the average charging time for all charging tasks completed by this charging pile in the last 24 hours. The average charging time is calculated by the charging pile controller after recording the start and end times of each charging session.

[0050] The data acquisition module collects the request data for each vehicle waiting to be charged in the following way: it receives the charging request message actively reported by the vehicle's onboard terminal and parses the request data from the message. The request data contains four data fields: current location coordinates, remaining battery power, desired charging amount, and urgency level tag. The current location coordinates are a geographic coordinate pair containing longitude and latitude values, both of which are floating-point numbers in degrees, obtained by the vehicle's onboard satellite positioning module. The remaining battery power is a floating-point number in kilowatt-hours, representing the remaining battery power of the vehicle at the time the charging request is sent, read by the vehicle's battery management system. The desired charging amount is also a floating-point number in kilowatt-hours, representing the amount of electricity the user wants to receive, input by the user through the onboard terminal's interface or automatically estimated by the onboard terminal based on the distance to the navigation destination. The data type of the urgency level label is an enumeration value, and the possible values ​​of the enumeration value are high urgency, medium urgency, and low urgency. The urgency level label is calculated and set by the on-board terminal of the vehicle to be charged before sending the charging request message.

[0051] The specific implementation method for determining the urgency level label of the vehicle's onboard terminal is as follows: the onboard terminal obtains the remaining battery power of the vehicle and the navigation distance from the vehicle's current location to the nearest charging station cluster. The remaining battery power of the vehicle is read from the vehicle's battery management system and recorded as follows: The unit is kilowatt-hours. The navigation distance from the current location of the vehicle to be charged to the nearest charging station is obtained as follows: The vehicle terminal calls the map navigation application software, using the current location coordinates as the starting point and the geographical coordinates of the nearest charging station as the ending point, to request a route planning service. The path length returned by the route planning service is taken as the navigation distance, denoted as . The unit is kilometers. The on-board terminal calculates the remaining battery power. Calculate the driving range of the vehicle to be charged, denoted as . The unit is kilometers, and the driving range is calculated as follows: based on the remaining battery power... Divide by the average energy consumption per kilometer of the vehicle to be charged. The average energy consumption per kilometer is calculated by the on-board terminal based on the total energy consumption of the vehicle over the past 100 kilometers of driving, divided by the driving distance. The average energy consumption per kilometer is denoted as... The unit is kilowatt-hours per kilometer, which indicates the driving range. The in-vehicle terminal uses the ratio of navigation distance to remaining driving range as a battery risk factor. The formula for calculating the battery risk factor is: ; in, It represents the energy risk coefficient of the vehicle to be charged, and is a dimensionless ratio. This indicates the navigation distance from the current location of the vehicle to be charged to the nearest group of charging stations, in kilometers; This indicates the driving range of the vehicle to be charged, in kilometers. The remaining battery power of the vehicle to be charged Divide by average power consumption per kilometer To obtain, that is , The unit is kilowatt-hour. The unit is kilowatt-hours per kilometer. Electricity risk factor. The range of values ​​for is real numbers greater than zero. A value less than 1 indicates that the navigation distance is less than the driving range, and the vehicle to be charged has sufficient remaining power to reach the charging station cluster; when A value greater than or equal to 1 indicates that the navigation distance exceeds or equals the driving range, and there is a high risk that the vehicle waiting to be charged may not be able to reach the charging station cluster.

[0052] After calculating the battery risk coefficient, the vehicle-mounted terminal sorts and marks the coefficients. It acquires the battery risk coefficients of all vehicles requesting charging within the nearby area and sorts them in descending order. The top 30% of vehicles are marked as high urgency, the middle 40% as medium urgency, and the bottom 30% as low urgency. The division is based on the 30th and 70th percentiles of the sorted sequence, respectively. The 30th percentile corresponds to the position in the sorted sequence where the total number of vehicles is multiplied by 0.3 and rounded up, and the 70th percentile corresponds to the position in the sorted sequence where the total number of vehicles is multiplied by 0.7 and rounded up. Vehicles ranked from the 1st to the 30th percentile are assigned a high urgency label; vehicles ranked from the 30th percentile plus one to the 70th percentile are assigned a medium urgency label; and vehicles ranked from the 70th percentile plus one to the last are assigned a low urgency label. When the total number of vehicles requesting charging in the vicinity is less than 10, the three vehicles with the highest battery risk coefficient are marked as high urgency, the three vehicles with the lowest battery risk coefficient are marked as low urgency, and the remaining vehicles are marked as medium urgency. The onboard terminal writes the determined urgency label into the corresponding field of the charging request message and sends the charging request message to the data acquisition module via wireless communication.

