Full-automatic AGV (Automatic Guided Vehicle) charging collaborative optimization system

Abstract

The invention relates to the technical field of electric vehicles, and discloses a full-automatic AGV charging collaborative optimization system which comprises an energy consumption prediction module, a consensus mechanism design module and an AGV scheduling optimization module. The energy consumption prediction module is used for acquiring a pose topological feature vector and a potential margin attenuation gradient, constructing a multi-dimensional input vector, and generating a dynamic energy consumption prediction value based on an energy consumption track residual model of multi-physics field coupling; the consensus mechanism design module is used for constructing and verifying a dynamic consensus factor and providing a priority decision basis; the AGV scheduling optimization module is used for discretizing an operation space of a battery swap station, generating a space-time folding unit, constructing a space-time conflict resolution atlas, and generating an optimal battery swap sequence and a collision avoidance path; through multi-module cooperation, efficient cooperation of multi-AGV charging and battery swap is realized, and congestion of the battery swap station and resource waste are avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of electric vehicles, and more specifically, to a collaborative optimization system for automatic charging and battery swapping of AGV vehicles. Background Art

[0002] With the continuous improvement of the automation level of logistics warehousing, as the core equipment for warehousing operations, the efficient charging and battery swapping management of AGVs (Automated Guided Vehicles) has become a key link to ensure the stable operation of the system. When multiple AGVs are simultaneously in a critical power state and converge on the same battery swapping station, traditional scheduling methods often lead to congestion or even paralysis of the battery swapping station due to energy consumption prediction deviation, single priority decision-making, and insufficient resolution of spatio-temporal conflicts, seriously affecting warehousing efficiency.

[0003] In the prior art, the Chinese patent application with the publication number CN113902289A proposes an AGV path planning method based on a flexible spatio-temporal network model, which optimizes the path and energy consumption through a hybrid heuristic algorithm. However, this method only focuses on the time and energy consumption objectives in path planning, and does not deeply integrate the dynamic impact of AGV motion states (such as trajectory curvature, battery aging, etc.) on energy consumption, making it difficult to accurately predict the remaining battery life under complex working conditions. This may lead to an AGV with "high-curvature driving + aging battery" running out of power midway due to insufficient battery life estimation, or a low-energy-consumption AGV occupying the battery swapping station resources in advance, exacerbating congestion. The Chinese patent application with the publication number CN118068836A discloses a multi-AGV obstacle avoidance and path planning method and system based on deep learning, which realizes multi-AGV obstacle avoidance and path planning based on deep learning and adapts to environmental changes through reinforcement learning. However, its priority decision-making depends on static power thresholds or task priorities, and does not construct a multi-dimensional evaluation system covering charging urgency, spatial conflict risk, and task timeliness, making it impossible to dynamically balance multiple constraints within the limited space of the battery swapping station, and easily causing problems such as "urgent task AGVs being delayed in scheduling due to spatial conflicts" or "multiple AGVs flooding into the battery swapping station resulting in robotic arm preemption".

[0004] In the prior art, when solving the problem of efficient and orderly battery swapping when multiple AGV vehicles are simultaneously in a critical power state and converge on the same battery swapping station, the battery swapping priority decision only depends on a single index (such as power or task priority), lacks a quantitative evaluation of spatial conflict risk and task timeliness, and cannot meet the multi-objective optimization requirements. Summary of the Invention

[0005] The present invention is applicable to scenarios such as large-scale intelligent warehousing, logistics centers, and automated factories. For AGV fleets of multiple models and multiple task types, under high-frequency charging and battery swapping requirements, it solves the congestion problem caused by centralized battery swapping when multiple AGVs reach the critical power level.

[0006] To overcome the above-mentioned defects of the prior art, the present invention provides a fully automatic AGV car charging and battery swapping collaborative optimization system. By integrating the AGV pose topological features and the battery potential margin attenuation gradient through an energy consumption prediction module, a multi-physical field coupled energy consumption trajectory residual model is constructed to achieve accurate energy consumption prediction. The consensus mechanism design module integrates the charging urgency, spatial conflict entropy value, and task time penalty term to generate a credible dynamic consensus factor to quantify the multi-dimensional priorities. With the help of the AGV scheduling optimization module, the spatio-temporal resources of the battery swapping station are discretized, a spatio-temporal conflict resolution map is constructed, and the optimal battery swapping sequence and collision avoidance path are generated. It breaks through the limitation of traditional single-dimensional scheduling, realizes the collaborative analysis of multi-physical fields of "motion - energy consumption - aging", significantly improves the scheduling efficiency and reliability of the battery swapping station in the multi-AGV critical power scenario, and avoids congestion and resource waste caused by prediction deviation, priority imbalance, and spatio-temporal conflict.

[0007] To achieve the above object, the present invention provides the following technical solutions: A fully automatic AGV car charging and battery swapping collaborative optimization system, comprising: An energy consumption prediction module: used to obtain the pose topological feature vector and the potential margin attenuation gradient of the AGV car, construct a multi-dimensional input vector, and generate an adaptive dynamic energy consumption prediction value based on the multi-dimensional input vector and a pre-constructed multi-physical field coupled energy consumption trajectory residual model; A consensus mechanism design module: used to construct a dynamic consensus factor that provides a priority decision basis for the AGV charging and battery swapping scheduling optimization, and perform cross-node verification on the constructed dynamic consensus factor to obtain a credible dynamic consensus factor; An AGV scheduling optimization module: used to discretize the operation space of the battery swapping station to generate spatio-temporal folding units; based on the dynamic energy consumption prediction value, the credible dynamic consensus factor, and the spatio-temporal folding units, construct a spatio-temporal conflict resolution map of the battery swapping station, and generate the optimal battery swapping sequence and collision avoidance path of the AGV.

[0008] Further, the method for obtaining the pose topological feature vector of the AGV car is: obtaining the pose topological features of the AGV car in real time to form a pose topological feature vector; the pose topological features at least include three-dimensional coordinates.

[0009] Further, the method for constructing the multi-dimensional input vector is: normalizing the pose topological feature vector and the potential margin attenuation gradient, and splicing the normalized pose topological feature vector and the potential margin attenuation gradient into a multi-dimensional input vector.

[0010] Further, the consensus mechanism design module includes a dynamic consensus factor construction unit and a dynamic consensus factor verification unit; The dynamic consensus factor construction unit is used to determine the urgency weight reflecting the charging urgency of the AGV, the spatial conflict entropy value reflecting the spatial conflict risk of the AGV, and the task timeliness penalty term reflecting the task timeout risk of the AGV. According to the urgency weight, the spatial conflict entropy value, and the task timeliness penalty term, a dynamic consensus factor is constructed to provide a priority decision basis for the AGV charging and swapping scheduling optimization.

[0011] Furthermore, the dynamic consensus factor verification unit is used to build a distributed collaborative verification layer based on the Byzantine fault tolerance mechanism, and cross-node verification is performed on the constructed dynamic consensus factor through the distributed collaborative verification layer to obtain a trusted dynamic consensus factor.

[0012] Furthermore, the method for determining the urgency weight includes: Obtain the current remaining power DE of the AGV; according to the current remaining power DE of the AGV and the potential margin attenuation gradient, obtain the activation exponential growth function; set the initial urgency weight E1. When DE≥S1, the urgency weight E = E1; when DE<S1, dynamically increase the initial urgency weight E1 according to the activation exponential growth function to obtain the urgency weight E, where S1 is the power urgency threshold.

[0013] Furthermore, the method for determining the spatial conflict entropy value includes: Based on the three-dimensional coordinates of the AGV in the pose topological feature vector, obtain the relative distances between multiple AGVs; According to the relative distances between multiple AGVs, introduce a virtual repulsive force field and establish a congestion model of the AGV in the operation area of the swapping station; According to the established congestion model of the AGV in the operation area of the swapping station, obtain the spatial conflict entropy value reflecting the spatial conflict risk of the AGV.

[0014] Furthermore, the method for determining the task timeliness penalty term includes: Obtain the deadline and the current time of the task currently executed by the AGV; introduce an inverse proportional function of the deadline and the current time of the task; according to the inverse proportional function, generate a task timeliness penalty term reflecting the task timeout risk of the AGV.

[0015] Furthermore, the method for discretizing the operation space of the swapping station includes: Extract the four-dimensional spatio-temporal coordinates of the resource elements of the swapping station to form a state matrix of the operation space of the swapping station; perform compression mapping on the state matrix of the operation space of the swapping station to generate a spatio-temporal folding unit reflecting the spatio-temporal dependence relationship between the resource elements of the swapping station.

[0016] Furthermore, the method for constructing the spatio-temporal conflict resolution map of the swapping station includes: Map the dynamic energy consumption prediction value and the credible dynamic consensus factor to the space-time folding unit to form a six-dimensional state matrix; based on the six-dimensional state matrix, construct a space-time conflict resolution map for the battery swapping station; each AGV corresponds to a node in the space-time conflict resolution map of the battery swapping station.

[0017] Further, the method for generating the optimal battery swapping order of the AGV includes: Construct a node coloring model according to the space-time conflict resolution map of the battery swapping station; according to the node coloring model, traverse and color the nodes of the map according to the credible dynamic consensus factor; Gradually eliminate the adjacent nodes of the same color in the space-time conflict resolution map of the battery swapping station through the space-time permutation algorithm; through iterative optimization, output the final node coloring scheme, and the final node coloring scheme corresponds to the optimal battery swapping order of the AGV.

[0018] Compared with the prior art, the beneficial effects of the present invention are: The full-automatic AGV car charging and battery swapping collaborative optimization system of the present invention improves the charging and battery swapping efficiency and reliability of multiple AGVs in the critical power scenario through the organic collaboration of multiple modules. The energy consumption prediction module establishes a physical coupling relationship between the motion trajectory and the battery state through cross-domain data fusion, accurately captures the non-linear energy consumption growth under complex working conditions such as "high-curvature driving + aging battery", provides a reliable remaining battery life estimate for the scheduling system, and avoids the centralized congestion of multiple AGVs at the battery swapping station due to prediction deviation. The dynamic consensus factor constructed by the consensus mechanism design module integrates the charging urgency, space conflict risk and task timeout risk, solves the problem of single priority decision-making basis in traditional methods, ensures that the scheduling system can dynamically balance multiple constraints, avoids decision-making deviations in scenarios such as "urgent but not crowded" and "crowded but not urgent", and at the same time ensures data reliability through cross-node verification. The AGV scheduling optimization module performs space-time discretization processing on the operation space of the battery swapping station, constructs a space-time conflict resolution map, combines the dynamic energy consumption prediction value and the credible dynamic consensus factor, generates the optimal battery swapping order and collision avoidance path, effectively processes the time window conflict and priority scheduling problem of multiple AGVs, avoids space resource competition and equipment collision, and realizes a battery swapping process of "ordered in time and conflict-free in space". The entire full-automatic AGV car charging and battery swapping collaborative optimization system realizes the upgrade from "single power monitoring" to "multi-physical field collaborative analysis of motion-energy-aging", significantly improves the scheduling efficiency of the battery swapping station in high-frequency usage scenarios, and ensures the efficient operation of the warehousing and logistics system. Description of the Drawings

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0020] Figure 1 It is a functional module diagram of a full-automatic AGV car charging and battery swapping collaborative optimization system in the present invention; Figure 2 It is a method flow chart for obtaining the potential margin attenuation gradient of an AGV car in a full-automatic AGV car charging and battery swapping collaborative optimization system of the present invention; Figure 3 It is a method flow chart for determining the emergency weight E in a full-automatic AGV car charging and battery swapping collaborative optimization system of the present invention; Figure 4 It is a method flow chart for determining the spatial conflict entropy value in a full-automatic AGV car charging and battery swapping collaborative optimization system of the present invention; Figure 5 It is a method flow chart for determining the task timeliness penalty term in a full-automatic AGV car charging and battery swapping collaborative optimization system of the present invention; Figure 6 It is a schematic diagram of a scenario where multiple AGVs in the embodiments of the present invention are simultaneously in a critical power state and gather to go to the same battery swapping station; Figure 7 It is a schematic diagram of the AGV car with the most urgent task being preferentially charged in the embodiments of the present invention; Figure 8 It is a method flow chart for iteratively searching for the optimal battery swapping order through a space-time replacement algorithm based on the space-time conflict resolution map of the battery swapping station in a full-automatic AGV car charging and battery swapping collaborative optimization system of the present invention; Figure 9 It is a method flow chart of a full-automatic AGV car charging and battery swapping collaborative optimization method in the embodiments of the present invention. Specific embodiments

[0021] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0022] Embodiment 1:

[0023] Please refer to Figure 1As shown in the figure, this embodiment provides a full-automatic AGV car charging and battery swapping collaborative optimization system, including: An energy consumption prediction module, configured to obtain the pose topological feature vector and potential margin attenuation gradient of the AGV car, construct a multi-dimensional input vector, and generate an adaptive dynamic energy consumption prediction value based on the multi-dimensional input vector and a pre-constructed energy consumption trajectory residual model with multi-physical field coupling; Traditional methods independently process motion data and battery data, resulting in the energy consumption model ignoring the collaborative attenuation effect of "high-curvature driving + aging battery". The energy consumption prediction module establishes a physical coupling relationship between the motion trajectory and the battery state through cross-domain data fusion, enabling the energy consumption trajectory residual model to accurately predict the energy consumption changes under complex working conditions, providing a reliable estimated remaining battery life for the scheduling system, and avoiding the centralized congestion of multiple AGVs at the battery swapping station caused by prediction deviation.

