A heterogeneous unmanned platform cooperative charging scheduling method and system

Abstract

This application relates to a collaborative charging scheduling method and system for heterogeneous unmanned platforms. Based on a charging behavior model constructed from the charging data of each unmanned platform, a charging feature vector is determined for each platform, including charging urgency and charging duration. Based on these feature vectors, multiple unmanned platforms are divided into at least one cluster. A scheduling objective function is set to minimize the time-averaged error between the actual total load of the cluster and the grid reference load. Simultaneously, a first virtual queue representing the backlog of charging demand and a second virtual queue representing priority based on charging urgency are constructed. Based on the first and second virtual queues, the scheduling objective function, and the Lyapunov drift penalty algorithm, the charging scheduling strategy for the current scheduling slot is solved. Thus, this application achieves real-time tracking of the grid reference load without requiring precise prediction of future charging demand and load changes, while prioritizing the charging needs of unmanned platforms with high urgency.

Description

Technical Field

[0001] This application relates to the field of intelligent dispatching of power systems, and in particular to a method and system for collaborative charging dispatching of heterogeneous unmanned platforms. Background Technology

[0002] The world is currently undergoing a new round of technological revolution and industrial transformation, with unmanned and intelligent equipment technologies experiencing explosive growth. Unmanned platforms, as a disruptive new type of equipment integrating mechanization, informatization, and intelligence, possess significant advantages such as flexible use, high comprehensive combat effectiveness, and suitability for hazardous environments, and are being widely applied in both military and civilian fields. Various intelligent unmanned platforms have broken through the limitations of single dimensions, achieving increasing collaborative applications in multi-dimensional spaces including the air, ground, surface, and underwater. However, limited by current high-efficiency power supply technology, various unmanned platforms generally suffer from bottlenecks such as short endurance and the need for frequent recharging.

[0003] Although heterogeneous unmanned platforms are not yet as numerous as widespread electric vehicles, they serve as a special type of "high-density mobile payload," possessing strong mission coordination and spatiotemporal aggregation capabilities. Various unmanned platforms typically execute high-value missions in "swarms." The "pulsating" load surges generated by these large-scale swarms simultaneously accessing refueling, coupled with the stringent energy security requirements in complex environments such as the field and at sea, make research into their energy supply and scheduling strategies particularly urgent and of great strategic significance. Summary of the Invention

[0004] This application provides a method and system for collaborative charging scheduling of heterogeneous unmanned platforms, which can realize real-time tracking of the reference load of the power grid, while prioritizing the charging needs of unmanned platforms with high urgency.

[0005] In a first aspect, embodiments of this application provide a method for collaborative charging scheduling of heterogeneous unmanned platforms, the method comprising: Real-time acquisition of charging data and grid reference load from multiple arriving unmanned platforms; A charging behavior model is constructed based on the charging data of each unmanned platform to determine the charging feature vector of each unmanned platform; the charging feature vector includes charging urgency and charging duration. Based on the charging feature vectors of each unmanned platform, the multiple unmanned platforms are divided into at least one virtual scheduling cluster; The scheduling objective function is set with the goal of minimizing the time average error between the actual total load of the cluster and the reference load of the power grid. At the same time, a first virtual queue and a second virtual queue are constructed. The first virtual queue is used to represent the charging demand backlog status of each virtual scheduling cluster, and the second virtual queue is used to represent the priority constraint based on the charging urgency. Based on the first virtual queue, the second virtual queue, the scheduling objective function, and the Lyapunov drift penalty algorithm, the charging scheduling strategy for the current scheduling slot is solved.

[0006] Secondly, embodiments of this application also provide a heterogeneous unmanned platform collaborative charging scheduling system, the system comprising: The acquisition module is used to acquire charging data and grid reference load from multiple arriving unmanned platforms in real time. The determination module is used to determine the charging behavior model based on the charging data of each unmanned platform, and to determine the charging feature vector of each unmanned platform; the charging feature vector includes charging urgency and charging duration. The partitioning module is used to divide the multiple unmanned platforms into at least one virtual scheduling cluster based on the charging feature vectors of each unmanned platform. The construction module is used to set a scheduling objective function with the goal of minimizing the time average error between the actual total load of the cluster and the reference load of the power grid, and to construct a first virtual queue and a second virtual queue. The first virtual queue is used to represent the charging demand backlog status of each virtual scheduling cluster, and the second virtual queue is used to represent the priority constraints based on charging urgency. The solution module is used to solve the charging scheduling strategy for the current scheduling slot based on the first virtual queue, the second virtual queue, the scheduling objective function, and the Lyapunov drift penalty algorithm.

[0007] Thirdly, embodiments of this application also provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-described method.

[0008] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the above-described method.

[0009] This application provides a method and system for collaborative charging scheduling of heterogeneous unmanned platforms. The method includes: constructing a charging behavior model based on charging data from each unmanned platform; determining a charging feature vector for each unmanned platform, including charging urgency and charging duration; dividing multiple unmanned platforms into at least one cluster based on these feature vectors; setting a scheduling objective function with the goal of minimizing the time-averaged error between the actual total load of the cluster and the grid reference load; simultaneously constructing a first virtual queue representing the backlog of charging demand and a second virtual queue representing priority based on charging urgency; and solving for the charging scheduling strategy for the current scheduling slot based on the first virtual queue, the second virtual queue, the scheduling objective function, and the Lyapunov drift penalty algorithm. Thus, this application establishes a charging behavior model, dynamically aggregates and clusters large-scale unmanned platforms, and on this basis, designs an online scheduling algorithm using Lyapunov optimization theory. This allows for real-time tracking of the grid reference load without requiring precise prediction of future charging demand and load changes, while prioritizing the charging needs of unmanned platforms with high urgency. Attached Figure Description

[0010] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0011] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] One or more embodiments are illustrated by way of example with reference numerals in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0013] Figure 1 A flowchart illustrating a collaborative charging scheduling method for heterogeneous unmanned platforms provided in this application embodiment; Figure 2 This is one of the schematic diagrams illustrating the dynamic aggregation effect of heterogeneous unmanned platforms based on the K-means algorithm in the embodiments of this application; Figure 3 This is the second schematic diagram illustrating the dynamic aggregation effect of heterogeneous unmanned platforms based on the K-means algorithm in the embodiments of this application; Figure 4 This is a comparison diagram of the convergence process of the time-averaged error ratio evolving with time slots in the embodiments of this application; Figure 5This is one of the schematic diagrams illustrating the influence of key parameters on the time-averaged error ratio in the embodiments of this application; Figure 6 This is the second schematic diagram illustrating the impact of key parameters on the time-averaged error ratio in the embodiments of this application. Figure 7 This is one of the statistical charts showing the charging completion rate of different clusters in the embodiments of this application; Figure 8 This is the second statistical chart showing the charging completion rate of different clusters in the embodiments of this application. Figure 9 This is a schematic diagram of the structure of a heterogeneous unmanned platform collaborative charging scheduling system in an embodiment of this application; Figure 10 This is a schematic diagram of the structure of a computer device according to an embodiment of this application. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0015] The following disclosure provides numerous different embodiments or examples for implementing various structures of this application. To simplify the disclosure, specific examples of components and arrangements are described below. These are merely examples and are not intended to limit the scope of this application. Furthermore, reference numerals and / or letters may be repeated in different examples. Such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0016] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0017] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0018] It should also be further understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0019] As used in this specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrases "if determined" or "if [described condition or event] is detected" may be interpreted, depending on the context, as "once determined," "in response to determination," "once [described condition or event] is detected," or "in response to detection of [described condition or event]."

[0020] To address the technical challenges of handling randomness and balancing real-time performance with global performance in the charging scheduling of large-scale unmanned platforms in existing technologies, this application provides a collaborative charging scheduling method and system for heterogeneous unmanned platforms. This method enables real-time tracking of the grid reference load while prioritizing the charging needs of unmanned platforms with high urgency, thus balancing real-time performance with global performance.

[0021] Figure 1 This is a flowchart illustrating a collaborative charging scheduling method for heterogeneous unmanned platforms provided in an embodiment of this application. Figure 1 As shown, the heterogeneous unmanned platform collaborative charging scheduling method includes the following steps: S101: Real-time acquisition of charging data and grid reference load from multiple arriving unmanned platforms.