[0053] See Figure 5 This invention provides an intelligent scheduling and management system for new energy vehicle charging piles, including a data acquisition module, a dynamic graphing module, a topology sampling module, a priority calculation module, and a scheduling generation module. The data acquisition module collects real-time status data of each charging pile in the charging pile group and request data from each vehicle waiting to be charged. The dynamic graphing module constructs a dynamic scheduling graph based on the real-time status data of each charging pile and the request data of each vehicle waiting to be charged. The dynamic scheduling graph includes charging pile nodes, vehicle nodes, and edges connecting the charging pile nodes and vehicle nodes. The topology sampling module performs hierarchical walk sampling on the dynamic scheduling graph to obtain the local topology feature vector of each charging pile node. The priority calculation module calculates the comprehensive scheduling priority of each charging pile node based on its local topology feature vector. The scheduling generation module generates a sequence of scheduling instructions based on the comprehensive scheduling priority of each charging pile node.

[0054] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent scheduling and management of charging piles for new energy vehicles, characterized in that, include: Collect real-time status data of each charging pile in the charging pile group and request data of each vehicle waiting to be charged. A dynamic scheduling graph is constructed based on the real-time status data of each charging pile and the request data of each vehicle to be charged. The dynamic scheduling graph includes charging pile nodes, vehicle nodes, and edges connecting the charging pile nodes and vehicle nodes. Hierarchical walk sampling is performed on the dynamic scheduling graph to obtain the local topological features of each charging pile node; The comprehensive scheduling priority of each charging pile node is calculated based on the local topology characteristics of each charging pile node. A sequence of scheduling instructions is generated based on the comprehensive scheduling priority of each charging pile node.

2. The intelligent scheduling and management method for new energy vehicle charging piles according to claim 1, characterized in that, The specific method for constructing a dynamic scheduling diagram based on the real-time status data of each charging pile and the request data of each vehicle waiting to be charged is as follows: Each charging pile is mapped to a charging pile node, and each vehicle to be charged is mapped to a vehicle node; For each vehicle node, calculate the spatial distance between the vehicle node and each charging pile node and the expected waiting time, wherein the expected waiting time is determined based on the current queue length and average charging time of the charging pile node. When the spatial distance is less than a preset distance threshold and the expected waiting time is less than a preset time threshold, an edge is established between the vehicle node and the charging pile node, and an initial weight is assigned to the edge. The initial weight is inversely proportional to both the spatial distance and the expected waiting time.

3. The intelligent scheduling and management method for new energy vehicle charging piles according to claim 2, characterized in that, The initial weights of the edges in the dynamic scheduling graph are updated dynamically over time. After each scheduling instruction is executed, the spatial distance and expected waiting time between the affected vehicle node and the charging pile node are recalculated and the initial weights of the corresponding edges are updated.

4. The intelligent scheduling and management method for new energy vehicle charging piles according to claim 2, characterized in that, The specific method for performing hierarchical walk sampling on the dynamic scheduling graph to obtain the local topological feature vector of each charging pile node is as follows: Each charging pile node is used as the starting node, and multiple walking step size levels are set, with each level corresponding to a walking step size range; Within each level, a random walk is performed starting from the initial node according to the initial weight of the edges. The charging pile nodes and vehicle nodes visited during the walk are recorded, and all the visited nodes are arranged in the order of visit to form the walk sequence for that level. The traversal sequences of all levels corresponding to the same charging pile node are concatenated to form a multi-level traversal path for that charging pile node. The multi-level traversal path of each charging pile node is uniquely encoded and transformed to obtain the local topological feature vector of that charging pile node.