[0024] The energy consumption prediction module includes: A feature fusion unit, configured to obtain the pose topological feature vector and potential margin attenuation gradient of the AGV car, and construct a multi-dimensional input vector; Furthermore, the feature fusion unit in the embodiment of the present application is further configured to perform the following steps: Step S1110: Real-time obtain the pose topological features of the AGV car to form a pose topological feature vector; the pose topological features include three-dimensional coordinates, motion vector angle, and trajectory curvature dynamic parameters; Specifically, the three-dimensional coordinates use the Cartesian coordinate system to record the position (x, y, z) of the AGV in the storage space in real time, which is used to determine its absolute space occupancy; the motion vector angle is defined as the angle θ between the current motion direction of the AGV and the x-axis of the global coordinate system, which is calculated in real time by analyzing the rotational speed difference of the drive wheels or the inertial navigation system, reflecting the heading change of the AGV. For example, when making a right-angle turn, θ suddenly changes from 0° to 90°, indicating that the motion trajectory needs to be adjusted. The dynamic parameter of the trajectory curvature is based on the curvature radius estimation algorithm of the Frenet frame. By continuously collecting pose data to fit the local trajectory, the curvature radius ρ is calculated. This parameter quantifies the non-linear complexity of the trajectory. The smaller the ρ value (such as a sharp turn with ρ = 1 meter), the higher the requirement for maneuverability, and the corresponding energy consumption increases due to frequent steering drives. The above parameters constitute the pose topological feature vector [x, y, z, θ, ρ], providing basic data at the geometric level for subsequent analysis of the motion energy consumption characteristics of the AGV in the storage space. The traditional energy consumption prediction model only relies on linear distance or speed parameters and does not consider the non-linear impact of the trajectory curvature on energy consumption (such as a sudden increase in the motor load caused by frequent steering). By introducing the dynamic parameter of the trajectory curvature, the maneuverability of the AGV (such as the degree of bending of the obstacle avoidance path) is associated with the motor energy consumption. Combined with the potential margin decay gradient in step S1120, it jointly characterizes the coupling relationship of "motion state - battery decay". If step S1110 is missing, the energy consumption trajectory residual model will not be able to distinguish the energy consumption differences between straight-line driving and complex paths, resulting in the charging station scheduling strategy ignoring the real-time motion requirements of the AGV.

[0025] Step S1120: Obtain the potential margin decay gradient of the AGV car. Furthermore, as Figure 2 shown, step S1120 includes: Step S1121: Collect the charge and discharge current data of the AGV car battery in real time, perform integral operation on the current data according to the time series, and obtain the change in the cumulative charge from the last charging completion to the current moment. Step S1122: Obtain the battery aging coefficient reflecting the current degree of battery performance degradation. Step S1123: Input the change in the cumulative charge and the battery aging coefficient into the pre-constructed non-linear mapping function to obtain the potential margin decay gradient of the current AGV car; the non-linear mapping function takes the change in the cumulative charge and the battery aging coefficient as inputs and the potential margin decay gradient as the output.

[0026] Specifically, the change in cumulative charge ΔQ is obtained by the battery management system (BMS) collecting the charging and discharging current in real time and integrating the current over time from the moment when the last charge was completed to the current moment. ΔQ reflects the cumulative amount of charge consumed or supplemented by the battery during operation. A positive value indicates charging, and a negative value indicates discharging. For example, when continuously carrying goods, ΔQ is negative, and its absolute value increases with the running time. The battery aging coefficient SoH is determined by analyzing the historical cycle charge and discharge times of the battery, the open-circuit voltage decay curve in the static state, and combining the electrochemical impedance spectroscopy (EIS) model or the capacity decay fitting algorithm. The value range of SoH is 0 - 1, where 1 represents a brand-new state and 0 represents complete failure. For example, after a certain battery has been cycled 500 times, the measured capacity drops to 80% of the rated capacity, then SoH = 0.8, indicating that its actual available energy has decayed by 20%. Based on the physical and chemical mechanisms during the battery charging and discharging process (such as the diffusion resistance of lithium ions in the electrode material, the decrease in electrolyte conductivity with aging), a mapping relationship with the input of ΔQ and SoH and the output of the potential margin decay gradient (γ) is established. This function is fitted through experimental data to characterize the property that the potential of an aging battery drops faster under the same charge change. For example, when SoH = 0.6, the γ value corresponding to the same ΔQ is 30% higher than when SoH = 1, reflecting the accelerating effect of the increased internal resistance caused by aging on potential decay. The finally obtained γ value characterizes the potential decay rate of the battery considering the current energy consumption and aging state, providing a key basis for dynamically evaluating the remaining endurance of the AGV. Battery aging leads to a prediction deviation of the actual available charge (SoC) under the same charge consumption. Traditional methods do not take SoH into account and cannot accurately predict the "false charge" phenomenon of aging batteries.

[0027] Step S1130, normalize the pose topological feature vector and the potential margin decay gradient, and splice the normalized pose topological feature vector and potential margin decay gradient into a multi-dimensional input vector.

[0028] Specifically, the three-dimensional coordinates (x, y, z) and the motion vector angle θ are converted into relative coordinates (s, φ, k) in the Frenet frame, where s is the arc length along the tangent direction of the trajectory, φ is the angle between the tangent and the principal normal, and k is the trajectory curvature (k = 1 / ρ). This transformation eliminates the influence of the absolute position in the global coordinate system and focuses on the local geometric features of the AGV's own trajectory. For example, it uniformly maps the path curvatures in different regions to relative values, facilitating the model to compare the energy consumption characteristics under different working conditions. The trajectory curvature k is normalized, and SoH and γ are mapped to the 0-1 interval. Through such processing, the interference of dimensional differences on model training is eliminated. For example, it avoids the three-dimensional coordinates (numerical range 0-100 meters) dominating the model weights during input and ensures the effective influence of the battery state parameters (0-1 interval). The finally spliced multi-dimensional input vector provides input data with a unified dimension for the subsequent multi-physics field coupling model, solving the training deviation problem caused by the direct fusion of heterogeneous data. If this step is missing, the model needs to additionally add a feature scaling layer, increasing the computational complexity and reducing the real-time performance.

[0029] The full-automatic AGV car charging and battery swapping collaborative optimization system aims to solve the problem of congestion in the battery swapping station when multiple AGVs reach the critical battery level. The core challenge lies in how to integrate the AGV motion state and the battery state to achieve accurate energy consumption prediction and scheduling priority calculation. The feature fusion unit specifically solves the problems of heterogeneous data fusion obstacles and inaccurate model input. The dimensional differences between the pose data (meters, radians) and the electric potential data (ampere-hours, dimensionless) lead to an imbalance in parameter weights during model training. For example, directly inputting the three-dimensional coordinates (large numerical range) and SoH (small numerical range) will cause the model to overly focus on the spatial position and ignore the significant impact of battery aging on energy consumption. The energy consumption of the AGV is jointly determined by the complexity of the motion trajectory (such as the curvature radius affecting the driving motor load) and the battery performance (such as increased internal resistance due to aging). Traditional methods independently process these two types of data and cannot capture the non-linear energy consumption growth in the scenario of "high-curvature driving + aging battery", which may lead to too high energy consumption prediction errors and further cause scheduling decision-making mistakes. Without the trajectory curvature parameter, the model cannot distinguish between straight-line driving (low energy consumption) and continuous curve driving (high energy consumption), which may cause low-energy-consuming AGVs to occupy the battery swapping station in advance, while high-energy-consuming AGVs stall due to depleted battery power, exacerbating congestion; if the battery aging coefficient is ignored, the endurance of aging batteries will be overestimated, causing them to power off before reaching the battery swapping station, blocking the passage and triggering a chain reaction.

[0030] The geometric features (curvature, heading) of the multi-dimensional input vector fusion motion trajectory and the electrochemical features (aging, power decay rate) of the battery state enable the energy consumption trajectory residual model to capture the coupling effect of "motion - energy consumption - battery aging". The potential margin decay gradient γ directly affects the dynamic adjustment of the urgency weight E. When the remaining power DE is lower than the threshold S1 (such as 15%), the exponential growth function is activated in combination with the γ value, enabling AGVs with severe aging and high decay gradients to obtain higher priorities, solving the problem of "being unable to distinguish the actual battery endurance based solely on the remaining power" and avoiding the over-inclination of scheduling resources towards new-battery AGVs. The three-dimensional coordinates in the pose topological features provide inputs for the calculation of the spatial conflict entropy value. A virtual repulsive potential field is constructed through the relative distance to quantify the congestion risk near the battery swapping station. For example, when two AGVs with small curvature radii run in parallel within 5 meters of the battery swapping station, the potential field strength increases non-linearly with the decrease in distance, prompting the scheduling system to prioritize arranging one of them to swap batteries in advance, reducing the probability of spatial collisions.

[0031] In the fully automatic AGV car charging and battery swapping collaborative optimization system, if the feature fusion unit is missing, then: The energy consumption prediction model fails: lacking the trajectory curvature parameter, it is unable to distinguish the energy consumption differences under different road conditions, resulting in straight-line running AGVs and curve-running AGVs being treated equally, and there may be a resource mismatch such as "low-energy consumption AGVs preempting the battery swapping station in advance, and high-energy consumption AGVs stagnating due to power exhaustion"; lacking the battery aging coefficient will overestimate the endurance of aging batteries, causing them to run out of power on the way to the battery swapping station, blocking the passage and triggering a chain of congestion.

[0032] The calculation of the scheduling priority is inaccurate: determining the urgency weight solely based on the remaining power and ignoring the rapid decay characteristics of aging batteries may cause AGVs that truly need charging urgently to be scheduled late, while new-battery AGVs are given priority, resulting in a mismatch between the resource allocation of the battery swapping station and the actual demand, exacerbating the congestion risk.

[0033] Heterogeneous data cannot be fused: Geometric and electrochemical dimensions are directly input into the model, resulting in training biases (such as the gradient descent algorithm converging to a local optimal solution). Ultimately, the generated battery swapping order and collision avoidance path lack the support of a physical mechanism and cannot meet the scheduling requirements under complex multi-AGV working conditions, reducing the processing efficiency of the battery swapping station.

[0034] The feature fusion unit systematically solves the technical obstacles of heterogeneous data fusion by constructing multi-dimensional input vectors, providing basic data with clear physical meaning and unified dimensions for subsequent energy consumption prediction, priority calculation, and spatio-temporal scheduling. Its core value lies in the cross-domain coupling of the motion characteristics of AGVs and battery states, enabling the entire scheduling system to be upgraded from "single power monitoring" to "motion-energy consumption-aging" multi-physical field collaborative analysis, fundamentally improving the scheduling efficiency and reliability of the battery swapping station in scenarios with multiple AGVs at critical battery levels, and avoiding congestion risks and resource waste caused by one-sided data.

