An intelligent loading method and system for air freight

Abstract

This invention provides an intelligent loading method and system for air cargo, including collecting information on cargo size, weight, center of gravity, and destination port; generating an initial plan using a layered filling strategy; iterative optimization using a simulated annealing algorithm; and selecting destruction and repair operators based on the current temperature. At high temperatures, operators with better historical exploration performance are prioritized, while at low temperatures, operators with better historical optimization performance are emphasized. The destruction operator analyzes load balance and stability based on the cargo support relationship diagram, removing cargo with insufficient stability. The repair operator evaluates each candidate position for the cargo, comprehensively evaluating multiple dimensions such as overall center of gravity impact, contact surface stability, cargo concentration at the same port, and local stability improvement, selecting the optimal position for insertion as a candidate load, and outputting the loading plan when preset termination conditions are met.

Description

Technical Field

[0001] This application belongs to the field of intelligent loading, and in particular relates to an intelligent loading method and system for air cargo. Background Technology

[0002] The loading process in air freight requires the rational placement of cargo of varying sizes, weights, and destinations into limited container space, while strictly adhering to aircraft center of gravity balance, container weight limits, and handling requirements for different types of cargo. Currently, air freight loading still largely relies on manual planning based on the experience of on-site operators. This method is not only inefficient and time-consuming, but also highly subjective, resulting in inconsistent quality and difficulty in ensuring optimal space utilization and loading stability. The limitations of manual loading become increasingly apparent, especially when dealing with large quantities of irregular cargo and complex constraints. Manual loading methods can no longer meet the high standards of precision and safety required by modern air logistics. Existing technologies mostly employ heuristic algorithms, such as greedy algorithms or genetic algorithms, to solve combinatorial optimization problems involving cargo packing. These are prone to getting trapped in local optima, leading to low container space utilization and difficulty in maximizing loading efficiency. Furthermore, current technologies only check support surfaces when assessing loading stability, failing to reliably evaluate the load transfer and distribution balance between cargo, posing safety risks. Existing optimization algorithms lack mechanisms to adjust search behavior based on the optimization process, making it difficult to achieve a balance between global exploration and local optimization, and their ability to perform collaborative optimization for multiple objectives is also relatively limited. Summary of the Invention

[0003] Existing technologies struggle to optimize space utilization and loading stability, often falling into local optima and failing to maximize loading efficiency.

[0004] In the first aspect, the present invention proposes an intelligent loading method for air cargo, comprising the following steps: An initial loading plan is generated by adopting a layered filling strategy, prioritizing the stacking of heavy cargo at the bottom of the container unit, and then filling the upper layer and remaining space with light cargo; The initial loading scheme is iteratively optimized, including: selecting a destruction operator and a repair operator based on the current temperature using a simulated annealing algorithm, wherein operators with better historical exploration performance are prioritized during high-temperature stages, and operators with better historical optimization performance are prioritized during low-temperature stages; executing the selected destruction operator to remove some cargo, wherein the destruction operator constructs a support relationship graph between cargoes, evaluates the load balance and support stability of each cargo, and identifies and removes cargoes with stability evaluation values ​​below a preset threshold; executing the selected repair operator to re-insert the removed cargo, generating a neighborhood loading scheme, wherein the repair operator evaluates all candidate loading positions for the cargo to be inserted and selects the optimal position for insertion based on a comprehensive evaluation value; and deciding whether to accept the neighborhood loading scheme based on simulated annealing criteria, while simultaneously updating the historical performance scores of the operators. Output the load allocation scheme when the preset termination conditions are met.

[0005] In another aspect, the present invention also proposes an intelligent cargo loading system, comprising the following modules: The generation module is used to generate an initial loading plan using a layered filling strategy, prioritizing the stacking of heavy cargo at the bottom of the container unit, and then filling the upper layer and remaining space with light cargo. An update module is used to iteratively optimize the initial loading scheme. This iterative optimization includes: selecting a destruction operator and a repair operator based on the current temperature using a simulated annealing algorithm, where operators with better historical exploration performance are prioritized during high-temperature phases, and operators with better historical optimization performance are prioritized during low-temperature phases; executing the selected destruction operator to remove some cargo, where the destruction operator constructs a support relationship graph between cargoes, evaluates the load balance and support stability of each cargo, and identifies and removes cargoes with stability evaluation values ​​below a preset threshold; executing the selected repair operator to re-insert the removed cargo, generating a neighborhood loading scheme, where the repair operator evaluates all candidate loading positions for the cargo to be inserted and selects the optimal position for insertion based on a comprehensive evaluation value; and deciding whether to accept the neighborhood loading scheme based on simulated annealing criteria, while simultaneously updating the historical performance scores of the operators. The output module is used to output the load allocation scheme when the preset termination conditions are met.