5. The intelligent scheduling and management method for new energy vehicle charging piles according to claim 4, characterized in that, The specific method for calculating the comprehensive scheduling priority of each charging pile node based on the local topology feature vector of each charging pile node is as follows: The local topological feature vector of each charging pile node is input into a pre-trained priority regressor, which outputs an initial priority score. Simultaneously, the remaining available power and current fault flag bit in the real-time status data of the charging pile node are obtained. If the current fault flag bit indicates that there is a fault, the comprehensive scheduling priority of the charging pile node is set to zero. If there is no fault, the initial priority score of the charging pile node is multiplied by the normalized value of the remaining available power to obtain the overall scheduling priority of the charging pile node.

6. The intelligent scheduling and management method for new energy vehicle charging piles according to claim 5, characterized in that, The normalized value of the remaining available power is the ratio of the remaining available power of the charging pile node to the maximum remaining available power of all charging pile nodes in the charging pile group.

7. The intelligent scheduling and management method for new energy vehicle charging piles according to claim 5, characterized in that, The specific method for generating the scheduling instruction sequence based on the comprehensive scheduling priority of each charging pile node is as follows: All charging pile nodes are sorted from high to low according to the overall scheduling priority to obtain the charging pile priority queue; Each vehicle waiting to be charged is sorted from highest to lowest according to the urgency of its request data to obtain the vehicle emergency queue. One vehicle to be charged is taken out from the vehicle emergency queue in sequence, and the charging pile node with the highest overall scheduling priority is selected from the charging pile priority queue. A scheduling instruction is generated to schedule the vehicle to be charged to the charging pile node. Remove the matched charging pile node from the charging pile priority queue and continue processing the next vehicle to be charged until the vehicle emergency queue is empty or the charging pile priority queue is empty. Combine all scheduling instructions into a scheduling instruction sequence according to the generation order.

8. The intelligent scheduling and management method for new energy vehicle charging piles according to claim 7, characterized in that, When the spatial distance between a vehicle to be charged taken from the vehicle emergency queue and the charging pile node with the highest overall scheduling priority selected from the charging pile priority queue is greater than a preset distance threshold, the charging pile node is skipped and the next charging pile node with the highest overall scheduling priority is selected for matching. The real-time status data includes the current operating voltage, current operating current, internal temperature, remaining available power, current queue length, and average charging time for each charging pile. The request data includes the current location coordinates, remaining battery power, desired charging amount, and urgency level tag for each vehicle to be charged.

9. The intelligent scheduling and management method for new energy vehicle charging piles according to claim 4, characterized in that, The specific method for setting multiple walking step length levels is as follows: Determine the maximum shortest path length between all charging pile nodes and vehicle nodes in the dynamic scheduling graph; Divide the maximum and shortest path length by three and round up to obtain the first step length threshold. Multiply the maximum and shortest path length by two-thirds and round up to obtain the second step length threshold. The first level is defined as the range of walking step length from one to the first step length threshold, the second level is defined as the range of the first step length threshold plus one to the second step length threshold, and the third level is defined as the range of the second step length threshold plus one to the maximum shortest path length.

10. A smart scheduling and management system for new energy vehicle charging piles, characterized in that, include: The data acquisition module is used to collect real-time status data of each charging pile in the charging pile group and request data of each vehicle waiting to be charged. The dynamic graph construction module is used to construct a dynamic scheduling graph based on the real-time status data of each charging pile and the request data of each vehicle to be charged. The dynamic scheduling graph includes charging pile nodes, vehicle nodes and edges connecting charging pile nodes and vehicle nodes. The topology sampling module is used to perform hierarchical walk sampling on the dynamic scheduling graph to obtain the local topology feature vector of each charging pile node; The priority calculation module is used to calculate the comprehensive scheduling priority of each charging pile node based on the local topology feature vector of each charging pile node. The scheduling generation module is used to generate a sequence of scheduling instructions based on the comprehensive scheduling priority of each charging pile node.

Images