[0035] The energy consumption prediction unit generates dynamic energy consumption prediction values with adaptive characteristics based on multi-dimensional input vectors and a pre-constructed energy consumption trajectory residual model with multi-physical field coupling.

[0036] Specifically, the energy consumption prediction unit solves the problem of insufficient accuracy in AGV energy consumption prediction under complex working conditions by integrating multi-dimensional physical characteristics and data-driven modeling methods, avoiding resource allocation imbalance and congestion risks in the battery swapping station caused by prediction deviations, and providing key inputs for subsequent construction of dynamic consensus factors and resolution of spatio-temporal conflicts in the battery swapping station. The energy consumption trajectory residual model with multi-physical field coupling is a composite prediction model that integrates the motion physical characteristics of AGVs (such as pose topology features) and the electrochemical characteristics of batteries (such as the attenuation law of potential margin). Its core architecture includes four parts: an input layer, a multi-modal feature fusion layer, a residual calculation layer, and an adaptive output layer. Input layer: Receives the multi-dimensional input vectors generated by the feature fusion unit, including pose topology feature vectors (three-dimensional coordinates, motion vector angles, dynamic parameters of trajectory curvature) and potential margin attenuation gradients (reflecting the battery energy consumption rate and aging state).

[0037] Multi-modal feature fusion layer: Adopts a spatio-temporal attention mechanism to perform cross-modal association on pose data in the spatial dimension and battery state data in the time dimension, and extracts their coupled features (for example, the additional consumption of high-curvature trajectories on battery energy consumption).

[0038] Residual calculation layer: Through training with historical data, a residual database of "predicted energy consumption" and "actual energy consumption" is established to identify prediction deviation patterns under different working conditions (such as the reduction in battery life caused by battery aging).

[0039] Adaptive output layer: Based on the prediction model corrected by residuals, outputs dynamic energy consumption prediction values for a future period of time, which are adaptively adjusted according to the real-time state of AGVs. An ensemble of decision trees generated by integrating the random forest algorithm uses the AdaBoost mechanism to strengthen the weights of abnormal working condition samples, and an online learning module is configured to update model parameters in real time.

[0040] The model training sample dataset contains the following core dimensions: Pose trajectory data: Collect the historical running trajectory of the AGV in the warehouse, including the three-dimensional coordinate sequence, the motion vector angle, and the dynamic parameters of the trajectory curvature (calculated by the Frenet frame curvature radius estimation algorithm, reflecting the degree of path bending).

[0041] Battery state data: Include the charge and discharge current sequence, the change in cumulative charge, the battery aging coefficient (SoH, calculated jointly by the battery capacity decay curve and the internal resistance growth model, characterizing the degree of battery performance degradation), and the potential margin decay gradient (obtained by coupling the change in cumulative charge and the aging coefficient through a non-linear mapping function, quantifying the battery energy decay rate).

[0042] Measured energy consumption data: The energy consumption value per unit time collected in real-time by the AGV battery management system (BMS), used as the supervision signal for model training.

[0043] The training samples cover different load conditions (no load / full load), path types (straight / curved), and battery aging stages (new battery / aged battery) to ensure the generalization ability of the model for complex scenarios.

[0044] Adopt an ensemble learning framework, generate multiple decision trees using the random forest algorithm, and each decision tree is trained based on a randomly sampled training subset and feature subset to reduce the model variance. Introduce the adaptive boosting (AdaBoost) mechanism, increase the weight of samples with larger prediction residuals in the previous round, forcing subsequent decision trees to focus on learning difficult-to-fit conditions (such as abnormal energy consumption under rapid acceleration and high-curvature paths). After the model is deployed, receive new AGV operation data in real-time through the online learning module, and regularly update the decision tree weights to achieve adaptive adjustment to dynamic factors such as battery aging and changes in path planning strategies.

[0045] Input the multi-dimensional input vector into the trained energy consumption trajectory residual model to generate dynamic energy consumption prediction values with adaptive characteristics; the "adaptive characteristics" means that the model can dynamically adjust the prediction parameters according to the real-time input pose and battery state data. For example, when the battery aging coefficient of a certain AGV exceeds the threshold, the model automatically increases the weight of the potential margin decay gradient to improve the prediction accuracy of the battery life reduction of this AGV.

[0046] Through multi-physics field coupling modeling, combine the motion characteristics of the AGV with the battery characteristics to solve the problem that "single-dimensional data cannot comprehensively reflect the energy consumption law". For example, the coupling of the dynamic parameters of the trajectory curvature and the potential margin decay gradient can identify the abnormal energy consumption growth under the combined condition of "high-maneuverability path + aged battery". The residual model and the adaptive training mechanism solve the problem of "the adaptability of historical data to real-time working conditions", enabling the model to dynamically adjust the prediction parameters with the AGV usage cycle (battery aging) and working environment (path changes).

[0047] In the scenario of high-frequency battery swapping for multiple AGVs, without accurate energy consumption prediction, it is impossible to accurately judge the true battery life demand of AGVs, which may lead to "incorrect battery level judgment" (such as insufficient estimation of remaining battery level resulting in power outage during operation, or excessive early battery swapping occupying resources). It is difficult to identify the impact of battery aging on battery life, resulting in rigid scheduling strategies (such as allocating battery swapping priorities according to new battery standards and ignoring the urgent charging needs of aging AGVs). The dynamic impact of pose trajectory on energy consumption is ignored. For example, the actual energy consumption of an AGV with frequent turning is higher than that of an AGV moving in a straight line. If the prediction model does not consider the trajectory curvature, it will lead to unreasonable allocation of battery swapping time windows.

[0048] If this step is omitted and a fixed energy consumption model is adopted (such as only estimating battery life based on the remaining battery level), it will lead to distorted calculation of battery swapping priorities, unable to distinguish between "low battery level due to battery aging" and "low battery level due to high-load operation", and may misallocate priorities (such as preferentially scheduling AGVs with aging batteries but sufficient actual battery life and ignoring high-load AGVs that truly need urgent battery swapping). The resolution of spatio-temporal conflicts fails. The time window allocation without energy consumption prediction cannot match the actual battery life demand of AGVs, which may lead to multiple AGVs flooding into the battery swapping station simultaneously, causing congestion and even collisions.

[0049] Combining the dynamic energy consumption prediction value with the remaining battery level in the energy consumption prediction unit can more accurately quantify the charging urgency of AGVs. For example, for two AGVs with a remaining battery level of 15% each, the model can identify that one of them is about to run out of battery life due to battery aging (with a large attenuation gradient) and the other has stable energy consumption due to a flat current path (with a small attenuation gradient) through the attenuation gradient of the potential margin, thus reasonably distinguishing their charging priorities and avoiding the rough scheduling of "determining priorities based on the absolute value of the battery level". Through the coupled analysis of pose trajectory and energy consumption, the time when the AGV arrives at the battery swapping station and the energy consumption rate can be predicted, and the battery swapping time window can be planned in advance. For example, for an AGV about to enter a path with a high curvature, the model predicts that its energy consumption will increase significantly, and thus allocate battery swapping resources for it in advance to avoid scheduling chaos caused by temporary emergency battery swapping. The adaptive training mechanism enables the model to adapt to dynamic changes such as battery aging and adjustment of path planning strategies. For example, when the warehouse layout changes, resulting in an increase in the average driving curvature of AGVs, the model automatically corrects the influence weight of the trajectory curvature on energy consumption through online learning to ensure that the prediction accuracy does not decline during long-term operation.

[0050] The consensus mechanism design module is used to construct a dynamic consensus factor that provides a basis for priority decision-making for the optimization of AGV charging and swapping scheduling, and perform cross-node verification on the constructed dynamic consensus factor to obtain a credible dynamic consensus factor; Traditional methods only sort by power or task priority and cannot balance the multi-objective optimization of charging urgency, space congestion, and task timeliness. The dynamic consensus factor in the consensus mechanism design module integrates the three elements, enabling the scheduling system to dynamically weigh multiple constraints and avoid decision-making biases in scenarios of "urgent but not crowded" and "crowded but not urgent". In a warehousing environment, an AGV node may report incorrect remaining power or task status due to electromagnetic interference, resulting in misjudgment of priorities by the scheduling system. The cross-node verification mechanism reduces the risk of misjudgment of abnormal data through cross-verification of multi-node data, ensuring that the resource allocation of the swapping station is based on reliable data and avoiding "global congestion caused by a single point of failure".

[0051] The consensus mechanism design module includes: A dynamic consensus factor construction unit, configured to determine an urgency weight reflecting the charging urgency of the AGV, a space conflict entropy value reflecting the space conflict risk of the AGV, and a task timeliness penalty term reflecting the task timeout risk of the AGV, and construct a dynamic consensus factor based on the urgency weight, the space conflict entropy value, and the task timeliness penalty term to provide a priority decision basis for the charging and swapping scheduling optimization of the AGV; Furthermore, the dynamic consensus factor construction unit in the embodiment of the present application is further configured to perform the following steps: Step S2110, determining the urgency weight E; Furthermore, as Figure 3 shown, step S2110 includes: Step S2111, obtaining the current remaining power DE of the AGV; Step S2112, obtaining an activation exponential growth function according to the current remaining power DE of the AGV and the potential margin decay gradient; Step S2113, setting an initial urgency weight E1, when DE≥S1, E = E1; when DE<S1, dynamically increasing the initial urgency weight E1 according to the activation exponential growth function to obtain the urgency weight E, where S1 is the power urgency threshold.

[0052] Specifically, the urgency weight E is a core parameter reflecting the charging urgency of the AGV, and its function is to quantify the urgent need of the AGV for charging resources and provide a priority decision basis for the dynamic consensus factor. In the scenario of centralized charging and swapping of multiple AGVs, judging the charging priority solely based on the absolute value of the remaining power has defects - for example, two AGVs with the same remaining power may have significant differences in actual charging urgency due to different battery aging degrees or energy consumption rates. Step S2110 solves the problem of "static power thresholds unable to reflect dynamic charging demands" by integrating the remaining power and the potential margin decay gradient.

[0053] The state of charge (SoC) of the battery is collected in real time through the AGV battery management system (BMS), and the current remaining power DE is obtained after normalization. The value range of DE is [0, 1], where 0 indicates that the battery is completely depleted and 1 indicates that it is fully charged. This parameter is a basic indicator for measuring the current energy reserve of the AGV. However, using DE alone cannot distinguish between "low power due to high energy consumption conditions" and "low power due to battery aging". Therefore, it is necessary to perform dynamic correction in combination with the potential margin decay gradient. The potential margin decay gradient reflects the potential decay rate after considering battery aging and cumulative energy consumption, and is a key parameter characterizing the available degree of the remaining energy of the battery. The activation exponential growth function takes DE and the potential margin decay gradient as inputs, and its mathematical mapping relationship is determined by analyzing the electrochemical characteristics of the battery: when DE is higher than the power urgency threshold S1 (such as 0.15), the AGV has sufficient time to plan the battery replacement path, and the urgency remains the initial urgency weight E1; when DE is lower than S1, the output value of the function increases exponentially as DE decreases, and the greater the potential margin decay gradient (that is, the faster the battery energy decays), the higher the growth rate. The design logic of the activation exponential growth function is as follows: in the low power state, the rapid potential decay caused by battery aging or high energy consumption will exacerbate the charging urgency, and the exponential growth characteristic can non-linearly amplify the priority of such emergency situations. For example, an AGV has a relatively large potential margin decay gradient due to performing high-frequency turning tasks (high trajectory curvature). Even if its DE is slightly higher than S1, the activation function will still moderately increase the urgency according to the gradient value, avoiding scheduling delays caused by simply relying on the remaining power threshold. Using the exponential growth function to process low power scenarios conforms to the electrochemical characteristics of the battery - when the battery is in a critical state of charge, the potential decay rate increases significantly with the increase of the aging degree. The non-linear mapping can accurately capture this dynamic change and avoid the lag of the fixed threshold.

[0054] Set the initial urgency weight E1 as the reference value when DE ≥ S1. This value is determined through historical data statistics and reflects the demand level of the AGV for charging resources under normal conditions. When DE < S1, perform dynamic correction on E1 according to the activation exponential growth function. Step S2110 avoids the rough scheduling of "only relying on power" - for example, for two AGVs with a DE of 12% each, one has a potential margin decay gradient of 0.8 V / s (high decay rate) due to battery aging, and the other has a gradient of 0.3 V / s (low decay rate) due to a flat path. The urgency weight E of the former will be significantly higher than that of the latter, so it will obtain battery replacement resources first. This differential processing can accurately identify the AGVs that really need to be charged urgently and avoid scheduling unfairness caused by battery performance differences.