[0022] It should be noted that in certain scenarios (such as emergency rescue), large-scale heterogeneous unmanned platforms performing continuous tasks require coordinated charging scheduling at temporarily constructed charging base stations (e.g., containing dozens of charging ports, with total power supply capacity constrained by the grid's reference load). Scheduling time slots (e.g., 15 minutes) are pre-set. During peak hours (e.g., mission breaks 10:00-12:00, 14:00-16:00), unmanned platforms return to base for charging in a concentrated manner, while a small number of unmanned platforms take turns charging during off-peak hours. Priority must be given to ensuring the charging needs of unmanned platforms performing urgent tasks (high urgency), while simultaneously mitigating grid load fluctuations.

[0023] In practice, charging base stations deploy data acquisition modules, enabling real-time acquisition of charging data from any unmanned platform upon its arrival. This charging data includes energy storage parameters and status data, such as battery capacity, rated charging power, state of charge upon arrival, and target state of charge. The grid reference load is dynamically generated and distributed by the regional grid dispatch center based on the real-time operating status of the grid. The charging base station establishes a connection with the grid dispatch center and can receive reference load commands from the dispatch center in real time.

[0024] Among them, the power grid dispatch center, based on the load monitoring data of the entire network, distinguishes between peak periods (such as daytime peak electricity consumption) and off-peak periods (such as nighttime off-peak electricity consumption), and sets different power grid reference loads P for each. re The f(t) threshold is lower during peak hours (e.g., 5kW) to limit charging load and alleviate grid pressure; and higher during off-peak hours (e.g., 10kW) to encourage charging and achieve peak shaving and valley filling. After receiving instructions, the charging base station converts the grid reference load into a constraint benchmark for charging scheduling to ensure that the cluster charging load does not exceed the grid control requirements.

[0025] Here, the unmanned platforms described in this application embodiment include, but are not limited to, unmanned aerial vehicles (UAVs), unmanned ground vehicles (UGVs), and unmanned surface vessels (USVs). The heterogeneous unmanned platform clusters in this application embodiment can adopt a heterogeneous form. Specifically, a heterogeneous unmanned platform cluster refers to a mixed formation composed of unmanned platforms equipped with different functional modules and possessing differentiated performance parameters (such as battery capacity, rated charging power, and range requirements). Typical examples include heterogeneous clusters of the same type composed of different models of UAVs, or hybrid heterogeneous clusters composed of UAVs and cross-type equipment such as UGVs and USVs. Among these, the battery characteristics, charging interfaces, and mission urgency of each heterogeneous unmanned platform in the cluster differ significantly.

[0026] It should be noted that in the charging scheduling scenario, the scheduling of heterogeneous drone swarms is particularly challenging: First, the charging power and demand duration of each drone model vary greatly, which can easily cause load fluctuations; second, the urgency of drone tasks differs, making it difficult to unify charging priorities; and third, the swarm size is large and dynamically enters and exits, which traditional static scheduling algorithms cannot adapt to, requiring a balance between real-time performance and grid load stability, thus placing higher demands on the flexibility and accuracy of scheduling strategies.

[0027] S102: A charging behavior model is constructed based on the charging data of each unmanned platform to determine the charging feature vector of each unmanned platform.

[0028] Here, the charging feature vector includes at least two feature dimensions: charging urgency and charging duration. Charging urgency reflects the urgency with which the unmanned platform needs to complete its charging within a limited dwell time, and is a key indicator reflecting the user's level of urgency.

[0029] In practical implementation, a charging behavior model for unmanned equipment that considers random characteristics is pre-constructed. That is, the process of "modeling the arrival and dwelling behavior of unmanned platforms" is initiated first. This charging behavior model is a model that characterizes the charging-related behavioral patterns of unmanned platforms based on their charging data. Then, the charging feature vectors of each unmanned platform are extracted based on the constructed charging behavior model.

[0030] S103: Based on the charging feature vectors of each unmanned platform, the multiple unmanned platforms are divided into at least one virtual scheduling cluster.

[0031] In practical implementation, dynamic aggregation based on charging urgency involves dynamically grouping unmanned platforms according to their charging urgency, forming several levels with different urgency levels, and formulating and simplifying scheduling strategies accordingly. In other words, feature clustering integrates disordered and scattered individual unmanned platforms into ordered scheduling units (virtual scheduling clusters), thereby reducing the scheduling dimensionality and the complexity of subsequent optimization solutions, effectively addressing the problem of insufficient real-time computation in large-scale cluster scenarios. This application does not specify a particular clustering algorithm; any algorithm that can cluster a large number of unmanned platforms based on their charging feature vectors is acceptable. For example, K-means clustering, density-based spatial clustering of applications with noise (DBSCAN), hierarchical clustering, etc., can be used.

[0032] Here, considering that the charging feature vector contains two core dimensions, charging urgency and charging duration, which directly determine the charging priority and resource consumption requirements of the equipment, clustering is reasonable and necessary. Therefore, the charging feature vector can be used for clustering.

[0033] Among them, the virtual scheduling cluster is a logical scheduling unit formed by clustering charging feature vectors, rather than a physical equipment grouping. Its core features are as follows: (1) Virtuality: The division of the cluster exists only at the scheduling algorithm level and does not change the physical location or affiliation of the unmanned platform; (2) Dynamism: Each scheduling time slot will update the composition and cluster center of the cluster according to the status of newly arrived unmanned platforms and equipment that has left, ensuring that the cluster division is always adapted to the real-time charging scenario; (3) Scheduling unit attributes: Subsequent charging decisions (charging / not charging) and power allocation are all executed on the basis of the virtual scheduling cluster, rather than for a single piece of equipment, which greatly reduces the scheduling complexity of large-scale clusters.

[0034] S104: Set a scheduling objective function with the goal of minimizing the time average error between the actual total load of the cluster and the reference load of the power grid, and simultaneously construct a first virtual queue and a second virtual queue.

[0035] The first virtual queue is used to represent the charging demand backlog status of each virtual scheduling cluster, and the second virtual queue is used to represent the priority constraint based on charging urgency.

[0036] In its implementation, this application's embodiments take into account multiple objective constraints, including "meeting charging demand," "prioritizing high-urgency equipment," and "grid load tracking." It achieves automatic trade-offs among these objectives through a dual-virtual queue and scheduling objective function mechanism. Specifically, the scheduling objective function is set with the goal of "minimizing the time-averaged error between the actual total load of the cluster and the grid reference load." Simultaneously, dual virtual queues are constructed. The first virtual queue represents the backlog of charging demand in each virtual scheduling cluster, while the second virtual queue represents priority constraints based on charging urgency. The overall aim is to balance grid-side stability (achieving peak shaving and valley filling) with equipment-side demand (ensuring priority charging for high-urgency equipment). Ultimately, the scheduling objective function clarifies the optimization direction, and the dual queues dynamically regulate the charging rhythm, achieving the effect of accurately tracking the grid's charging load requirements while ensuring the orderly guarantee of equipment charging demand.

[0037] Here, a dynamic scheduling model based on virtual queues is constructed. The process begins with "demand queue construction," preparing a virtual queue to represent the backlog of charging demands for each heterogeneous unmanned platform cluster. The queue status is updated by calculating the difference between newly arriving demands and the current charging amount to ensure overall charging demand is met. Next, "urgency queue construction" is performed, preparing a virtual queue with a penalty function. Through a penalty mechanism negatively correlated with charging urgency, high-urgency heterogeneous unmanned platform clusters receive higher backlog values, thus gaining priority in scheduling. The "objective function setting" establishes the optimization objective of minimizing the time-averaged error between the total system cluster load and the grid reference load. Finally, it checks if the virtual queue parameters have been initialized; if so, the online optimization solution phase begins.

[0038] S105: Based on the first virtual queue, the second virtual queue, the scheduling objective function, and the Lyapunov drift penalty algorithm, solve the charging scheduling strategy for the current scheduling slot.

[0039] In practical implementation, relying on the previously constructed scheduling objective function and dual virtual queue constraints, the long-term load tracking optimization problem is transformed into an instantaneous optimization problem for the current scheduling slot using the Lyapunov drift penalty algorithm. Specifically, the first virtual queue provides the basis for the charging demand backlog status, the second virtual queue clarifies priority constraints, and the scheduling objective function locks in the load tracking direction. By calculating the Lyapunov drift term of the queue state and combining it with the weighted objective function term to construct the drift penalty term, minimizing the upper bound of this penalty term yields the charging decision (charge / not charge, power allocation) for each virtual scheduling cluster in the current time slot, integrating them to form the final charging scheduling strategy. This avoids the need for precise prediction of future charging demand and grid load, achieving real-time dynamic scheduling. It can balance stable grid load tracking with priority charging of high-urgency equipment, ensuring the real-time performance, reliability, and optimization of the scheduling strategy.