[0006] This invention employs a layered filling strategy, prioritizing heavier loads over lighter ones, to quickly generate a loading scheme with a low center of gravity and good initial stability. In subsequent iterative optimizations, by constructing a support relationship diagram between cargoes, unstable local areas can be identified and prioritized for adjustment, thus improving the scheme's stability. During the cargo re-insertion phase, the impact of loading positions on the overall center of gravity of the container unit, contact stability with adjacent cargoes, the degree of aggregation of cargoes at the same destination port, and the improvement of local structures were evaluated, ensuring that each adjustment was made towards multi-objective optimization. The resulting loading scheme not only improves space utilization but also ensures structural stability and safety during cargo transportation. Attached Figure Description

[0007] Figure 1 A flowchart of the first embodiment; Figure 2 A schematic diagram illustrating the strategy for selecting operators; Figure 3 A schematic diagram of the operator's historical performance score update mechanism. Detailed Implementation

[0008] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0009] In the first embodiment, the present invention proposes an intelligent loading method for air cargo, see [link to relevant documentation]. Figure 1 This includes the following steps: S1. An initial loading plan is generated by adopting a layered filling strategy, prioritizing the stacking of heavy cargo at the bottom of the container unit, and then filling the upper layer and remaining space with light cargo. The density threshold is set, for example, to 800 kg per cubic meter. All goods to be loaded are divided into heavy and light cargo. All heavy cargo is sorted by base area from largest to smallest and placed sequentially at the bottom of the virtual three-dimensional space of the container unit, using a left-to-back-bottom corner priority algorithm. After all heavy cargo is placed or the bottom space is insufficient, all light cargo is sorted by volume from largest to smallest and filled from bottom to top and from inside to outside into the space above the heavy cargo and the gaps between the heavy cargo. In one embodiment, the container unit is an aircraft, preferably a small bulk cargo aircraft. In another embodiment, the aircraft loading area is divided into multiple container units.

[0010] In an optional embodiment, the three-dimensional dimensions, weight, and center of gravity of the cargo set to be loaded are obtained in advance; The method of generating an initial loading plan using a layered filling strategy, which prioritizes stacking heavy cargo at the bottom of the container unit and then fills the upper layer and remaining space with light cargo, includes: The cargo to be loaded is divided into heavy cargo and light cargo according to the preset density threshold. The heavy goods are arranged in descending order of weight and stacked from the bottom of the container unit towards the center along a predetermined path. The lighter goods are arranged in descending order of volume and filled into the remaining space above the heavier goods.

[0011] Specifically, the system automatically reads freight order data from the Warehouse Management System (WMS) or Enterprise Resource Planning (ERP) database via an interface program to obtain the length, width, height, and weight of goods in batches. In one embodiment, it also obtains the destination port code and special handling requirements such as whether the goods are fragile or invertible. For goods with irregular shapes, a 3D laser scanning device is used to automatically scan and model them, obtaining the external contour dimensions and volume of the goods. The center of gravity information is initially set as the geometric center based on the assumption of material homogeneity. Assuming a preset density threshold of 800 kg / m³, existing goods A have dimensions of 2 meters long, 1 meter wide, and 1 meter high, a volume of 2 m³, and a weight of 2000 kg. The calculated density of goods A is 1000 kg / m³, which is greater than the threshold, so it is classified as heavy goods. Goods B have dimensions of 1 meter long, 1 meter wide, and 1 meter high, a volume of 1 m³, and a weight of 500 kg. The calculated density of goods B is 500 kg / m³, which is less than the threshold, so it is classified as light goods. In this way, all cargo lists to be loaded are processed into two sets: a heavy cargo list and a light cargo list.