[0055] Step S2110 makes up for the defect that a single power threshold cannot reflect the dynamic performance of the battery. It introduces battery aging and energy consumption rate factors through the potential margin attenuation gradient to achieve multi-dimensional quantification of charging urgency. It provides the key input of the "urgency" dimension for the dynamic consensus factor, so that the subsequent power exchange order generation can give priority to high-urgency AGVs and reduce the risk of power outages caused by power exhaustion. The urgency weight is used as the priority basis of the node coloring model. High-weight AGVs are given priority in the power exchange time window in the taboo search algorithm to ensure that urgent tasks are executed first. The dynamic energy consumption prediction value in the energy consumption prediction module provides a dynamic evaluation of the endurance for the urgency weight (for example, even if the power of a high-energy consumption AGV is still good, it may trigger high urgency due to insufficient endurance), and the urgency weight provides a reverse constraint on the scheduling priority for the energy consumption prediction (such as giving priority to the power exchange needs of high-energy consumption AGVs). The combination of the two avoids the waste of resources caused by "power misjudgment".

[0056] Step S2120, determining the spatial conflict entropy value; Furthermore, if Figure 4 As shown, step S2120 includes: Step S2121, based on the three-dimensional coordinates of the AGV in the posture topological feature vector, obtaining the relative distances between the multiple AGVs; Step S2122, introducing a virtual repulsive potential field according to the relative distances between the multiple AGVs, and establishing a congestion model of the AGVs in the operation area of ​​the battery swap station; Step S2123, based on the established crowding model of AGV in the battery swap station operation area, obtain a spatial conflict entropy value reflecting the spatial conflict risk of AGV.

[0057] Specifically, the spatial conflict entropy value is used to quantify the spatial collision risk of multiple AGVs in the operation area of ​​the battery swap station, and is an important component of the dynamic consensus factor. In the limited operating space of the battery swap station (such as the charging station and the robot arm activity area), the relative distance between AGVs is too close, which can easily cause path conflicts, resulting in reduced battery swap efficiency and even equipment damage. Step S2120 converts the spatial conflict risk into a computable entropy value indicator by constructing a congestion model.

[0058] Based on the three-dimensional coordinates of the AGV in the pose topological feature vector, calculate the Euclidean distance between any two AGVs. The calculation of the relative distance needs to consider the key areas of the operation space of the battery swapping station, such as the area within 1 meter around the charging pile position, the boundary of the robotic arm activity area, etc. These areas are high-incidence zones of spatial conflicts. Introduce the virtual repulsive force potential field theory, regarding each AGV as a mass point with repulsive force, and the magnitude of the repulsive force is inversely proportional to the square of the relative distance - that is, the closer the distance, the stronger the repulsive force. The mathematical relationship of the potential field strength function is determined by analyzing the AGV size and the distribution of obstacles in the operation space, ensuring that the repulsive force can be ignored outside the safe distance (such as 0.5 meters), while the repulsive force increases sharply within the dangerous distance (such as 0.2 meters). Introduce the virtual repulsive force potential field to transform the relative distance between AGVs into a non-linear repulsive force, which conforms to the physical constraints of the operation space of the battery swapping station - a small change in distance within a narrow space can trigger a high collision risk, and the non-linear characteristic of the potential field model can amplify this risk, prompting the scheduling system to avoid it in advance.

[0059] The construction process of the congestion model is as follows: Potential field initialization: Define the repulsive force range of each AGV; Repulsive force calculation: For each AGV, accumulate the repulsive force vectors generated by all surrounding AGVs to obtain the resultant repulsive force of the AGV at the current position; Risk quantification: Map the magnitude and direction of the resultant repulsive force to a spatial conflict risk value. The higher the risk value, the more serious the congestion degree of the area.

[0060] Based on the output of the congestion model, use the information entropy theory to normalize the spatial conflict risk, map the resultant repulsive force value to the interval [0, 1], and eliminate the influence of dimensions; calculate the risk distribution uniformity of all AGVs in the battery swapping station area through the Shannon entropy formula. The larger the entropy value, the more concentrated the conflict risk. The spatial conflict entropy value reflects the degree of disorder in the distribution of AGVs in the operation space of the battery swapping station. For example, when multiple AGVs are concentrated within 1 meter in front of the charging pile position, the entropy value increases significantly, indicating the existence of serious potential spatial conflicts.

[0061] Step S2120 converts the conflict risk in the spatial dimension into a quantifiable entropy value index, solving the problem of "difficult to evaluate the spatial position relationship of multiple AGVs in real time" and providing spatial constraint conditions for the scheduling algorithm. As the input of the "spatial conflict" dimension of the dynamic consensus factor, the subsequent spatio-temporal conflict resolution graph can identify high-risk areas and actively avoid spatial overlap when allocating the battery swapping time window. The spatial conflict entropy value guides the setting of the repulsive force source in the artificial potential field method (step S3244). AGVs in high-entropy value areas are given a greater repulsive force weight to avoid entering the dangerous distance during path planning.

[0062] Step S2130, determine the task timeliness penalty term; Furthermore, asFigure 5 As shown, step S2130 includes: Step S2131, obtain the deadline and the current time of the task currently being executed by the AGV; based on the deadline and the current time, obtain the remaining time of the task; Step S2132, introduce an inverse proportional function of the deadline and the current time of the task; the function value of the inverse proportional function increases non-linearly as the remaining time of the task decreases; Step S2133, generate a task timeliness penalty term reflecting the risk of task timeout of the AGV according to the inverse proportional function.

[0063] Specifically, the task timeliness penalty term is a quantitative indicator reflecting the risk of task timeout of the AGV, and its role is to convert the time urgency of the task into a decision-making basis for scheduling priority. In the logistics and warehousing scenario, the AGV may undertake multiple transportation tasks simultaneously, and some tasks have strict requirements for the completion time (such as the temperature control task for fresh food transportation and the immediate material supply task for the production line). If only the power and space factors are considered and the task timeliness is ignored, it may lead to delays in key tasks and affect the overall efficiency of the warehousing system. Step S2130 solves the problem of "the risk of task timeout cannot be quantified" by introducing the inverse proportional relationship between the task deadline and the current time.

[0064] Obtain the deadline T of the task currently being executed by the AGV in real time through the Warehouse Management System (WMS) end (i.e., the latest completion time required by the task) and the current time T now . These two time parameters need to be synchronized with the positioning system and the task allocation system of the AGV to ensure the consistency of the timestamps. For example, an AGV undertakes the task of delivering goods to the sorting port, and the WMS sets the T of this task end as the current time T now plus 30 minutes. Introduce a non-linear mapping function to map the remaining time of the task (T end - T now ) to the task timeliness risk value. The core characteristic of this function is that as T end approaches T now (i.e., the task deadline is approaching), the function value shows a non-linear increasing trend. Using the inverse proportional function to quantify the timeliness risk conforms to the time-sensitive characteristics of warehousing tasks - the tolerance of tasks near the deadline to delays decreases exponentially, and the rapid growth characteristic of the inverse proportional function is consistent with the actual risk change trend, ensuring the immediate response of the scheduling system to emergency tasks.

[0065] The specific construction logic is as follows: Calculation of the remaining time: Define the remaining time of the task ΔT = T end - T nowWhen ΔT ≤ 0, the task has timed out; Non - linear mapping: Adopt the form of inverse proportional function, such as f(ΔT) = 1 / ΔT (ΔT > 0), and handle the case of ΔT ≤ 0 through a piece - wise function (in this case, the function value takes a preset maximum value). The non - linear characteristic of this function can significantly amplify the risk weight of tasks near the deadline. For example, the risk value of a task with 10 minutes remaining is twice that of a task with 20 minutes remaining, and the risk value of a task with 5 minutes remaining is four times that, forming a gradient - based evaluation sensitive to timeliness.

[0066] Normalize the output value of the inverse proportional function to obtain the task timeliness penalty term P. The normalization process needs to combine the task timeliness requirements of the warehousing system. For example, map the function value to the interval [0, 1], where 0 represents no timeout risk and 1 represents that the task is about to time out. Specifically, linearly scale f(ΔT) so that its maximum value does not exceed the risk upper limit preset by the system; when ΔT ≤ 0, P is directly assigned the value of 1 to ensure that timed - out tasks obtain the highest penalty term and trigger the scheduling response first. Step S2130 avoids systematic risks caused by the scheduling strategy ignoring task timeliness. For example, Figures 6 - 7 as shown, Figure 6 in the example, the remaining battery levels of two AGV cars are both 11%, reaching the critical state, and they both go to the battery replacement station at the same time. However, the remaining task time of AGV - 01 is 20 minutes, and the remaining task time of AGV - 02 is 40 minutes. The task of AGV - 01 is more urgent and has a higher priority. Therefore, as Figure 7 shown, give priority to arranging AGV - 01 to enter the battery replacement station for battery replacement to ensure its priority battery replacement and on - time task completion, and avoid the stagnation of the entire production line due to battery replacement delay.

[0067] Step S2140: Generate a dynamic consensus factor by weighted fusion of the urgency weight, spatial conflict entropy value, and task timeliness penalty term.

[0068] Specifically, the dynamic consensus factor is a multi - dimensional priority index that integrates the urgency weight E, spatial conflict entropy value S, and task timeliness penalty term P. By quantifying the comprehensive scheduling priority of AGVs, it provides a sorting basis for the subsequent node coloring model and spatio - temporal replacement algorithm. For example, if an AGV has a high E value (urgent battery), a medium S value (general spatial conflict), and a very high P value (task about to time out), its dynamic consensus factor C will be significantly higher than that of other AGVs, and it will be preferentially allocated a battery replacement time window in the scheduling.

[0069] The dynamic consensus factor construction unit solves the key problem of "single priority decision basis" in the charging and swapping scheduling of multiple AGVs by constructing a dynamic consensus factor. Traditional scheduling algorithms usually only use the remaining battery power as the priority basis, without considering the impact of battery aging and energy consumption rate on the urgency of charging, without quantifying the collision risk of multiple AGVs near the swapping station, and without incorporating the impact of task timeliness on the scheduling priority. However, the dynamic consensus factor construction unit in this embodiment realizes a technical leap from "single-dimensional power priority" to "multi-modal comprehensive evaluation" through the fusion of three factors. The dynamic consensus factor ensures that the scheduling system can synchronously respond to multi-dimensional constraints such as power urgency, space safety, and task timeliness. As the input of the node coloring model, the dynamic consensus factor enables the tabu search algorithm to allocate the charging and swapping time window according to the priority order of "high urgency, low space conflict risk, and high task timeliness", avoiding the preemption of swapping station resources caused by priority chaos. The priority dimension of the dynamic consensus factor is combined with the time dimension of the space-time folding unit to form a three-dimensional constraint of "priority-time-space" in the six-dimensional state matrix, enabling the space-time conflict resolution map of the swapping station to optimize both the time window allocation and the space path planning simultaneously, and avoiding the global sub-optimal solution caused by single-dimensional optimization.

[0070] If the dynamic consensus factor construction unit is missing in the fully automatic AGV car charging and swapping collaborative optimization system and only the battery power is used as the basis, it may cause AGVs with aging batteries but sufficient power to preempt resources, while AGVs with urgent tasks but sufficient power are forced to wait, resulting in task timeouts; lacking the space conflict entropy value, the scheduling system cannot identify crowded areas near the swapping station, which may lead to multiple AGVs entering a narrow space simultaneously, causing robotic arm collisions or path blockages; without incorporating task timeliness, the task scheduling of the warehousing system is separated from the AGV charging and swapping strategy, and the timeliness of key tasks cannot be guaranteed.