[0040] In one possible implementation, the charging behavior model is used to characterize the arrival process and dwell time patterns of the unmanned platform; the charging behavior model constructed based on the charging data of each unmanned platform in S102, which determines the charging feature vector of each unmanned platform, includes the following steps: Step 1021: Obtain the dwell time of each unmanned platform through the charging behavior model.

[0041] Among them, the dwell time refers to the total time from when the unmanned platform arrives at the charging base station until it completes charging and leaves. It is a key parameter for calculating the charging urgency (charging urgency is directly related to the dwell time; the shorter the dwell time and the longer the charging time, the higher the urgency).

[0042] In practical implementation, the charging behavior model encompasses the arrival process and dwell time patterns of the unmanned platform. Dwell time can be obtained in various ways. Specifically, it can be obtained by combining the unmanned platform's task scheduling information (such as the latest departure time of the next task), subtracting the current scheduling slot and the estimated departure preparation time, to arrive at the maximum possible dwell time; or by using historical charging data of the same type of equipment to calculate the charging time from the current state of charge to the target state of charge, and then adding the time spent on routine equipment checks and data interactions to comprehensively determine the dwell time. By integrating the above-mentioned relevant data and historical patterns through the charging behavior model, the dwell time of each piece of equipment can be accurately output, providing a core basis for subsequent scheduling.

[0043] The following is an example of a method for obtaining dwell time, namely, the charging behavior model includes a non-homogeneous Poisson process model characterizing the arrival process of the unmanned platform and a conditional gamma distribution model describing the dwell time of the unmanned platform; step 1021, obtaining the dwell time of each unmanned platform through the charging behavior model, includes the following steps: Step 1021a: Determine the number of newly arriving unmanned platforms in the current time slot based on the time-varying arrival rate function predefined in the non-homogeneous Poisson process model; the time-varying arrival rate function sets different arrival rates according to the time-slot characteristics of the actual scenario.

[0044] In practical implementation, a non-homogeneous Poisson process (NHPP) is used to characterize the arrival process of the unmanned platform, and a time-varying arrival rate function λ(t) is used to describe the arrival pattern of the unmanned platform in different time periods. Different arrival rates are set according to the time characteristics of the actual scenario; for example, 500 vehicles / hour during peak periods (duty-concentrated periods) and 200 vehicles / hour during off-peak periods. The expected number of arrivals for any scheduling time slot t is dynamically output through λ(t). Compared with the traditional homogeneous Poisson process, the non-homogeneous Poisson process can reflect the time-varying characteristics of the arrival rate, which is more in line with the arrival patterns of heterogeneous unmanned platforms driven by missions (such as concentrated deployments in military missions and emergency rescue).

[0045] Step 1021b: Determine the dwell time of each newly arrived unmanned platform based on the shape and scale parameters of the conditional gamma distribution model corresponding to the current time slot.

[0046] In practice, the conditional gamma distribution is used to describe the dwell time of the unmanned platform. Based on the distribution pattern of arrival scheduling slots, the shape parameter α and scale parameter θ of the conditional gamma distribution dynamically change with the arrival time of the unmanned platform to characterize the differences in the dwell time of unmanned platforms arriving at different times. For example, α=2 and θ=1 are set during the day, and α=5 and θ=2 are set at night. For each arriving unmanned platform k, the corresponding α and θ parameters are queried according to its arrival time t (scheduling slot) to generate the dwell time of that unmanned platform. The duration of the stay is affected by the type of mission and environmental conditions (such as short missions during the day and long-endurance preparations at night).

[0047] Furthermore, based on the shape and scale parameters corresponding to the conditional gamma distribution model of the current time slot, the dwell time of each newly arrived unmanned platform is determined, specifically including: Based on the time period category to which the current time slot belongs, retrieve the shape parameter and scale parameter associated with the time period category from the pre-built table of correspondence between time period category and gamma distribution parameter; Using the shape and scale parameters associated with the time period category, a conditional gamma distribution model is constructed. The conditional gamma distribution model is used to characterize the stochastic characteristics of the dwell time of the unmanned platform under the time period category. For each newly arrived unmanned platform in the current time slot, a random sampling operation is performed from the conditional gamma distribution model corresponding to the time slot category to generate the initial dwell time of the unmanned platform; Based on the task type and preset adjustment rules of each unmanned platform, the initial dwell time of the unmanned platform is adjusted to obtain the dwell time of each unmanned platform.

[0048] During the operation of heterogeneous unmanned platforms, the operational intensity, task type distribution, and environmental constraints of the equipment vary significantly across different time periods, resulting in a markedly random characteristic in their dwell behavior at charging stations or relay points. Using probability distributions with fixed parameters or deterministic duration assumptions makes it difficult to accurately describe the dwell patterns in real-world scenarios, potentially leading to insufficient charging resource allocation or idle scheduling. Therefore, this embodiment introduces a gamma distribution model based on time-period category conditionalization, which can more precisely generate dwell times that conform to the current operational context.

[0049] Specifically, a day or typical task cycle can be divided into several time period categories, each corresponding to a typical operating state, such as a high-concurrency job period, a low-load standby period, or a nighttime maintenance window. Based on historical operating data, the shape and scale parameters of the gamma distribution are fitted to each time period category, and a correspondence table between the time period category and the two parameters is established. This correspondence table is configured and fixed in the operating environment before deployment.

[0050] Upon entering any time slot, the time period category to which the time slot belongs is identified, and the shape and scale parameters associated with that time period category are retrieved from the corresponding relationship table. A gamma distribution instance is constructed using the retrieved shape and scale parameters. This gamma distribution instance serves as the conditional gamma distribution model, and its probability density function is entirely determined by the current time period category. It can reflect the distribution characteristics of the dwell time of the unmanned platform under that time period due to the combined effects of factors such as mission rhythm, operational procedures, or external interference.

[0051] For each newly arrived unmanned platform in the current time slot, an independent random sampling is performed from the conditional gamma distribution model to generate a positive real number as the initial dwell time of the unmanned platform. This sampled value follows the empirical statistical law of the current time period, avoiding the bias caused by the global uniform assumption.

[0052] Then, by combining the task type of each unmanned platform with the preset adjustment rules, the initial dwell time is adjusted. Task types include reconnaissance, transportation, inspection, and relay communication, each with different dwell behavior constraints. Preset adjustment rules can include setting a maximum dwell time limit for time-sensitive tasks, increasing buffer time for heavy-load transportation tasks based on cargo class, or extending the waiting window for equipment requiring collaborative operations. By integrating task semantic information into the dwell time generation process, the final output dwell time conforms to both macro-level time-segment patterns and meets the micro-level needs of individual tasks.

[0053] By using conditional gamma distribution modeling driven by time period categories, we have achieved a dynamic and probabilistic characterization of the dwell behavior of unmanned platforms. By combining task type for posterior adjustment, we have improved the scenario adaptability and scheduling feasibility of dwell time generation. This effectively supports the accurate execution of subsequent charging resource allocation, path replanning, and multi-agent collaborative decision-making, and avoids resource conflicts or service delays caused by distorted dwell time estimation.

[0054] Step 1022: For any unmanned platform, calculate the corresponding charging time based on the battery capacity, rated charging power, state of charge at arrival, and target state of charge in the charging data of the unmanned platform.

[0055] Here, the charging time directly determines the charging resource requirements of the unmanned platform. It is necessary to calculate accurately based on the battery characteristics (capacity differences) and mission requirements (target SOC) of different unmanned platforms to avoid insufficient or wasted resource allocation.

[0056] The charging time can be calculated using a formula. ;in, For unmanned platforms Battery capacity, For unmanned platforms The state of charge upon arrival, For unmanned platforms The target state of charge, For unmanned platforms The rated charging power.

[0057] Step 1023: Determine the charging urgency of the unmanned platform based on its dwell time and charging time.

[0058] Here, we define unmanned platforms. charging urgency Charging time Duration of stay The ratio, i.e. Among them, the urgency of charging. The larger the value, the more advanced the unmanned platform. The more electricity that needs to be replenished within a limited stay time, the higher the urgency of charging, and the more urgently it needs to be prioritized for dispatch.

[0059] Step 1024: Using the charging urgency and charging time of the unmanned platform as core dimensions, determine the charging feature vector of the unmanned platform.

[0060] Here, the urgency of charging is taken into account. and charging time For unmanned platforms Constructing the charging feature vector: Among them, the charging feature vector space reflects both the user's urgency and the scale of their charging needs.

[0061] In one possible implementation, the step S103, which involves dividing the multiple unmanned platforms into at least one virtual scheduling cluster based on the charging feature vectors of each unmanned platform, includes the following steps: Step 1031: Determine the number of clusters based on the scale of the heterogeneous unmanned platform, and select multiple initial cluster centers using the K-means algorithm.