[0012] Assuming the heavy goods list contains three items, sorted in descending order of weight as follows: Weighing 3000kg, Weighing 2500kg, Weighing 2200 kg. The container is a standard shipping container, and the planned loading path is to start from the front two sides of the bottom of the container, move backward in a zigzag pattern, and gradually fill towards the center. Place it on the front left side of the bottom of the container. Place it on the front right side, Placed next to The process is repeated until all heavy cargo is placed at the bottom of the container, creating a stable foundation with a low center of gravity. The Z-shaped rearward and gradually inward filling method prioritizes placing heavy cargo in the front left and right corners of the container unit, for example, first the left front, then the right front, following an alternating "left-right-left-right" Z-shaped trajectory to ensure weight balance on both sides. Once the designated areas on both sides are filled, subsequent heavy cargo is added to the gaps between the cargo on the left and right sides, moving towards the center. This front-to-back, outside-to-inside filling method avoids unilateral load distribution in the initial stages of cargo stacking and ensures that the center of gravity of the final heavy cargo layer converges towards the geometric center of the container unit, thus maximizing the structural stability and flight safety of air freight.

[0013] After the heavy goods are stacked, irregular gaps will form at the top. At this point, the list of light goods is processed, assuming it includes... The volume is 8m³. The volume is 6m³, arranged in descending order of volume. Find a container that can hold... The maximum available space, for example, a sufficiently large space above the middle of the heavy cargo layer, will allow for... Insert. Then for The next largest available space is then filled. This process continues until all light cargo is loaded or the remaining space cannot accommodate any more light cargo, thus maximizing the use of the container's loading capacity while ensuring structural stability.

[0014] S2, iteratively optimize the initial loading scheme. The iterative optimization includes: selecting a destruction operator and a repair operator based on the current temperature using a simulated annealing algorithm, wherein in the high-temperature stage, operators with better historical exploration performance are prioritized, and in the low-temperature stage, operators with better historical optimization performance are prioritized; executing the selected destruction operator to remove some cargo, wherein the destruction operator constructs a support relationship graph between cargoes, evaluates the load balance and support stability of each cargo, and identifies and removes cargoes with stability evaluation values ​​below a preset threshold; executing the selected repair operator to re-insert the removed cargo, generating a neighborhood loading scheme, wherein the repair operator evaluates all candidate loading positions for the cargo to be inserted, and selects the optimal position for insertion based on a comprehensive evaluation value of at least four dimensions: the impact on the overall center of gravity of the container unit, the stability of the contact surface with adjacent cargoes, the degree of aggregation of cargoes at the same destination port, and the degree of improvement on the structural stability of the local area where the cargo is located; deciding whether to accept the neighborhood loading scheme according to the simulated annealing criterion, and updating the historical performance score of the operator simultaneously.

[0015] A pool of operators with various strategies is maintained, further divided into a destruction operator pool and a repair operator pool. For example, the destruction pool might contain operators such as removing the most unstable cargo, random removal, and removing cargo from the same destination port cluster. At algorithm startup, the historical exploration and optimization performance scores of all these operators are initialized to the same initial value, for example, 1. Without any historical data, each operator has an equal chance of being selected initially. As iterations progress, during the high-temperature phase of simulated annealing, operators that lead the search to novel solution regions (i.e., those with high exploration performance scores) are prioritized for breadth-first global search. However, in the low-temperature phase, operators that continuously improve the optimal solution (i.e., those with high optimization performance scores) are prioritized, achieving a smooth transition from global exploration to local optimization. Figure 2 As shown. Specifically, at the start of the iteration, the initial temperature is set to 100, and the cooling coefficient is 0.995; two scores are maintained for each operator: a historical exploration score and a historical optimization score. At each iteration, a weighting coefficient W is calculated based on the current temperature T. , The initial temperature is used as the coefficient, which decreases from 1 to 0 as the temperature decreases. The operator selection score is given by a formula: for example, equal to W multiplied by the historical exploration score + (1-W) multiplied by the historical optimization score. Based on the proportion of the comprehensive selection scores of all operators, a roulette wheel selection method is used to randomly select a destruction operator and a repair operator. A temperature cutoff point is set, which is determined according to the proportion of the initial temperature. For example, if the initial temperature is 100 and the termination temperature is 0.1, the temperature cutoff point is 50. Figure 2 As shown. When the current temperature is above 50°C, it is the high-temperature stage, where the goal is to extensively explore the solution space and avoid getting trapped in local optima. When the temperature is below 50°C, it is the low-temperature stage, where the goal is to optimize the current better solution, i.e., to perform a local search. For example... Figure 2 In the above, when the temperature is 40°C, the weight of historical exploration is 0.4 and the weight of historical optimization is 0.6.