[0071] The dynamic consensus factor construction unit constructs a dynamic consensus factor covering power urgency, space safety, and task timeliness through the multi-dimensional fusion of the urgency weight, space conflict entropy value, and task timeliness penalty term, solving the priority decision problem in the AGV charging and swapping scheduling in complex warehousing scenarios. As the key hub connecting front-end data collection and back-end time-space scheduling, the dynamic consensus factor construction unit enables the fully automatic AGV car charging and swapping collaborative optimization system to respond to multiple constraints in real time, avoiding scheduling deviations caused by single indicators, fundamentally improving the collaborative efficiency of multi-AGV charging and swapping, ensuring the orderly operation of the swapping station in high-frequency usage scenarios, and providing core technical support for the intelligent upgrade of the warehousing logistics system.

[0072] The dynamic consensus factor verification unit is used to build a distributed collaborative verification layer based on the Byzantine fault tolerance mechanism, and cross-node verification of the constructed dynamic consensus factor is performed through the distributed collaborative verification layer to obtain a credible dynamic consensus factor.

[0073] Specifically, in the logistics and warehousing scenario, when multiple AGVs report dynamic consensus factors through wireless communication, problems such as data transmission errors, node failures, or malicious tampering may occur, resulting in the scheduling system receiving incorrect priority information and causing chaos in the battery swapping order. The distributed collaborative verification layer adopts the Byzantine Fault Tolerance (BFT) mechanism. Its core is to ensure data consistency through a voting arbitration mechanism on the premise that some nodes may send incorrect or false data. A star-shaped communication network is established, and each AGV node sends the dynamic consensus factor to the central scheduling node. At the same time, the nodes broadcast data to each other to form a cross-verification loop. After the central scheduling node collects the dynamic consensus factors of all nodes, a majority vote check is performed on the factor values of each AGV. For example, when the urgency weight reported by a certain node significantly deviates from the average value of other nodes (exceeding the preset threshold), it is determined as abnormal data, and a retransmission or isolation mechanism is triggered. The fault tolerance threshold is set according to the scale of the battery swapping station. For example, when the proportion of faulty nodes is less than 33%, consensus can still be reached through the valid data of the remaining nodes, ensuring the reliability of the scheduling strategy in case of partial equipment failures.

[0074] Based on the BFT mechanism, the Isolation Forest algorithm is introduced to perform a secondary screening on the verified dynamic consensus factors. The three dimensions of the dynamic consensus factor (urgency weight, spatial conflict entropy value, task timeliness penalty term) are used as input features to form a multi-dimensional feature vector. Multiple decision trees are constructed by randomly selecting feature attributes and split points, and the average path length of each data point in the decision tree is calculated. The path length of abnormal data (such as extreme values caused by sensor failures) is significantly shorter than that of normal data, so it can be effectively identified. The Isolation Forest model is incrementally trained regularly with the latest reliable data to adapt to dynamic scenarios such as AGV battery aging and task type changes, avoiding a decrease in detection accuracy due to environmental changes.

[0075] If the dynamic consensus factor verification unit is missing, data inconsistency may occur, leading to scheduling chaos. For example, an AGV reports an incorrect high urgency weight due to a sensor failure, resulting in the system incorrectly allocating priorities and causing resource preemption at the battery swapping station; there is also a risk of malicious node attacks. For example, in a wireless communication environment, false data may be injected into the system, causing key task AGVs to be unable to swap batteries in a timely manner and affecting the continuity of warehousing operations. The root cause of these problems lies in the lack of a data credibility guarantee mechanism, making the priority decision basis of the dynamic consensus factor unreliable and ultimately resulting in the failure of the battery swapping order and path planning.

[0076] The dynamic consensus factor verification unit constructs a reliable data verification system in a distributed system through the Byzantine Fault Tolerance mechanism and the Isolation Forest algorithm, ensuring the reliability of the dynamic consensus factor and solving the data consistency problem in multi-node communication.

[0077] The AGV scheduling optimization module is used to discretize the operation space of the battery swapping station to generate spatio-temporal folding units; based on the dynamic energy consumption prediction value, the credible dynamic consensus factor, and the spatio-temporal folding units, construct a spatio-temporal conflict resolution map of the battery swapping station, and generate the optimal battery swapping order and collision avoidance path of the AGV; The scheduling of the battery swapping station involves four-dimensional constraints of three-dimensional space (length × width × height) and time dimension. The computational complexity of directly solving the optimal solution by traditional methods increases exponentially with the number of AGVs (for example, the number of time window combinations of 5 AGVs reaches 120, and that of 10 AGVs reaches 3,628,800), which cannot meet the real-time requirements; the spatio-temporal folding theory in the AGV scheduling optimization module transforms the problem into polynomial-level complexity through a compression mapping that maintains topological homeomorphism, and is applicable to the working conditions where 10 - 50 AGVs swap batteries simultaneously, which is common in warehousing scenarios. Traditional methods separately process the spatial layout and time scheduling, resulting in "spatially feasible but time-conflicting" (such as the charging pile position being occupied when the AGV arrives) or "time-ordered but spatially colliding" (such as the distance between parking positions being less than the safety threshold). The six-dimensional state matrix in the AGV scheduling optimization module integrates all constraints, and the occupancy interval needs to simultaneously meet the available time of the charging pile position, the robotic arm having no task, and the safety distance from the previous AGV (such as ≥ 2 meters), ensuring that the resource allocation is completely consistent in the spatio-temporal dimension.

[0078] The AGV scheduling optimization module includes: The spatial discrete unit is used to discretize the operation space of the battery swapping station to generate spatio-temporal folding units; Furthermore, the spatial discrete unit in the embodiment of the present application is further used to perform the following steps: Step S3110: Extract the four-dimensional spatio-temporal coordinates of the resource elements of the battery swapping station to form the operation space state matrix of the battery swapping station; the four-dimensional spatio-temporal coordinates introduce the time dimension on the basis of the three-dimensional Euclidean space; the resource elements of the battery swapping station include charging pile positions, robotic arm activity areas, and AGV parking positions; Step S3120: Use the spatio-temporal folding theory to perform a compression mapping on the operation space state matrix of the battery swapping station to generate spatio-temporal folding units that reflect the spatio-temporal dependence relationship between the resource elements of the battery swapping station.

[0079] Specifically, the efficient utilization of the operation space of the battery swapping station depends on accurately modeling the spatio-temporal states of the resource elements. The spatial discrete unit combines the three-dimensional space and the time dimension to solve the problem of "difficulty in quantifying spatio-temporal conflicts during the battery swapping process of multiple AGVs".

[0080] The four-dimensional space-time coordinates introduce the time dimension on the basis of the three-dimensional Euclidean space. The three-dimensional space coordinates take the geometric center of the battery swapping station as the origin, the X-axis is along the arrangement direction of the charging piles, the Y-axis is perpendicular to the ground upwards, and the Z-axis points to the direction of the battery swapping station entrance, accurately describing the spatial positions of the charging pile positions, the robotic arm activity areas, and the AGV parking positions; the time dimension records the occupancy time intervals of each resource element (such as a certain AGV occupying a charging pile position during the time period from t1 to t2). For each resource element, its four-dimensional coordinates are recorded to form a state matrix of the battery swapping station operation space with a dimension of M×4, where M is the total number of resource elements. This matrix is updated in real time to reflect the spatio-temporal occupancy status of each area in the battery swapping station. Traditional three-dimensional space modeling cannot distinguish the occupancy status of the same position at different times, resulting in the inability to identify conflicts when the time windows of multiple AGVs overlap; after introducing the time dimension, spatio-temporal occupancy analysis can be accurate to the millisecond level, for example, determining whether two AGVs use the same charging pile position in adjacent time slices.

[0081] Adopt the space-time folding theory to map the high-dimensional space-time state into a low-dimensional computable unit. Specifically, identify the space-time dependence relationships in the battery swapping station operation space. For example, the occupancy of the robotic arm activity area is necessarily accompanied by the use of adjacent charging pile positions, forming space-time correlation constraints; perform dimensionality reduction on the state matrix through principal component analysis (PCA), retain the main feature components of the space-time correlation, and generate space-time folding units. Each unit contains a set of space-time constraint conditions (such as "when charging pile position A is occupied at time t, adjacent parking position B is unavailable within the time period of t±Δt"); ensure that the folding transformation does not change the space-time conflict logic of the resource elements. For example, the minimum safety distance constraint of the AGV in the robotic arm activity area remains effective after folding. There are strong space-time correlations among the resource elements in the battery swapping station operation space (such as the battery swapping process must sequentially use the parking position, the robotic arm, and the charging pile). The folding transformation enables the low-dimensional model to still reflect the real conflict logic by retaining these correlation features, significantly reducing the computational complexity.

[0082] Spatial discrete units combine the spatial occupancy of AGVs (such as the occupancy of parking positions) with time windows (such as the start / end times of battery swapping) through four-dimensional spatio-temporal coordinates to form a basis for conflict judgment of "integration of space and time"; spatio-temporal folding units compress complex four-dimensional state matrices into low-dimensional features to avoid the infeasibility of algorithms caused by the curse of dimensionality. As a basic data structure, spatio-temporal folding units are combined with dynamic energy consumption prediction values and credible dynamic consensus factors to form a six-dimensional state matrix including space, time, priority, and energy consumption, providing an underlying data model for the spatio-temporal conflict resolution map; the priority dimension in the spatio-temporal folding unit is directly related to the credible dynamic consensus factor to ensure that high-priority AGVs preferentially occupy resources in the state matrix after spatio-temporal folding, such as reserving the time window of the charging pile position in advance; the addition of the energy consumption dimension enables the spatio-temporal folding unit to reflect the battery life requirements of AGVs. For example, AGVs with low predicted energy consumption values are assigned later battery swapping time windows to balance resource utilization efficiency. The spatio-temporal constraint conditions (such as the time occupancy of the robotic arm activity area) in the spatio-temporal folding unit directly guide the setting of the gravitational source and repulsive source of the artificial potential field method to ensure that the AGV paths have no conflicts in both the spatial and temporal dimensions.

[0083] If the spatial discrete unit is missing, the space planning is only based on the static layout and cannot dynamically perceive the time occupancy status of the charging pile positions, resulting in multiple AGVs applying for the same resource simultaneously; lacking spatio-temporal correlation constraints, the node coloring model cannot accurately judge the spatio-temporal conflicts between AGVs, and the battery swapping order and path planning lose the key basis, ultimately leading to congestion in the battery swapping station.

[0084] The scheduling decision generation unit constructs a spatio-temporal conflict resolution map for the battery swapping station based on the dynamic energy consumption prediction value, the credible dynamic consensus factor, and the spatio-temporal folding unit, and generates the optimal battery swapping order and collision avoidance path for the AGV. Furthermore, the scheduling decision generation unit in the embodiment of the present application is further configured to perform the following steps: Step S3210: Map the dynamic energy consumption prediction value and the credible dynamic consensus factor to the spatio-temporal folding unit to form a six-dimensional state matrix; where the six dimensions refer to the three dimensions of space, the time dimension, the priority dimension, and the energy consumption dimension.

[0085] Specifically, the dynamic energy consumption prediction value generated by the energy consumption prediction module (reflecting the endurance ability of the AGV) and the credible dynamic consensus factor obtained by the consensus mechanism design module (characterizing the scheduling priority) are passed through the spatio-temporal folding unit generated by the dimensionality expansion embedding space discrete unit (including the spatio-temporal dependence relationship of the swapping station resources). The specific operation is as follows: The three dimensions of space follow the swapping station coordinate system defined by the space discrete unit to accurately describe the spatial coordinates of the AGV at the charging pile position, the robotic arm activity area, and the parking position; the time dimension records the time interval when the AGV enters / leaves the swapping station, which is aligned with the time stamp of the spatio-temporal folding unit; the priority dimension directly uses the verified credible dynamic consensus factor value in the consensus mechanism design module. The higher the value, the higher the scheduling priority (for example, the AGV with a high urgency weight, a low spatial conflict risk, and an urgent task timeliness has the highest priority); the energy consumption dimension normalizes the dynamic energy consumption prediction value obtained by the energy consumption prediction module to the interval [0, 1]. The smaller the value, the tighter the remaining endurance (for example, 0.1 means the endurance is about to run out, and 0.9 means the endurance is sufficient). For each AGV, its current pose coordinates, the predicted swapping time window, the dynamic consensus factor, and the energy consumption prediction value are extracted to form a six-dimensional vector. The six-dimensional vectors of all AGVs constitute the global six-dimensional state matrix of the swapping station. This matrix is updated in real time to comprehensively depict the spatio-temporal occupancy, priority requirements, and energy consumption status of each AGV in the swapping station.