[0062] In practice, the number of clusters J is set according to the scale of the heterogeneous unmanned platforms (for example, J=20, which can be dynamically adjusted according to the total number (scale) of unmanned platforms), and the initial cluster center (i.e., the initial centroid) is selected using the K-means algorithm. To avoid clustering getting trapped in local optima, the number of clusters J needs to balance computational complexity and intra-cluster homogeneity. If J is too small, the intra-cluster differences will be large, resulting in low scheduling accuracy; if J is too large, the computational load will increase, and real-time performance will decrease.

[0063] Step 1032: Using the charging feature vector of each unmanned platform as input, the K-means algorithm is used to minimize the sum of squared distances between the charging feature vector of each unmanned platform in the cluster and the corresponding cluster center, thereby dividing the multiple unmanned platforms into at least one virtual scheduling cluster.

[0064] In practical implementation, the charging feature vector of each unmanned platform is used. As input, the K-means algorithm is used to minimize the sum of squared distances between the charging feature vectors (intra-cluster samples) of each unmanned platform in the cluster and the corresponding cluster center. The optimization objective is... ,in, Represents the number of clusters. For multiple unmanned platforms with similar characteristics The set of unmanned platforms that make up the j-th cluster, Let the j-th cluster be the cluster center, and divide all unmanned platforms into J clusters. The core of clustering is "homogeneity within clusters and differentiation between clusters", so that the urgency and charging time characteristics of unmanned platforms within the same cluster are similar, and a unified scheduling strategy can be adopted to reduce the number of scheduling decision variables (from N unmanned platforms to J clusters).

[0065] Step 1033: In each scheduling time slot, for any newly arrived unmanned platform, it is assigned to the nearest cluster center according to the charging feature vector corresponding to the unmanned platform. At the same time, unmanned platforms that have left the charging base station are removed, the cluster centers are recalculated, and the composition of the virtual scheduling cluster is updated.

[0066] In practice, during each scheduling slot (e.g., 15 minutes per slot), for a newly arriving unmanned platform, the distance between its charging feature vector and all current cluster centers is calculated, and it is assigned to the nearest cluster. Simultaneously, cluster centers that have left the charging base station are removed, and the cluster centers of each cluster are recalculated to update the cluster composition. This dynamic updating ensures that the clustering results adapt to the real-time inflow and outflow of unmanned platforms, avoiding scheduling lag caused by static clustering (e.g., newly added high-urgency unmanned platforms can be quickly assigned to the corresponding cluster and given priority scheduling).

[0067] It should be noted that this application is based on K-means dynamic aggregation based on charging urgency. The process begins with "virtual scheduling cluster initialization," which sets the number of clusters and prepares initial cluster centers. The "K-means algorithm" minimizes the distance between samples within a cluster and the cluster center, dividing the massive number of heterogeneous unmanned platforms into several virtual scheduling clusters based on charging urgency and charging duration characteristics. "Dynamic aggregation update" is then performed, where, in each scheduling time slot, newly arriving and departing unmanned platforms are assigned in real-time to the nearest cluster center based on their charging feature vectors, thus updating the cluster composition.

[0068] Figure 2 This is one of the schematic diagrams illustrating the dynamic aggregation effect of heterogeneous unmanned platforms based on the K-means algorithm in the embodiments of this application. Specifically, Figure 2 This demonstrates the dynamic clustering distribution effect of heterogeneous unmanned platforms in the feature space of "charging urgency - charging duration". Among them, Figure 2 In the demonstration of the dynamic aggregation effect of heterogeneous unmanned platforms based on the K-means algorithm, the horizontal axis represents the normalized charging urgency, and the vertical axis represents the normalized charging duration. These two dimensions together constitute the scheduling feature space of the unmanned platforms. Different colored scatter points in the figure represent individual heterogeneous unmanned platforms assigned to different virtual scheduling clusters; scatter points of the same color belong to the same virtual scheduling cluster. The black "×" symbols in the figure represent the centers of each cluster, reflecting the average feature level of the unmanned platform group within that virtual scheduling cluster.

[0069] Figure 3 This is the second schematic diagram illustrating the dynamic aggregation effect of heterogeneous unmanned platforms based on the K-means algorithm in this application embodiment; specifically, Figure 3 The statistics on the number of heterogeneous unmanned platforms in each virtual scheduling cluster after clustering are displayed. Figure 3The distribution of heterogeneous unmanned platforms within each virtual scheduling cluster was statistically analyzed. The cluster index on the horizontal axis corresponds to the number of the J virtual scheduling clusters generated by the algorithm (J=20 in the figure), representing equipment groups with different charging urgency and charging demand characteristics. The vertical axis represents the total number of heterogeneous unmanned platforms (unmanned equipment) dynamically aggregated into the j-th virtual scheduling cluster in the current statistical scheduling slot. This distribution intuitively reflects the composition ratio of various charging demands within the current system. For example, virtual scheduling cluster j=18 includes 143 (number of clusters) unmanned platforms.

[0070] Further, the step of updating the virtual scheduling cluster composition in each scheduling time slot includes: for any newly arrived unmanned platform in each scheduling time slot, calculating the distance between the charging feature vector of the unmanned platform and all current cluster centers, assigning the unmanned platform to the nearest virtual scheduling cluster, and simultaneously identifying and removing unmanned platforms that have left the charging base station; based on the updated set of unmanned platforms within the cluster, recalculating the cluster centers of each virtual scheduling cluster to obtain the updated virtual scheduling cluster; determining whether the updated virtual scheduling cluster meets the convergence condition or reaches the preset maximum number of iterations; if not, returning to the step of calculating the distance between the newly arrived unmanned platform and each cluster center, and continuing to iteratively update the virtual scheduling cluster; if satisfied, then the updated virtual scheduling cluster is the virtual scheduling cluster of the current scheduling time slot.

[0071] In practice, the clustering results (updated virtual scheduling clusters) are checked to see if they meet the convergence condition (e.g., the change in cluster centers is less than a set threshold) or the maximum number of iterations has been reached. If not, the cluster distances are recalculated and the iteration continues. If they meet the condition, the aggregated J virtual scheduling clusters are output. The total power of each cluster is determined, and the cluster is used as the smallest scheduling unit to proceed to the next step.

[0072] In one possible implementation, the step of setting the scheduling objective function in S104 includes: calculating the actual total load of the cluster in each scheduling time slot within the total number of scheduling time slots for cluster scheduling; the actual total load of the cluster is the sum of the charging power of all virtual scheduling clusters; calculating the difference between the actual total load of the cluster and the grid reference load in each scheduling time slot and squaring it; taking the expected value of the squared differences of all scheduling time slots and summing them; dividing the sum by the total number of scheduling time slots; and taking the limit when the total number of scheduling time slots approaches infinity to obtain the scheduling objective function.

[0073] Here, we define the objective function for optimizing scheduling: to minimize the actual total load on the cluster. With grid reference load The objective function expression is: (The time average error between the two points is taken as the objective.) Where T is the total number of scheduling slots in the cluster; The actual total cluster load for scheduling time slot t, j represents the index of the virtual scheduling cluster, which is used as the index variable for the summation operation; J represents the total number of virtual scheduling clusters, that is, the total number of scheduling units obtained after clustering. Let represent the charging decision variable for the j-th virtual scheduling cluster in scheduling slot t, and let take the value of a binary discrete value. When When, it means that the j-th virtual scheduling cluster is in a charging state in scheduling time slot t; when When, it means that the j-th virtual scheduling cluster does not charge in scheduling time slot t; The load (actual charging amount) of the j-th virtual scheduling cluster. This is the reference load for the power grid.

[0074] In one possible implementation, the step of constructing the first virtual queue and the second virtual queue in S104 includes: For any virtual scheduling cluster, a first virtual queue is constructed; based on the charging demand backlog status of any virtual scheduling cluster in the current scheduling time slot, the actual charging amount in the current scheduling time slot, and the charging demand of the newly arrived unmanned platform in the current scheduling time slot, the charging demand backlog status of the virtual scheduling cluster in the next scheduling time slot is determined, and the first virtual queue is updated.

[0075] In practice, a first virtual queue (demand queue) is constructed. This queue is used to characterize the charging demand backlog status of each unmanned platform cluster. Its queue status update depends on the difference between newly arriving charging demands and the current scheduling slot's charging capacity. Maintaining the stability of this queue satisfies the overall charging demand. The update formula for the first virtual queue is: ;in, The virtual scheduling cluster representing the newly arriving unmanned platform in the current scheduling time slot t. The charging needs, The actual charging amount in the current scheduling time slot t. The charging demand backlog state of the virtual scheduling cluster for the next scheduling time slot t+1; The charging demand backlog state of the virtual scheduling cluster in the current scheduling time slot t; This indicates that a non-negative value is being used. By keeping the queue stable (i.e., the queue backlog does not grow indefinitely), the overall charging demand is ensured to be met.