[0086] Step S3220, based on the six-dimensional state matrix, construct a spatio-temporal conflict resolution map of the swapping station through the variational Bayesian inference method; each AGV corresponds to a node in the spatio-temporal conflict resolution map of the station; Specifically, step S3220 uses a probabilistic graphical model to model the spatio-temporal conflicts among AGVs. The specific implementation is as follows: Define nodes, where each node represents the six-dimensional state of an AGV. The node attributes include spatial coordinates, time window, priority, and energy consumption value; establish edges. If the spatial distance between two AGVs is less than the safety threshold (such as 1.5 meters) or the time windows overlap by more than a preset time (such as 5 seconds), then an undirected edge is established between the corresponding nodes, indicating the existence of spatio-temporal conflicts; train the conflict probability between nodes through historical swapping data. For example, "when the priority difference between two AGVs exceeds the threshold and the spatial distance is less than 1 meter, the conflict probability increases to 80%". In view of the high-dimensional characteristics of the six-dimensional state matrix, the mean field variational method is used to approximately decompose the joint probability distribution into the product of marginal distributions of each dimension, and the KL divergence is minimized through iterative optimization to approximately solve the conflict probability distribution in the high-dimensional space. This method avoids the curse of dimensionality problem of traditional Monte Carlo simulations, reduces the computational complexity from exponential to polynomial level, and is suitable for real-time scheduling scenarios.

[0087] In the scenario of centralized battery swapping for multiple AGVs, without the six-dimensional state matrix and variational Bayes, the dynamic energy consumption, priority, and spatio-temporal occupancy data cannot be coordinated, which may cause the AGV with tight battery life to lose power due to the lag in time window allocation. Traditional methods only rely on static spatial distance to judge conflicts and cannot dynamically evaluate the comprehensive impact of priority and time window overlap (for example, high-priority AGVs should be given priority to use the conflict area). The six-dimensional state matrix in step S3220 provides a unified data framework, and variational Bayes inference realizes the probability quantification of multi-dimensional conflicts, enabling the scheduling system to dynamically balance "resource occupancy - priority - battery life demand". For example, it preferentially allocates conflict-free time windows for high-priority AGVs with insufficient battery life, avoiding system congestion caused by single-dimensional decision-making.

[0088] Step S3230: Based on the spatio-temporal conflict resolution map of the battery swapping station, iteratively search for the optimal battery swapping order through the spatio-temporal replacement algorithm; Furthermore, as Figure 8 shown, step S3230 includes: Step S3231: Construct a node coloring model according to the spatio-temporal conflict resolution map of the battery swapping station; Step S3232: According to the node coloring model, use the tabu search algorithm to traverse and color the nodes of the map according to the credible dynamic consensus factor; Step S3233: Gradually eliminate the adjacent nodes of the same color in the spatio-temporal conflict resolution map of the battery swapping station through the spatio-temporal replacement algorithm; Step S3234: Through iterative optimization, output the final node coloring scheme, and the final node coloring scheme corresponds to the optimal battery swapping order of the AGV.

[0089] Specifically, based on the spatio-temporal conflict resolution map of the battery swapping station, each AGV corresponds to a node in the map, and each node represents an AGV in a specific spatio-temporal state; define "color" as the battery swapping time window (for example, color A corresponds to 10:00 - 10:15, color B corresponds to 10:15 - 10:30), and establish a constraint rule: "adjacent nodes (AGVs with spatio-temporal conflicts) are prohibited from being dyed the same color", construct a node coloring model, which transforms the AGV scheduling problem into a classic "graph coloring problem", and use the mature node coloring algorithm in graph theory to solve the conflict-free time window allocation scheme to ensure that conflict-free AGVs occupy the same or adjacent resources within the same time window. By extracting the spatio-temporal occupancy intervals, calculate the spatial distance and time window overlap degree between any two AGVs. If the spatial distance is less than the distance safety threshold (such as 2 meters) and the time window overlap exceeds the set time (such as 5 minutes), it is determined as an adjacent node, and a different color constraint needs to be imposed. Transform the spatio-temporal occupancy of AGVs into the relationship between graph nodes and edges, and the existence of edges represents spatio-temporal conflicts, which is convenient for using mature graph coloring algorithms (such as the Welch-Powell algorithm) to resolve conflicts. For example, the connecting edges between AGVs with a spatial distance less than 2 meters and overlapping time windows are forced to have different colors (time windows). The node attributes integrate dynamic consensus factors (priorities) and energy consumption prediction values (battery swapping time requirements), so that color allocation not only satisfies non-overlapping time, but also can give priority to ensuring AGVs with high urgency and high timeliness requirements, and solve the defect of "one-size-fits-all" scheduling.

[0090] According to the AGV scheduling priority characterized by the credible dynamic consensus factor, all nodes are sorted from high to low according to the priority. A taboo list is established to record the time window combinations allocated recently, avoiding repeated search for invalid solutions. For example, if AGV-01 has been allocated the red time window (14:00-14:10), then it is prohibited to allocate red to adjacent nodes in subsequent searches, and this operation is recorded in the taboo list for a certain number of iterations (such as 10 times). Starting from the high-priority nodes, all feasible time windows are traversed (that is, not occupied by adjacent nodes and meeting the lower limit of battery replacement time), and the color allocation scheme with the least global conflict is selected. For example, AGVs with a high urgency weight are preferentially allocated the earliest available time window to ensure that they can charge as soon as possible to avoid power failure during operation. The taboo search algorithm efficiently searches for near-optimal solutions in the solution space by combining the priority sorting and taboo list mechanisms, balancing the search speed and the quality of the solution, and is especially suitable for the complex scenario of multi-AGV dynamic scheduling. During the search process, the tried invalid solutions are recorded to prevent the algorithm from falling into a loop search. For example, it is avoided to repeatedly exchange the time windows of the same pair of AGVs without progress. When a new AGV enters the battery replacement station area or the state of an existing AGV changes (such as a sudden drop in battery power), the scheme is adjusted through local replacement instead of global recalculation to improve the system response speed, which is suitable for the real-time scheduling requirements of the dynamic entry and exit of AGVs in the warehousing environment. The node priority sorting directly uses the credible dynamic consensus factor, which integrates the battery power urgency, space conflict risk, and task timeliness, enabling the coloring process to give priority to high-priority AGVs. For example, AGVs with a battery power lower than 15% and carrying urgent orders are preferentially allocated time windows to reduce the power failure risk caused by their waiting. The space-time folding unit compresses the four-dimensional space-time coordinates of the battery replacement station resources, providing a standardized space-time occupancy description for node coloring, enabling the time window allocation of different AGVs to uniformly refer to the available time of resources such as charging pile positions and robotic arm activity areas.

[0091] Real-time scanning node coloring model to identify all adjacent node pairs of the same color (i.e., AGVs that have spatio-temporal conflicts but are assigned the same time window). For example, it is detected that both AGV-02 and AGV-03 are assigned the blue time window, and the distance between their parking positions in the battery swapping station is less than the distance safety threshold (1.5 meters), then they are determined as a conflict node pair. For conflict node pairs, the time window of the AGV with a higher priority is preferentially retained, and an available window is re-assigned to the AGV with a lower priority (e.g., AGV-02 keeps the blue window, and AGV-03 is adjusted to the green window), and the conflict status after the replacement is re-checked. When replacing, the minimum interval requirement of the time window needs to be met (e.g., the interval between consecutive battery swapping operations is at least 5 minutes to ensure that the previous AGV completely leaves the battery swapping station). Repeat the conflict detection and replacement operations until the colors of all adjacent nodes are different, that is, there is no time window overlap and the spatial occupancy is safe without conflicts. The spatio-temporal replacement algorithm gradually eliminates global conflicts by locally adjusting the time window, ensuring that the battery swapping operations are executed orderly in the time dimension and avoiding the problem of spatial resource preemption caused by time overlap.

[0092] Set the maximum number of iterations (e.g., 200 times) or the minimum conflict threshold (e.g., the number of conflicts remains unchanged for 10 consecutive iterations), and terminate the search when either condition is met. For example, after 150 iterations, the number of conflict node pairs no longer decreases, which is regarded as reaching a local optimal solution. Map the color order in the final node coloring scheme to the order of the battery swapping time windows. For example, red (14:00 - 14:10) corresponds to the AGV with the first priority, blue (14:10 - 14:20) corresponds to the AGV with the second priority, and so on, to form a battery swapping sequence arranged in chronological order. If it is detected during the iteration that the battery level of a certain AGV is lower than the emergency threshold (e.g., 10%), automatically increase its node priority and forcefully assign the nearest available time window to ensure the system robustness in case of emergencies.

[0093] The core goal of the fully automatic AGV charging and battery swapping collaborative optimization system is to avoid congestion when multiple AGVs are centralized for battery swapping. Step S3230 specifically solves the time window conflict problem, priority scheduling imbalance and high solution space complexity. The time window conflict problem, that is, when multiple AGVs apply for battery swapping at the same time, directly allocating time windows is prone to overlap, causing competition for spatial resources such as robot arm preemption and parking space conflicts. For example, if two AGVs enter the battery swapping station at the same time, the channel may be blocked because the robot arm can only serve one. Priority scheduling imbalance, that is, the traditional scheduling method only sorts according to the power supply, without considering multi-dimensional factors such as task timeliness and spatial congestion, may cause emergency task AGVs (such as vehicles carrying goods that need to be shipped out of the warehouse urgently) to be delayed. The solution space complexity is high, that is, the scheduling of the battery swapping station involves four-dimensional constraints of three-dimensional space and time. The computational complexity of directly solving the optimal solution increases exponentially with the number of AGVs (NP-hard problem), and an efficient heuristic algorithm is required to reduce the computational cost. By converting the scheduling problem into a node coloring problem, step S3230 uses graph theory and heuristic algorithms to generate a near-optimal solution within a reasonable time, avoiding the computational explosion problem of traditional precise algorithms.

[0094] Through the "different colors of adjacent nodes" constraint, it is ensured that the battery swap time windows of any two AGVs do not overlap, or the spatial distance is greater than the safety threshold when overlapping. For example, when AGV-01 uses charging station position 1 from 14:00 to 14:10, the time window of AGV-02 is automatically adjusted to after 14:10 to avoid both occupying the same area at the same time, reduce the competition between the robot arm and the parking space, and improve the resource utilization rate of the battery swap station. The time-space replacement algorithm allows dynamic adjustment of the time window. When an AGV is delayed in arriving at the battery swap station due to path planning deviation, resources can be reallocated by exchanging time windows without recalculating the entire scheduling plan. For example, AGV-03 arrived 1 minute late due to shelf obstruction. The system automatically replaced its time window with the subsequent window of AGV-04 to ensure that the overall scheduling process is not interrupted and avoid global failure of traditional fixed scheduling plans due to local disturbances.

[0095] If step S3230 is missing, the system cannot handle the time window conflicts of multiple AGVs, which may cause multiple AGVs to enter the battery swap station at the same time, causing robot arm preemption, parking space blockage, and even physical collisions, which may cause the battery swap station to be paralyzed in serious cases. In the absence of node sorting based on dynamic consensus factors, emergency AGVs may be blocked by low-priority task AGVs. For example, an AGV with only 10% battery power may lose power while waiting for the unloaded AGV in front to swap power, increasing the cost of manual intervention. Directly solving the optimal solution of the four-dimensional spatiotemporal mechanism requires traversing all possible time window combinations, which cannot be completed within a reasonable time, resulting in scheduling delays and exacerbating the risk of congestion.

[0096] Step S3230 systematically solves the time window conflict and priority scheduling problems during the battery swapping of multiple AGVs by constructing a node coloring model and combining the tabu search and spatio-temporal permutation algorithms. It transforms the complex spatio-temporal resource allocation into a node coloring problem in graph theory, and uses heuristic algorithms to generate an efficient and ordered battery swapping sequence within a reasonable time, ensuring that high-priority AGVs are charged first and avoiding spatial resource competition. It realizes the calculation from priority to the actual time window allocation, provides a constraint basis in the time dimension for the subsequent collision avoidance path planning, and fundamentally improves the scheduling efficiency and reliability of the battery swapping station in the scenario of multiple AGVs with critical battery levels, avoiding congestion and resource waste caused by disordered battery swapping.