[0076] For any virtual scheduling cluster, a second virtual queue is constructed; based on the urgency queue status of the virtual scheduling cluster in the current scheduling time slot, the charging decision variables of the current scheduling time slot, the penalty function for negative correlation of charging urgency, and the indicator function, the urgency queue status of the virtual scheduling cluster in the next scheduling time slot is determined, and the second virtual queue is updated; the indicator function is determined according to the charging demand backlog status of the first virtual queue of the virtual scheduling cluster.

[0077] In practical implementation, a second virtual queue is constructed to represent priority constraints based on urgency. This queue introduces a penalty function negatively correlated with charging urgency, making it easier for high-urgency unmanned platform clusters to obtain scheduling priority. Specifically, a second virtual queue (urgency queue) is constructed. The queue created This is used to characterize priority constraints based on urgency, quantifying the degree to which high-urgency clusters are "delayed in gratification." The second virtual queue state update formula is: ,in, The urgency queue state of the virtual scheduling cluster for the next scheduling time slot t+1; The urgency queue state of the virtual scheduling cluster in the current scheduling time slot t. The charging decision variable (0 or 1) for the j-th cluster. A monotonically decreasing penalty function is set (the higher the cluster's urgency, the more...). (The larger the value) For indicator functions ( (Take 1 if it's urgent, otherwise take 0). This queue uses a penalty mechanism negatively correlated with charging urgency to ensure that high-urgency clusters... It is easier to accumulate, thus gaining higher priority in subsequent scheduling.

[0078] Here, we set up a queue. and The initial value (usually set to 0) confirms the penalty function. Specific forms (such as) ),in (where k is the average urgency of the j-th cluster and k is the adjustment coefficient). After initialization, the online optimization solution phase begins.

[0079] In one possible implementation, the step S105, which involves solving the charging scheduling strategy for the current scheduling slot based on the first virtual queue, the second virtual queue, the scheduling objective function, and the Lyapunov drift penalty algorithm, includes the following steps: Step 1051: Construct a Lyapunov function based on the first and second virtual queue states of all virtual scheduling clusters; the Lyapunov function is used to characterize the combined congestion degree of charging demand backlog and urgency delay of the cluster under the current scheduling time slot.

[0080] Here, the embodiments of this application transform the long-term optimization problem into a time-slot-by-time real-time optimization problem. Without the need to predict future information, the optimal charging decision for each time slot is obtained through an efficient solution algorithm, thus meeting the real-time requirements of large-scale systems.

[0081] In practical implementation, the Lyapunov function is defined. This function characterizes the combined congestion level of the cluster's charging demand backlog and urgency delay under the current scheduling slot. The larger the queue backlog, the higher the congestion level. The higher the value, the more the function is based on the demand queue (first virtual queue). ) and urgency queue (second virtual queue) )set up.

[0082] Specifically, first, the total number of clusters is determined to be J, and then the demand queue state of each cluster from the 1st cluster to the Jth cluster is calculated. The squared value of the urgency queue state for each cluster. The square value; then the corresponding square value for each cluster. Square value and The squared values ​​are summed to obtain the sum of the squared demand queue and the squared urgency queue for each of the J clusters; these J sums are then aggregated and summed again, and finally, the total sum is multiplied by 1 / 2 to obtain the Lyapunov function. The result is that this function is used to characterize the overall level of congestion.

[0083] Step 1052: Calculate the expected difference between the Lyapunov function values ​​of the current scheduling slot and the next scheduling slot to obtain the Lyapunov drift term.

[0084] Here, the Lyapunov drift term is defined as That is, the expected difference of the Lyapunov function values ​​of adjacent time slots.

[0085] Specifically, the Lyapunov function value for scheduling time slot t+1 is first calculated. Lyapunov function value with respect to scheduling time slot t The difference reflects the change in the Lyapunov function between adjacent time slots; then, the set of all cluster demand queue states in scheduling time slot t is used. (Include The set of all cluster urgency queue states in scheduling slot t. (Include Given the condition, calculate the expected value of the above difference. Ultimately, Lyapunov drift was achieved. Its core is to characterize the expected change in the congestion level of adjacent time slots given that the current queue state is known.

[0086] Step 1053: Add the Lyapunov drift term to the weighted scheduling objective function term to construct the Lyapunov drift penalty term; the weighted scheduling objective function term is obtained by the expected product of the control parameters and the squared errors of the actual total load of the current scheduling time slot cluster and the grid reference load.

[0087] Here, the Lyapunov drift penalty is constructed. W is a control parameter (weight) used to balance "demand satisfaction" and "load tracking error" (the larger W is, the more accurate the cluster's tracking of the grid reference load is); finally, the long-term optimization problem is transformed into a problem of minimizing the upper bound of the drift penalty term in each time slot, thereby achieving time correlation decoupling.

[0088] Specifically, the Lyapunov drift is first calculated. (i.e., the set of all cluster demand queue states in scheduling time slot t) The set of all cluster urgency queue states in scheduling time slot t Given the condition, the expected difference of the Lyapunov function values ​​of adjacent time slots); then calculate the actual total system load of scheduling time slot t. With reference load command value of the power grid The difference, after squaring the difference, is... and Calculate the expected value E based on the given conditions; then multiply this expected value by the control parameter (weight) W to obtain the weighted expected value of the load tracking error; then perform Lyapunov drift... Adding this to the weighted expectation term, the Lyapunov drift penalty term is finally constructed.

[0089] Step 1054: Transform the long-term optimization objective into a time-slot optimization problem that minimizes the upper bound of the Lyapunov drift penalty term in the current scheduling time slot. By solving the time-slot optimization problem, generate the charging scheduling strategy for the current scheduling time slot.

[0090] Here, this embodiment utilizes Lyapunov optimization theory to transform the long-term average error minimization problem into a real-time optimization problem for each scheduling slot t. Specifically, this problem aims to maximize the difference between the queue backlog release benefit and the reference load tracking error penalty. That is, in each scheduling slot t, the following optimization problem is solved to determine the charging decision.

[0091] Specifically, the online rolling optimization solution is based on Lyapunov shift-penalty. The process begins with a "Lyapunov optimization problem transformation," constructing a Lyapunov function to represent the system congestion level. The long-term reference load tracking problem is transformed into a minimization of the upper bound for each time slot using a "drift-penalty" technique. Next, a "relaxed linear programming solution" is performed, relaxing the complex 0-1 integer programming charging decision variables into continuous variables. A convex optimization method is then used to find the optimal strategy for the current time slot. A "threshold recovery mechanism" restores the continuous solution to discrete charging control commands (charging or not charging). The system then checks whether the scheduling command for the current time slot has been issued; if not, it recalculates. If yes, all queue states are updated to proceed to the next time slot, achieving real-time rolling scheduling without future prediction.

[0092] Figure 4 This is a comparison diagram of the convergence process of the time-averaged error ratio evolving with time slots in the embodiments of this application; Figure 4 This paper presents a comparison of the errors of the proposed Lyapunov optimization-based dynamic scheduling method (Proposed-LOA) with those of the existing uncoordinated charging strategy (UC) and static optimization-based centralized scheduling strategy (SO) in tracking the grid reference load.

[0093] Specifically, Figure 4 The evolution and convergence of the system's time-averaged error ratio with scheduling time slots under three different strategies are illustrated. The specific meanings of the three curves are as follows: Curve 1 (Proposed-LOA) represents the dynamic scheduling method based on Lyapunov optimization proposed in this application (i.e., the heterogeneous unmanned platform cooperative charging scheduling method); Curve 2 (Benchmark1:UC) represents the disordered charging strategy, i.e., the heterogeneous unmanned platform charges at maximum power upon access; Curve 3 (Benchmark2:SO) represents the centralized scheduling strategy based on static optimization, which depends on intraday forecast data.

[0094] Experimental results show that the Proposed-LOA algorithm (red line) proposed in this application has the lowest time-averaged error ratio (approximately 0.055), significantly outperforming the UC and SO strategies. This is mainly attributed to the online rolling optimization framework adopted in this application, which enables dynamic decision-making using real-time conditions, eliminating the influence of prediction errors in static optimization and exhibiting stronger robustness in mitigating grid load fluctuations.

[0095] Furthermore, the step of solving the time-slot optimization problem includes: transforming the time-slot optimization problem of minimizing the upper bound of the Lyapunov drift penalty term in the current scheduling time slot into an maximization revenue optimization problem; the revenue is the difference between the queue backlog release revenue and the grid reference load tracking error penalty term; constructing a mathematical model containing the revenue optimization problem using the charging decision variables of each virtual scheduling cluster as optimization variables; solving the mathematical model to obtain the charging decision results of each virtual scheduling cluster under the current scheduling time slot, and integrating them to form the charging scheduling strategy for the current scheduling time slot.

[0096] Here, we construct a time-slot-wise optimization problem: minimizing the upper bound of the drift penalty term is transformed into maximizing the profit problem, with the optimization objective being... The constraints are ,in, ; For clusters Binary charging decision variables (charging scheduling strategy). This indicates that the j-th cluster is being charged in this time slot. This indicates that the device is not charging. and These represent the demand backlog weight and urgency priority weight for the current time slot t, respectively, guiding the system to prioritize scheduling clusters with large backlogs or high urgency. For a single electric vehicle The rated charging power. This represents the total power scale of the j-th virtual scheduling cluster in the charging state. These are control parameters (weights) used to balance the ratio between "demand fulfillment" and "load tracking error". The larger the value, the more accurate the system's tracking of the grid reference load. This is the reference load for the power grid.

[0097] In practical implementation, firstly, a mathematical model is constructed, using the charging decision variables of each virtual scheduling cluster. As optimization variables (taking values ​​of 0 or 1, where 0 indicates no charging in the current time slot of the cluster, and 1 indicates charging), the optimization objective is "maximizing the difference between the queue backlog release revenue and the grid reference load tracking error penalty term." This is combined with the demand backlog constraints of the first virtual queue and the urgency priority constraints of the second virtual queue, forming a complete revenue optimization mathematical model. Next, the model is solved and a strategy is generated: the above mathematical model is solved using an adapted optimization algorithm (such as integer programming or greedy algorithms) to obtain the charging decision results for each virtual scheduling cluster under the current scheduling time slot (i.e., the charging decision for each cluster). (Value selection); Finally, the decision results of all clusters are integrated to form a directly executable current time slot charging scheduling strategy, achieving precise control over cluster charging behavior. This approach transforms abstract optimization objectives into computable mathematical problems, avoiding the need for precise predictions of future data and ensuring the real-time performance and feasibility of the scheduling strategy.

[0098] For example, a relaxed linear programming approach can be used to solve the problem: Since the above problem is a 0-1 integer programming problem, the computational complexity is high. First, the binary decision variables are... Relaxing the variables to be continuous within the interval [0,1] transforms the problem into a convex optimization problem. A convex optimization solver (such as the interior-point method) can then be used to quickly find the optimal continuous solution for the current time slot. (range of values) Relaxed linear programming can significantly reduce computational complexity, enabling real-time scheduling. Then, discrete decisions are recovered using a thresholding method: a threshold is set. (Usually taken as 0.5), for each cluster j, if Then determine (This time slot charges the cluster); if Then determine (This time slot does not charge this cluster). The continuous solution is transformed into an executable discrete charging decision instruction through the threshold recovery mechanism, thereby obtaining the optimal charging strategy for each time slot.

[0099] Figure 5 This is one of the schematic diagrams illustrating the influence of key parameters on the time-averaged error ratio in the embodiments of this application; Figure 6 This is the second schematic diagram illustrating the impact of key parameters on the time-averaged error ratio in the embodiments of this application. Among them, Figure 5 The influence of different control parameters W on the convergence results is shown. Figure 6 The effect of different queue penalty function forms on the error ratio is shown.

[0100] in, Figure 5 The control parameters (weights) in this application are shown. The impact on the performance of a collaborative charging system for heterogeneous unmanned platforms. The three curves in the figure reflect the effects of different... Under the given value, the convergence process of the time average error ratio of the actual load of the cluster tracking the reference load of the power grid.

[0101] Detailed analysis of curve differences: Curve 1 ( ): Corresponds to weight The settings are as follows. It can be seen that under these parameters, the cluster's time-averaged error ratio converges to the lowest value (approximately 0.055), indicating that the charging scheduling strategy for the heterogeneous unmanned platform can most accurately track the reference load and achieve the best optimization effect. Curve 3 ( ): Corresponds to weight The settings are as follows. The curve converges to a relatively high error level (approximately 0.083), indicating that when the weight parameter is too large, the algorithm prioritizes faster queue convergence when balancing virtual queue stability and load tracking error, thus sacrificing tracking accuracy of the reference load to some extent. Curve 2 ( (): This section shows different control parameters that fall between the two mentioned above. The regulating effect on the charging scheduling performance of heterogeneous unmanned platform clusters.

[0102] Specifically, Figure 7 This is one of the statistical charts showing the charging completion rate of different clusters in the embodiments of this application; Figure 8 This is the second statistical chart showing the charging completion rate of different clusters in the embodiments of this application. Figure 7 The diagram shows the charging completion rate of each cluster under the first penalty function (f1) setting. Figure 8 The charging completion rate of each cluster is shown under the second penalty function (f2) setting, and the two together verify the effectiveness of the high urgency priority guarantee mechanism.

[0103] For example, to verify the effectiveness of the heterogeneous unmanned platform collaborative charging scheduling method provided in this application embodiment, the following simulation scenario is constructed: the scheduling time slot is set to 15 minutes. Heterogeneous unmanned platform arrival rate. During peak hours (e.g., morning and evening rush hours), the capacity is set at 500 vehicles / hour, and during off-peak hours, it is set at 200 vehicles / hour. The dwell time of heterogeneous unmanned platforms follows a conditional Gamma distribution, with daytime parameters set to shape parameter α=2 and scale parameter θ=1, and nighttime parameters set to α=5 and θ=2. Considering the differences in heterogeneous equipment, the battery capacity is set to follow a mixed distribution: battery capacity for small unmanned platforms (e.g., UAVs) is between 0.5 and 2 kWh, and battery capacity for medium and large unmanned platforms (e.g., UGVs / USVs) is between 20 and 60 kWh. Under these parameters, the cooperative charging scheduling of this application is run, with the cluster number J set to 20. Experimental results (e.g.) Figure 4 As shown in the figure, the time average error ratio of the method provided in the embodiments of this application converges to about 0.055, which is significantly better than the disordered charging strategy.

[0104] The following is an example illustrating the implementation of the technical solution in a practical application scenario of the heterogeneous unmanned platform collaborative charging scheduling method provided in the embodiments of this application. In one example, in an emergency rescue scenario, 200 heterogeneous drones (including 3 models: 80 Type A reconnaissance drones, 60 Type B transport drones, and 60 Type C communication relay drones) return to their respective charging base stations after completing post-disaster reconnaissance and material delivery tasks. The charging base station needs to complete cluster charging scheduling within multiple consecutive scheduling time slots (each time slot is 15 minutes). The core requirements are: ① Tracking the reference load of the power grid (peak period ≤ 80kW, off-peak period ≤ 120kW); ② Prioritizing charging for drones with high urgency (those needing to perform emergency communication tasks again); ③ Avoiding the backlog of charging demand.

[0105] Here is the multi-slot scheduling implementation process (a total of 6 slots, slots 1-2 are the peak period of the power grid, and slots 3-6 are the off-peak period).

[0106] Specifically, time slot t=1: (1) Initial data acquisition and clustering, real-time acquisition of charging data of the first batch of 60 drones, and acquisition of the peak period reference load of the power grid issued by the power grid. Using "charging urgency + charging duration" as the charging feature vector, a clustering algorithm was used to divide 60 drones into 4 virtual scheduling clusters. The rated charging power calculation results for each cluster are: Cluster 1 = 24 × 0.5 = 12kW, Cluster 2 = 56 × 0.5 = 28kW, Cluster 3 = 12 × 1 = 12kW, Cluster 4 = 30 × 0.8 = 24kW. (2) Objective function and dual queue construction, with the goal of "minimizing the actual total load of the cluster and The scheduling objective function is constructed with the time average error as the target; the first virtual queue (demand backlog queue): records the uncharged demand of each cluster, the initial state is cluster 1=24, cluster 2=56, cluster 3=12, cluster 4=30; the second virtual queue (priority queue): sorted by urgency, cluster 1 and cluster 3 have higher priority than cluster 2 and cluster 4, the queue state is cluster 1=3, cluster 3=2, cluster 2=1, cluster 4=1 (the larger the value, the higher the priority). (3) Solving and executing the scheduling strategy. Algorithm solution: based on the first and second virtual queues and the objective function, the dynamic scheduling method of Lyapunov optimization is adopted to construct the drift penalty term and minimize its upper bound, transforming the long-term optimization problem into the current time slot revenue optimization problem. Optimization variables and constraints: the charging decision variables of each cluster are used. (0 = no charging, 1 = charging) is the optimization variable, with the following constraints: the total load of the current time slot is ≤80kW, and high-priority clusters are charged first. Solution results: x1=1 (cluster 1 is charging), x2=1 (cluster 2 is charging), x3=1 (cluster 3 is charging), x4=0 (cluster 4 is not charging), and the current actual total load P(1)=12+28+12=52kW (≤80kW, satisfying the grid constraints). Execution and status update: Cluster 1, cluster 2, and cluster 3 start charging; the first virtual queue is updated to cluster 1=0 (charging completed), cluster 2=28 (28 out of 56 units are charging), cluster 3=12 (2 time slots need to be charged, still backlogged), and cluster 4=30; the priority of the second virtual queue remains unchanged.