[0097] Step S3240 generates the collision avoidance path of AGVs in the battery swapping station based on the spatio-temporal conflict resolution map of the battery swapping station through a multi-granularity trajectory planning method; Furthermore, step S3240 includes: Step S3241 extracts the spatio-temporal occupancy intervals of each AGV from the spatio-temporal conflict resolution map of the battery swapping station; Step S3242 determines the scheduled battery swapping time window sequence of each AGV in combination with the optimal battery swapping sequence of the AGV; Step S3243 takes the time granularity as the first priority, performs sequential constraint and interval extension on the scheduled battery swapping time windows of each AGV to obtain the planning result in the time dimension; Step S3244 takes the space granularity as the second priority, searches for local paths through the artificial potential field method within each time slice to obtain the planning result in the space dimension; the time slice is obtained by dividing the operation space of the entire battery swapping station into a multi-dimensional grid map; Step S3245 synthesizes the planning results in the time dimension and the space dimension to generate the collision avoidance path of AGVs in the battery swapping station.

[0098] The optimal battery swapping sequence and the collision avoidance path of the AGV are sent to each AGV node.

[0099] Specifically, using the spatio-temporal conflict resolution map of the battery swapping station as the data source, for each AGV, extract the three-dimensional space area it occupies in the operation space of the battery swapping station (including the spatial positions of resource elements such as charging pile positions, robotic arm activity areas, AGV parking positions, etc.) and the corresponding time interval (i.e., the start time and end time of the battery swapping operation of the AGV in the battery swapping station), forming a unique spatio-temporal occupancy interval for each AGV. This interval reflects the spatio-temporal occupancy status of the AGV in the battery swapping station. The spatio-temporal occupancy status clarifies the exclusive or shared status of the AGV for the resources in the battery swapping station at a specific time. It provides precise spatio-temporal constraints for subsequent time window allocation and path planning to ensure that the resource usage of multiple AGVs in the battery swapping station does not overlap and conflict. The spatio-temporal occupancy interval clarifies when, where, and which resources each AGV occupies (e.g., charging pile position 1 is occupied by AGV-01 from 14:00 to 14:10), providing a direct basis for time window allocation. When the spatio-temporal occupancy intervals of two AGVs overlap spatially and partially overlap temporally, it is determined as a conflict, triggering the spatio-temporal replacement algorithm in step S3233 to adjust the time window to ensure resource exclusivity.

[0100] Take the optimal battery swapping sequence output by step S3230 as the constraint condition in the time dimension to clarify the order of each AGV entering the battery swapping station for battery swapping operations; according to this order, allocate a predetermined battery swapping time window for each AGV. The start and end times of the time window need to ensure sufficient safety spacing in time between the front and rear AGVs to avoid spatial usage conflicts caused by time overlap during the battery swapping process. Through the orderly allocation of time windows, ensure that the battery swapping operations form a strict priority queue in the time dimension, reducing resource competition caused by time overlap. In the time dimension, strictly sort the predetermined battery swapping time windows of each AGV according to the optimal battery swapping sequence to ensure that the battery swapping operations are carried out in order of priority from high to low; at the same time, set a reasonable time interval (by means of interval extension) between every two adjacent AGVs for battery swapping. This time interval needs to be greater than or equal to the shortest time required for the AGV to enter the battery swapping station, complete battery swapping preparation, perform battery swapping operations, and leave the battery swapping station to ensure sufficient safety spacing in time between the front and rear AGVs and completely eliminate the possibility of collision from the time dimension. The time dimension planning fundamentally eliminates the possibility of multiple AGVs simultaneously occupying the core areas of the battery swapping station (such as charging pile positions, robotic arm activity areas) through quantified safety spacing and priority order. Taking the time window order as the first constraint, ensure that the battery swapping operations of high-priority AGVs are not interfered with by low-priority AGVs. For example, the time window of an AGV with only 10% battery power will not be delayed by the path planning of other AGVs.

[0101] Divide the entire operation space of the battery swapping station into multiple grids, with each grid corresponding to a specific spatial position and time slice (e.g., 1 minute / slice); within each time slice, for the AGVs currently in the battery swapping station, taking its own kinematic constraints (such as maximum speed, minimum turning radius, etc.) as limiting conditions, adopt the artificial potential field method, set target positions such as charging pile positions, robotic arm activity areas, and AGV parking positions as gravitational sources, set other AGVs, obstacles, etc. as repulsive sources, calculate the resultant force received by the AGV at the current grid position, search for the locally optimal path from the current position to the target position according to the direction of the resultant force, so that the AGV can flexibly avoid other AGVs and obstacles in space. The gravitational source refers to target positions such as charging pile positions and AGV parking positions, which generate an attractive force on the AGV, and the magnitude of the gravitational force is inversely proportional to the distance, ensuring that the AGV preferentially moves towards the target position; the repulsive source refers to the grids currently occupied by other AGVs and obstacles (such as shelf pillars), which generate a repulsive force, and the magnitude of the repulsive force is inversely proportional to the square of the distance, to avoid collisions. The artificial potential field method enables the AGV to autonomously avoid dynamic obstacles in a complex spatial environment by simulating the interaction of physical fields, improving the real-time performance and flexibility of path planning. The artificial potential field method independently calculates the local path within each time slice, avoiding the complexity of global path planning, and is applicable to real-time obstacle avoidance in the narrow space of the battery swapping station (e.g., when AGVs are moving towards each other or approaching closely in a 1.5-meter-wide passage, the direction is automatically adjusted through the repulsive source).

[0102] Integrate the pre-planned battery swapping time window sequence and interval extension results in the time dimension with the local paths obtained by the artificial potential field method search within each time slice in the spatial dimension; in chronological order, connect the local paths within each time slice in sequence to form a complete trajectory for the AGV from entering the battery swapping station to completing the battery swapping and leaving. This trajectory meets the battery swapping sequence and safety interval requirements in terms of time and the collision avoidance requirements in terms of space, which is the final AGV collision avoidance path.

[0103] Traditional path planning only considers spatial obstacle avoidance and ignores resource occupancy in the time dimension (such as a charging pile position being occupied at a specific time), resulting in the AGV waiting due to busy resources after arriving at the battery swapping station, exacerbating congestion. The AGVs in the battery swapping station are in a dynamic moving state (such as entering, charging, leaving), and need to adjust the path in real time to avoid collisions with other AGVs and fixed facilities such as robotic arms and charging piles. High-priority AGVs need to use resources preferentially, but it is difficult for traditional methods to simultaneously meet the dual constraints of priority order and spatial safety in path planning. Step S3240 solves the above problems. By combining the priority order in the time dimension with local obstacle avoidance in the spatial dimension, step S3240 realizes a battery swapping process of "orderly in time and conflict-free in space", solving the problem of the separation of spatio-temporal constraints in traditional methods.

[0104] Through the interval extension of the time window (such as setting a 15-minute safety interval) and the artificial potential field method for obstacle avoidance in the spatial path, it is ensured that the spatial distance between any two AGVs in the swapping station is always greater than the distance safety threshold (2 meters), and the time windows do not overlap at all. For example, when AGV-01 occupies charging pile position 1 from 14:00 to 14:10, the time window of AGV-02 is assigned to 14:15 - 14:25, and its path planning automatically avoids the spatial area of charging pile position 1 to prevent mechanical arm collisions caused by approaching the charging pile simultaneously. The priority sorting in the time dimension ensures that AGVs with high urgency can swap batteries first, reducing their waiting time, while the local optimization in the spatial dimension shortens the movement time of AGVs in the swapping station (such as finding the shortest obstacle avoidance path through the artificial potential field method), improving the overall battery swapping efficiency. When an AGV deviates from its position due to sensor errors, the repulsive force source of the artificial potential field method will be adjusted in real time to force the AGV back to the planned path; if the battery swapping time is extended due to battery aging, the interval extension mechanism will automatically adjust the time windows of subsequent AGVs to avoid chain congestion. For example, if the charging time of AGV-03 is 5 minutes more than expected, the system will automatically postpone the time window of AGV-04 by 5 minutes and adjust its entry path through the artificial potential field method to avoid AGV-03 which is still charging. Combining the priority sorting with the dynamic consensus factor and the resource constraints of the space-time occupancy interval, the path planning not only satisfies the scheduling principle of "urgent first, then slow", but also ensures that the movement trajectory of each AGV conforms to its physical characteristics (such as the minimum turning radius, maximum speed), avoiding mechanical losses caused by unreasonable paths. For example, AGVs with a small curvature radius (such as model A-02) are automatically assigned wider turning grids during path planning to reduce tire wear.

[0105] If step S3240 is missing, there will be a lack of safety interval in the time window and spatial obstacle avoidance planning, and AGVs may have physical collisions in the swapping station, especially in narrow spaces such as the mechanical arm activity area, which may cause equipment damage or the swapping station to shut down. Multiple AGVs may simultaneously seize the charging pile positions, triggering a "deadlock" phenomenon (such as two AGVs driving towards the same charging pile at the same time and both being unable to complete the docking), resulting in a more than 50% decrease in the processing capacity of the swapping station. Even if step S3230 generates the optimal battery swapping order, without the time and space constraints of path planning, high-priority AGVs may be unable to swap batteries on time due to path congestion, leading to their power depletion and exacerbating system-level congestion.

[0106] Step S3240 uses a multi-granularity trajectory planning method to organically combine the priority order in the time dimension with real-time obstacle avoidance in the space dimension, systematically solving the spatio-temporal conflict problem of multiple AGVs in the battery swapping station. Through the accurate extraction of spatio-temporal occupancy intervals, the orderly allocation of time windows, and the local path optimization of the artificial potential field method, the dual goals of "swapping batteries in an orderly manner according to priority in time and moving safely without collision in space" are achieved. This ensures the feasibility of the scheduling scheme on the ground, forms a complete closed-loop of "priority sorting - path execution" with the node coloring model in step S3230, improves the operation efficiency and safety of the battery swapping station under complex working conditions, avoids equipment collisions and process chaos caused by the lack of spatio-temporal planning, and provides a reliable technical guarantee for the efficient coordination of multiple AGV charging and battery swapping in logistics warehousing.

[0107] Embodiment 2: Based on Embodiment 1, this embodiment provides a method for collaborative optimization of full-automatic AGV car charging and battery swapping, as Figure 9 shown, including: Step S1000, obtain the pose topological feature vector and potential margin attenuation gradient of the AGV car, construct a multi-dimensional input vector, and generate an adaptive dynamic energy consumption prediction value based on the multi-dimensional input vector and a pre-constructed energy consumption trajectory residual model coupled with multiple physical fields; Further, step S1000 includes: Step S1100, obtain the pose topological feature vector and potential margin attenuation gradient of the AGV car, and construct a multi-dimensional input vector; Further, step S1100 includes: Step S1110, obtain the pose topological features of the AGV car in real time to form a pose topological feature vector; the pose topological features include three-dimensional coordinates, motion vector angle, and trajectory curvature dynamic parameters; Step S1120, obtain the potential margin attenuation gradient of the AGV car; Further, step S1120 includes: Step S1121, collect the charge and discharge current data of the AGV car battery in real time, perform integral operation on the current data according to the time series to obtain the change in the cumulative charge from the completion of the last charge to the current moment; Step S1122, obtain the battery aging coefficient reflecting the current degree of battery performance degradation; Step S1123, input the change in the cumulative charge and the battery aging coefficient into a pre-constructed non-linear mapping function to obtain the potential margin attenuation gradient of the current AGV car; the non-linear mapping function takes the change in the cumulative charge and the battery aging coefficient as inputs and the potential margin attenuation gradient as the output.

[0108] Step S1130: Normalize the pose topological feature vector and the potential margin attenuation gradient, and concatenate the normalized pose topological feature vector and the potential margin attenuation gradient into a multi-dimensional input vector.

[0109] Step S1200: Generate an adaptive dynamic energy consumption prediction value based on the multi-dimensional input vector and the pre-constructed multi-physical-field-coupled energy consumption trajectory residual model.