[0107] Time slot t=2: Dynamic update and scheduling (peak period, (1) Data update: 40 new drones arrived (20 Type A and 20 Type C), and charging data was collected. The grid reference load remained at 80kW. (2) Clustering and queue update: The newly arrived drones were merged into the corresponding clusters (Type A was merged into cluster 2, and Type C was merged into cluster 4). The number of drones in cluster 2 was updated to 76 (rated power 38kW), and the number of drones in cluster 4 was updated to 50 (rated power 40kW). The first virtual queue was updated to cluster 2=76-28=48, cluster 3=12, and cluster 4=50. The second virtual queue added 10 high-urgency Type C drones (merged into the sub-queue of cluster 4, with priority increased to 2). (3) Scheduling and execution: Proposed-LOA solution yields x1=0 (cluster 1 completed), x2=1 (cluster 2 charging), x3=1 (cluster 3 charging), x4=1 (10 high-urgency C-type units charging, power 8kW), and the actual total load P(2)=38+12+8=58kW (≤80kW). After execution, cluster 2 has a backlog of 48-38=10 units, cluster 3 has a backlog of 12 units (still requires 1 time slot), and cluster 4 completes charging with high urgency.

[0108] Time slots 3-6: Off-peak period scheduling ( (1) Core change: The reference load of the power grid is increased to 120kW, the constraint is relaxed, and more clusters can be scheduled for charging at the same time. (2) Result of time slot 3: x2=1 (10 units remaining in cluster 2 + 30 new units, power 20kW), x3=1 (12 units remaining in cluster 3, power 12kW), x4=1 (40 units remaining in cluster 4, power 32kW), new cluster 5 (48 units of low urgency type B, power 48kW), x5=1, actual total load P(3)=20+12+32+48=112kW (≤120kW). (3) Dynamic adjustment of time slots 4-6: As the drones are charged one after another, the backlog of the first virtual queue gradually decreases, and the Proposed-LOA algorithm is dynamically adjusted. The values ​​were set to ensure that the total load always matched the 120kW reference load. By the end of time slot 6, all 200 drones had completed charging, with no backlog of demand. Drones with high urgency were prioritized for charging, and emergency missions were not delayed.

[0109] It should be noted that the embodiments of this application accurately characterize the randomness of heterogeneous unmanned platform behavior by constructing a charging behavior model based on a non-homogeneous Poisson process and conditional gamma distribution, solving the problem that traditional static models cannot adapt to fluctuations in real-world scenarios. Utilizing K-means clustering technology, large-scale single heterogeneous unmanned platforms are transformed into a small number of virtual scheduling units, significantly reducing the computational dimensionality and complexity of the scheduling algorithm, enabling the algorithm to maintain millisecond-level real-time response capabilities even in large-scale scenarios. The proposed online scheduling strategy based on Lyapunov optimization does not require precise prediction of future arrival information of heterogeneous unmanned platforms; it can make optimal decisions based solely on the current system state, exhibiting strong robustness. By introducing urgency weights and a dual-queue mechanism (demand queue and urgency queue), it effectively ensures the charging needs of users with high urgency while achieving accurate tracking of the grid reference load and smoothing load fluctuations, balancing grid-side stability with user-side fairness.

[0110] Based on the same application concept, this application also provides a heterogeneous unmanned platform collaborative charging scheduling system corresponding to the heterogeneous unmanned platform collaborative charging scheduling method provided in the above embodiments. Since the principle of the system in this application is similar to the heterogeneous unmanned platform collaborative charging scheduling method in the above embodiments of this application, the implementation of the system can refer to the implementation of the method, and the repeated parts will not be described again.

[0111] Figure 9 This is a schematic diagram of a heterogeneous unmanned platform collaborative charging scheduling system according to an embodiment of this application. Figure 9 As shown, the heterogeneous unmanned platform collaborative charging scheduling system 900 includes: The acquisition module 910 is used to acquire charging data and grid reference load of multiple arriving unmanned platforms in real time; The determination module 920 is used to determine the charging behavior model constructed based on the charging data of each unmanned platform, and to determine the charging feature vector of each unmanned platform; the charging feature vector includes charging urgency and charging duration. The partitioning module 930 is used to partition the multiple unmanned platforms into at least one virtual scheduling cluster based on the charging feature vectors of each unmanned platform. The construction module 940 is used to set a scheduling objective function with the goal of minimizing the time average error between the actual total load of the cluster and the reference load of the power grid, and to construct a first virtual queue and a second virtual queue. The first virtual queue is used to represent the charging demand backlog status of each virtual scheduling cluster, and the second virtual queue is used to represent the priority constraints based on charging urgency. The solver module 950 is used to solve the charging scheduling strategy for the current scheduling slot based on the first virtual queue, the second virtual queue, the scheduling objective function, and the Lyapunov drift penalty algorithm.

[0112] In this embodiment, a charging behavior model is constructed based on the charging data of each unmanned platform to determine a charging feature vector for each platform, including charging urgency and charging duration. Based on these feature vectors, multiple unmanned platforms are divided into at least one cluster. A scheduling objective function is set to minimize the time-averaged error between the actual total load of the cluster and the grid reference load. Simultaneously, a first virtual queue representing the backlog of charging demand and a second virtual queue representing priority based on charging urgency are constructed. Based on the first virtual queue, the second virtual queue, the scheduling objective function, and the Lyapunov drift penalty algorithm, the charging scheduling strategy for the current scheduling slot is solved. Thus, this application achieves real-time tracking of the grid reference load without requiring precise prediction of future charging demand and load changes, while prioritizing the charging needs of unmanned platforms with high urgency.

[0113] like Figure 10 As shown in the figure, this application provides a computer device including a processor 1011, a communication interface 1012, a memory 1013, and a communication bus 1014, wherein the processor 1011, the communication interface 1012, and the memory 1013 communicate with each other through the communication bus 1014. Memory 1013 is used to store computer programs; In one embodiment of this application, when the processor 1011 executes the program stored in the memory 1013, it implements the heterogeneous unmanned platform collaborative charging scheduling method provided in any of the foregoing method embodiments, including: Real-time acquisition of charging data and grid reference load from multiple arriving unmanned platforms; A charging behavior model is constructed based on the charging data of each unmanned platform to determine the charging feature vector of each unmanned platform; the charging feature vector includes charging urgency and charging duration. Based on the charging feature vectors of each unmanned platform, the multiple unmanned platforms are divided into at least one virtual scheduling cluster; The scheduling objective function is set with the goal of minimizing the time average error between the actual total load of the cluster and the reference load of the power grid. At the same time, a first virtual queue and a second virtual queue are constructed. The first virtual queue is used to represent the charging demand backlog status of each virtual scheduling cluster, and the second virtual queue is used to represent the priority constraint based on the charging urgency. Based on the first virtual queue, the second virtual queue, the scheduling objective function, and the Lyapunov drift penalty algorithm, the charging scheduling strategy for the current scheduling slot is solved.

[0114] It will be understood by those skilled in the art that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program may be stored in a storage medium, which is a computer-readable storage medium. The computer program is executed by at least one processor in the computer system to implement the process steps of the embodiments of the above methods.

[0115] Therefore, this application embodiment also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the heterogeneous unmanned platform collaborative charging scheduling method provided in any of the foregoing method embodiments.

[0116] The storage medium is a physical, non-transient storage medium, such as a USB flash drive, external hard drive, read-only memory (ROM), magnetic disk, or optical disk, or any other physical storage medium capable of storing program code. The computer-readable storage medium can be non-volatile or volatile.

[0117] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0118] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For example, the division of each unit is merely a logical functional division, and there may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0119] The steps in the methods of this application embodiment can be adjusted, merged, or deleted according to actual needs. The units in the apparatus of this application embodiment can be merged, divided, or deleted according to actual needs. Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0120] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a terminal, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0121] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0122] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Since these modifications and variations fall within the scope of the claims and their equivalents, this application also intends to include these modifications and variations.

[0123] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for collaborative charging scheduling of heterogeneous unmanned platforms, characterized in that, The method includes: Real-time acquisition of charging data and grid reference load from multiple arriving unmanned platforms; A charging behavior model is constructed based on the charging data of each unmanned platform to determine the charging feature vector of each unmanned platform; the charging feature vector includes charging urgency and charging duration. Based on the charging feature vectors of each unmanned platform, the multiple unmanned platforms are divided into at least one virtual scheduling cluster; The scheduling objective function is set with the goal of minimizing the time average error between the actual total load of the cluster and the reference load of the power grid. At the same time, a first virtual queue and a second virtual queue are constructed. The first virtual queue is used to represent the charging demand backlog status of each virtual scheduling cluster, and the second virtual queue is used to represent the priority constraint based on the charging urgency. Based on the first virtual queue, the second virtual queue, the scheduling objective function, and the Lyapunov drift penalty algorithm, the charging scheduling strategy for the current scheduling slot is solved.

2. The method according to claim 1, characterized in that, The charging behavior model is used to characterize the arrival process and dwell time patterns of the unmanned platform; the charging behavior model constructed based on the charging data of each unmanned platform determines the charging feature vector of each unmanned platform, including: The dwell time of each unmanned platform is obtained through the charging behavior model; For any unmanned platform, the corresponding charging time is calculated based on the battery capacity, rated charging power, state of charge upon arrival, and target state of charge in the charging data of the unmanned platform. The charging urgency of the unmanned platform is determined based on its dwell time and charging time. The charging urgency and charging duration of the unmanned platform are used as core dimensions to determine the charging feature vector of the unmanned platform.

3. The method according to claim 2, characterized in that, The charging behavior model includes a non-homogeneous Poisson process model characterizing the arrival process of the unmanned platform and a conditional gamma distribution model describing the dwell time of the unmanned platform; obtaining the dwell time of each unmanned platform through the charging behavior model includes: The number of newly arriving unmanned platforms in the current time slot is determined based on a time-varying arrival rate function predefined in the non-homogeneous Poisson process model; the time-varying arrival rate function sets different arrival rates according to the time-slot characteristics of the actual scenario. Based on the shape and scale parameters of the gamma distribution model corresponding to the current time slot, the dwell time of each newly arrived unmanned platform is determined.

4. The method according to claim 3, characterized in that, The process of determining the dwell time of each newly arrived unmanned platform based on the shape and scale parameters corresponding to the conditional gamma distribution model of the current time slot includes: Based on the time period category to which the current time slot belongs, retrieve the shape parameter and scale parameter associated with the time period category from the pre-built table of correspondence between time period category and gamma distribution parameter; Using the shape and scale parameters associated with the time period category, a conditional gamma distribution model is constructed. The conditional gamma distribution model is used to characterize the stochastic characteristics of the dwell time of the unmanned platform under the time period category. For each newly arrived unmanned platform in the current time slot, a random sampling operation is performed from the conditional gamma distribution model corresponding to the time slot category to generate the initial dwell time of the unmanned platform; By combining the task type and preset adjustment rules of each unmanned platform, the initial dwell time of the unmanned platform is adjusted to obtain the dwell time of each unmanned platform.

5. The method according to claim 1, characterized in that, The method of dividing the multiple unmanned platforms into at least one virtual scheduling cluster based on the charging feature vectors of each unmanned platform includes: The number of clusters is determined based on the scale of the heterogeneous unmanned platform, and multiple initial cluster centers are selected using the K-means algorithm; Using the charging feature vectors of each unmanned platform as input, the K-means algorithm is used to minimize the sum of squared distances between the charging feature vectors of each unmanned platform within the cluster and the corresponding cluster center, thereby dividing the multiple unmanned platforms into at least one virtual scheduling cluster. In each scheduling time slot, for any newly arrived unmanned platform, it is assigned to the nearest cluster center according to the charging feature vector corresponding to the unmanned platform. At the same time, unmanned platforms that have left the charging base station are removed, the cluster centers are recalculated, and the composition of the virtual scheduling cluster is updated.

6. The method according to claim 1, characterized in that, The steps for setting the scheduling objective function include: Within the total number of scheduling slots in the cluster scheduling, calculate the actual total load of the cluster for each scheduling slot; the actual total load of the cluster is the sum of the charging power of all virtual scheduling clusters; The difference between the actual total load of the cluster and the reference load of the power grid in each scheduling time slot is calculated and squared. The expected value of the squared difference of all scheduling time slots is calculated and then summed. The summation result is divided by the total number of scheduling time slots, and the limit when the total number of scheduling time slots approaches infinity is taken to obtain the scheduling objective function.

7. The method according to claim 1, characterized in that, The steps of constructing the first virtual queue and the second virtual queue include: For any virtual scheduling cluster, a first virtual queue is constructed; based on the charging demand backlog status of any virtual scheduling cluster in the current scheduling time slot, the actual charging amount in the current scheduling time slot, and the charging demand of the newly arrived unmanned platform in the current scheduling time slot, the charging demand backlog status of the virtual scheduling cluster in the next scheduling time slot is determined, and the first virtual queue is updated. For any virtual scheduling cluster, a second virtual queue is constructed; based on the urgency queue status of the virtual scheduling cluster in the current scheduling time slot, the charging decision variables of the current scheduling time slot, the penalty function for negative correlation of charging urgency, and the indicator function, the urgency queue status of the virtual scheduling cluster in the next scheduling time slot is determined, and the second virtual queue is updated; the indicator function is determined according to the charging demand backlog status of the first virtual queue of the virtual scheduling cluster.

8. The method according to claim 1, characterized in that, The step of solving the charging scheduling strategy for the current scheduling slot based on the first virtual queue, the second virtual queue, the scheduling objective function, and the Lyapunov drift penalty algorithm includes: Based on the first and second virtual queue states of all virtual scheduling clusters, a Lyapunov function is constructed; the Lyapunov function is used to characterize the combined congestion degree of charging demand backlog and urgency delay of the cluster under the current scheduling time slot. Calculate the expected difference between the Lyapunov function values ​​of the current scheduling slot and the next scheduling slot to obtain the Lyapunov drift term; The Lyapunov drift term is added to the weighted scheduling objective function term to construct the Lyapunov drift penalty term; the weighted scheduling objective function term is obtained by the expected product of the control parameters and the squared errors of the actual total load of the current scheduling time slot cluster and the grid reference load. The long-term optimization objective is transformed into a time-slot optimization problem that minimizes the upper bound of the Lyapunov drift penalty term in the current scheduling slot. By solving the time-slot optimization problem, the charging scheduling strategy for the current scheduling slot is generated.

9. The method according to claim 8, characterized in that, The steps for solving the time-slot optimization problem include: The time-slot optimization problem of minimizing the upper bound of the Lyapunov drift penalty term in the current scheduling time slot is equivalently transformed into a revenue optimization problem of maximizing the revenue; the revenue is the difference between the queue backlog release revenue and the grid reference load tracking error penalty term. Using the charging decision variables of each virtual scheduling cluster as optimization variables, a mathematical model incorporating the aforementioned revenue optimization problem is constructed. Solving the mathematical model yields the charging decision results of each virtual scheduling cluster under the current scheduling time slot, which are then integrated to form the charging scheduling strategy for the current scheduling time slot.

10. A heterogeneous unmanned platform collaborative charging scheduling system, characterized in that, The system includes: The acquisition module is used to acquire charging data and grid reference load from multiple arriving unmanned platforms in real time. The determination module is used to determine the charging behavior model based on the charging data of each unmanned platform, and to determine the charging feature vector of each unmanned platform; the charging feature vector includes charging urgency and charging duration. The partitioning module is used to divide the multiple unmanned platforms into at least one virtual scheduling cluster based on the charging feature vectors of each unmanned platform. The construction module is used to set a scheduling objective function with the goal of minimizing the time average error between the actual total load of the cluster and the reference load of the power grid, and to construct a first virtual queue and a second virtual queue. The first virtual queue is used to represent the charging demand backlog status of each virtual scheduling cluster, and the second virtual queue is used to represent the priority constraints based on charging urgency. The solution module is used to solve the charging scheduling strategy for the current scheduling slot based on the first virtual queue, the second virtual queue, the scheduling objective function, and the Lyapunov drift penalty algorithm.

Images