[0110] Step S2000: Construct a dynamic consensus factor that provides a priority decision basis for the AGV charging and swapping scheduling optimization, and perform cross-node verification on the constructed dynamic consensus factor to obtain a credible dynamic consensus factor; Furthermore, Step S2000 includes: Step S2100: Determine the urgency weight reflecting the urgency of AGV charging, the spatial conflict entropy value reflecting the risk of AGV spatial conflict, and the task timeliness penalty term reflecting the risk of AGV task timeout. According to the urgency weight, the spatial conflict entropy value, and the task timeliness penalty term, construct a dynamic consensus factor that provides a priority decision basis for the AGV charging and swapping scheduling optimization; Furthermore, Step S2100 includes: Step S2110: Determine the urgency weight E; Furthermore, Step S2110 includes: Step S2111: Obtain the current remaining battery power DE of the AGV; Step S2112: According to the current remaining battery power DE of the AGV and the potential margin attenuation gradient, obtain an activation exponential growth function; Step S2113: Set the initial urgency weight E1. When DE≥S1, E = E1; when DE<S1, dynamically increase the initial urgency weight E1 according to the activation exponential growth function to obtain the urgency weight E, where S1 is the battery power urgency threshold.

[0111] Step S2120: Determine the spatial conflict entropy value; Furthermore, Step S2120 includes: Step S2121: Obtain the relative distances between multiple AGVs; Step S2122: According to the relative distances between multiple AGVs, introduce a virtual repulsive force field and establish a congestion model of AGVs in the operation area of the swapping station; Step S2123: According to the established congestion model of AGVs in the operation area of the swapping station, obtain the spatial conflict entropy value reflecting the risk of AGV spatial conflict.

[0112] Step S2130: Determine the task timeliness penalty term; Furthermore, Step S2130 includes: Step S2131: Obtain the deadline of the current task being executed by the AGV and the current time. Step S2132: Introduce an inverse proportional function of the deadline of the task and the current time; the function value of the inverse proportional function increases non-linearly as the deadline of the task approaches. Step S2133: Generate a task timeliness penalty term reflecting the risk of AGV task timeout according to the inverse proportional function.

[0113] Step S2140: Generate a dynamic consensus factor by weighted fusion of the urgency weight, the spatial conflict entropy value, and the task timeliness penalty term.

[0114] Step S2200: Build a distributed collaborative verification layer based on the Byzantine fault tolerance mechanism, and perform cross-node verification on the constructed dynamic consensus factor through the distributed collaborative verification layer to obtain a trusted dynamic consensus factor.

[0115] Step S3000: Discretize the operation space of the battery swapping station to generate a spatio-temporal folding unit; based on the dynamic energy consumption prediction value, the trusted dynamic consensus factor, and the spatio-temporal folding unit, construct a spatio-temporal conflict resolution map of the battery swapping station, and generate the optimal battery swapping order and collision avoidance path of the AGV. Furthermore, Step S3000 includes: Step S3100: Discretize the operation space of the battery swapping station to generate a spatio-temporal folding unit. Furthermore, Step S3100 includes: Step S3110: Extract the four-dimensional spatio-temporal coordinates of the resource elements of the battery swapping station to form a state matrix of the operation space of the battery swapping station; the four-dimensional spatio-temporal coordinates introduce a time dimension on the basis of the three-dimensional Euclidean space; the resource elements of the battery swapping station include charging pile positions, robotic arm activity areas, and AGV parking positions. Step S3120: Use the spatio-temporal folding theory to perform compression mapping on the state matrix of the operation space of the battery swapping station to generate a spatio-temporal folding unit reflecting the spatio-temporal dependence relationship between the resource elements of the battery swapping station.

[0116] Step S3200: Based on the dynamic energy consumption prediction value, the trusted dynamic consensus factor, and the spatio-temporal folding unit, construct a spatio-temporal conflict resolution map of the battery swapping station, and generate the optimal battery swapping order and collision avoidance path of the AGV. Furthermore, Step S3200 includes: Step S3210: Map the dynamic energy consumption prediction value and the trusted dynamic consensus factor to the spatio-temporal folding unit to form a six-dimensional state matrix; where six dimensions refer to the three dimensions of space, the time dimension, the priority dimension, and the energy consumption dimension.

[0117] Step S3220: Based on the six-dimensional state matrix, construct a spatio-temporal conflict resolution map for the battery swapping station through the variational Bayesian inference method; each AGV corresponds to a node in the spatio-temporal conflict resolution map for the battery swapping station. Step S3230: Based on the spatio-temporal conflict resolution map for the battery swapping station, iteratively search for the optimal battery swapping order through the spatio-temporal replacement algorithm. Further, Step S3230 includes: Step S3231: According to the spatio-temporal conflict resolution map for the battery swapping station, construct a node coloring model. Step S3232: According to the node coloring model, use the tabu search algorithm to traverse and color the nodes of the map according to the credible dynamic consensus factor. Step S3233: Gradually eliminate the adjacent nodes of the same color in the spatio-temporal conflict resolution map for the battery swapping station through the spatio-temporal replacement algorithm. Step S3234: Through iterative optimization, output the final node coloring scheme, and the final node coloring scheme corresponds to the optimal battery swapping order of the AGV.

[0118] Step S3240: Based on the spatio-temporal conflict resolution map for the battery swapping station, generate a collision avoidance path for the AGV in the battery swapping station through the multi-granularity trajectory planning method. Further, Step S3240 includes: Step S3241: Extract the spatio-temporal occupancy intervals of each AGV from the spatio-temporal conflict resolution map for the battery swapping station. Step S3242: Combine the optimal battery swapping order of the AGV to determine the order of the predetermined battery swapping time windows for each AGV. Step S3243: Taking the time granularity as the first priority, perform sequential constraints and interval extension on the predetermined battery swapping time windows of each AGV to obtain the planning result in the time dimension. Step S3244: Taking the space granularity as the second priority, search for local paths through the artificial potential field method within each time slice to obtain the planning result in the space dimension; the time slice is obtained by dividing the entire operation space of the battery swapping station into a multi-dimensional grid map. Step S3245: Integrate the planning results in the time dimension and the space dimension to generate a collision avoidance path for the AGV in the battery swapping station.

[0119] The methods and systems of the present application can be implemented in many ways. For example, the methods and systems of the present application can be implemented through software, hardware, firmware, or any combination of software, hardware, and firmware. The above order of the steps for the method is only for illustration, and the steps of the method of the present application are not limited to the above specific described order unless otherwise specifically stated.

[0120] In addition, in the above technical solutions provided in the embodiments of the present application, the parts that are consistent with the corresponding technical solutions in the prior art in terms of implementation principles are not described in detail to avoid excessive elaboration.

[0121] As described above in the specific embodiments, the objectives, technical solutions, and beneficial effects of the present invention have been further described in detail. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A fully automatic AGV car charging and battery swapping collaborative optimization system, characterized in that, The system includes: An energy consumption prediction module: used to obtain the pose topological feature vector and potential margin attenuation gradient of the AGV, construct a multi-dimensional input vector, and generate an adaptive dynamic energy consumption prediction value based on the multi-dimensional input vector and a pre-constructed energy consumption trajectory residual model with multi-physical field coupling; A consensus mechanism design module: used to construct a dynamic consensus factor that provides a priority decision basis for the charging and swapping scheduling optimization of AGVs, and perform cross-node verification on the constructed dynamic consensus factor to obtain a credible dynamic consensus factor; An AGV scheduling optimization module: used to discretize the operation space of the swapping station to generate spatio-temporal folding units; based on the dynamic energy consumption prediction value, the credible dynamic consensus factor, and the spatio-temporal folding units, construct a spatio-temporal conflict resolution map of the swapping station, and generate the optimal swapping order and collision avoidance path of the AGV.

2. The fully automatic AGV vehicle charging and battery swapping collaborative optimization system according to claim 1, wherein, The method for obtaining the pose topological feature vector of the AGV is: obtaining the pose topological features of the AGV in real time to form a pose topological feature vector; the pose topological features at least include three-dimensional coordinates.

3. The fully automatic AGV car charging and battery swapping collaborative optimization system according to claim 2, wherein, The method for constructing the multi-dimensional input vector is: normalizing the pose topological feature vector and the potential margin attenuation gradient, and splicing the normalized pose topological feature vector and potential margin attenuation gradient into a multi-dimensional input vector.

4. A full-automatic AGV car charging and battery swapping collaborative optimization system according to claim 1, characterized in that, The consensus mechanism design module includes a dynamic consensus factor construction unit and a dynamic consensus factor verification unit; The dynamic consensus factor construction unit is used to determine the urgency weight reflecting the charging urgency of the AGV, the spatial conflict entropy value reflecting the spatial conflict risk of the AGV, and the task timeliness penalty term reflecting the task timeout risk of the AGV, and construct a dynamic consensus factor that provides a priority decision basis for the charging and swapping scheduling optimization of AGVs according to the urgency weight, the spatial conflict entropy value, and the task timeliness penalty term.

5. The fully automatic AGV car charging and battery swapping collaborative optimization system according to claim 4, characterized in that, The dynamic consensus factor verification unit is used to build a distributed collaborative verification layer based on the Byzantine fault tolerance mechanism, and perform cross-node verification on the constructed dynamic consensus factor through the distributed collaborative verification layer to obtain a credible dynamic consensus factor.

6. The fully automatic AGV car charging and battery swapping collaborative optimization system according to claim 4, characterized in that, The method for determining the urgency weight includes: Obtaining the current remaining battery power DE of the AGV; obtaining an activation exponential growth function according to the current remaining battery power DE of the AGV and the potential margin attenuation gradient; setting an initial urgency weight E1, when DE≥S1, the urgency weight E = E1; when DE<S1, dynamically increase the initial urgency weight E1 according to the activation exponential growth function to obtain the urgency weight E, where S1 is the battery power urgency threshold.

7. A fully automatic AGV car charging and battery swapping collaborative optimization system according to claim 4, characterized in that, The method for determining the spatial conflict entropy value includes: Based on the three-dimensional coordinates of the AGV in the pose topological feature vector, obtaining the relative distances between multiple AGVs; According to the relative distances between multiple AGVs, introducing a virtual repulsive force field to establish a congestion model of the AGV in the operation area of the swapping station; According to the established congestion model of the AGV in the operation area of the swapping station, obtaining the spatial conflict entropy value reflecting the spatial conflict risk of the AGV.

8. The fully automatic AGV car charging and battery swapping collaborative optimization system according to claim 4, wherein The method for determining the task timeliness penalty term includes: Obtain the deadline of the current task being executed by the AGV and the current time; introduce an inverse proportional function of the deadline of the task and the current time; generate a task timeliness penalty term reflecting the AGV task timeout risk according to the inverse proportional function.

9. The fully automatic AGV car charging and battery swapping collaborative optimization system according to claim 1, wherein The method for discretizing the operation space of the swapping station includes: Extract the four-dimensional spatio-temporal coordinates of the resource elements of the swapping station to form a state matrix of the operation space of the swapping station; perform a compression mapping on the state matrix of the operation space of the swapping station to generate a spatio-temporal folding unit reflecting the spatio-temporal dependence relationship between the resource elements of the swapping station.

10. A fully automatic AGV car charging and battery swapping collaborative optimization system according to claim 9, characterized in that, The method for constructing a spatio-temporal conflict resolution map of the swapping station includes: Map the dynamic energy consumption prediction value and the credible dynamic consensus factor to the spatio-temporal folding unit to form a six-dimensional state matrix; construct a spatio-temporal conflict resolution map of the swapping station based on the six-dimensional state matrix; each AGV corresponds to a node in the spatio-temporal conflict resolution map of the swapping station.

11. The fully automatic AGV car charging and battery swapping collaborative optimization system according to claim 10, characterized in that, The method for generating the optimal swapping sequence of the AGV includes: Construct a node coloring model according to the spatio-temporal conflict resolution map of the swapping station; traverse and color the nodes of the map according to the credible dynamic consensus factor according to the node coloring model; Gradually eliminate the adjacent nodes of the same color in the spatio-temporal conflict resolution map of the swapping station through the spatio-temporal permutation algorithm; output the final node coloring scheme through iterative optimization, and the final node coloring scheme corresponds to the optimal swapping sequence of the AGV.

Images