[0016] Traverse all goods in the loading scheme and construct a directed graph. If the top of goods A supports the bottom of goods B, then establish an edge from A to B. For each goods A, calculate the horizontal projection point of the total center of gravity of all goods B, C, and D above it. If the distance of this projection point from the center of the top surface of goods A exceeds one-third of the width or length of the top surface, the load balance is counted as low. At the same time, for each goods B, calculate the ratio of the contact area between the bottom surface and the goods below to the total bottom area. If this ratio is less than 70%, the support stability is counted as low. Combine the above two indicators to calculate a stability evaluation value between 0 and 1. For example, for any goods node, first calculate the load balance score, and use a linear function to reverse normalize the projection offset of the upper bearing center of gravity. That is, when the offset is 0, it is scored as 1 point, and when the offset reaches 1 / 3 of the top surface size threshold or more, it is scored as 0 points. At the same time, calculate the support area score. According to a preset weight, for example, each accounts for 50%, the above two scores are weighted and summed to obtain a comprehensive stability evaluation value between 0 and 1. Add all cargo with an assessment value below 0.4 to the removal list and remove it from the current loading plan.

[0017] For the first item to be removed from the list, iterate through all available space points within the container unit to generate a series of candidate loading positions. For each candidate position, calculate four scores: calculate the new total center of gravity of the entire container unit after inserting the item; the closer the new total center of gravity is to the center of the bottom surface of the container unit, the higher the score; calculate the total contact area between the item and the items below and around it; the larger the area, the higher the score; count the number of items within one meter of the container unit whose destination port is the same as the item's; the more items, the higher the score; evaluate whether the item provides better support for the unstable items above it after placement; the greater the improvement, the higher the score; normalize the four scores, calculate the weighted sum, and select the candidate position with the highest total score to insert the item. Repeat this process for the next item in the list.

[0018] Calculate the difference between the objective function value of the newly generated neighborhood loading scheme and the current loading scheme; if the difference is less than zero, that is, the new scheme is better, then the new scheme is unconditionally accepted as the current scheme; if the difference is greater than or equal to zero, that is, the new scheme is worse, then calculate an acceptance probability, which is the negative difference of the natural constant e divided by the power of the current temperature, and generate a random number between 0 and 1. If the random number is less than the calculated acceptance probability, then the worse scheme is still accepted, otherwise it is rejected; after accepting the scheme, update the performance score of the operators used this time (including the destruction operator and the repair operator) according to the result.

[0019] In an optional embodiment, the current temperature selection of the destruction operator and the repair operator based on the simulated annealing algorithm, wherein the operator with better historical exploration performance is preferentially selected in the high-temperature stage, and the operator with better historical optimization performance is preferentially selected in the low-temperature stage, includes: Compare the current temperature with the preset temperature boundary point to distinguish between high and low temperature stages; For any operator o in the operator set O, the historical exploration performance score is: Historical optimization performance score: ; When the temperature is high, the probability P(o) of operator o being selected is determined based on historical exploration performance scores: ; When the system is in a low-temperature phase, the probability P(o) of selecting operator o is determined based on historical optimization performance scores: ; Where j is the index of the operator in operator set O.

[0020] During the high-temperature phase, operators with stronger historical exploration performance are prioritized. Assume operator set O contains three operators. , , Their historical exploration performance score The scores are 20, 50, and 30 respectively. At this point, the total score is 100. Operator The probability of being selected is Operator The probability is Operator The probability is Operators that are more active in historical exploration There is a higher probability of selection, resulting in more diverse neighborhood loading schemes.

[0021] When entering a low-temperature phase, for example, when the current temperature drops to 150°C, the selection strategy switches to based on historical optimization performance scores. Assuming the historical optimization performance scores of the three operators mentioned above... The scores were 60, 15, and 5 respectively, with a total score of 80. At this point, the operator... The probability of selection becomes ,and The probability drops to , The probability is This allows operators that previously generated better solutions to be used later in the algorithm. Frequent use allows for fine-tuning of high-quality solutions.

[0022] In an optional embodiment, the structural stability-based failure operator removes some cargo. The failure operator constructs a support relationship graph between cargoes, evaluates the load balance and support stability of each cargo, and identifies and removes cargoes with stability evaluation values ​​below a preset threshold, including: Construct a directed support relationship graph between goods; Calculate the load balance of each item i. With support stability ; The load balance and support stability are weighted and summed to obtain the stability assessment value for each cargo i. The calculation formula is: ; in, and These are the preset weights for load balancing and support stability, respectively. Remove goods whose stability assessment value is lower than the preset threshold.

[0023] The destruction operator abstracts the current loading scheme as a directed support graph model. In this graph, each cargo is a node. If cargo A directly supports cargo B, then there is a directed edge from node A to node B. For example, if cargo C is placed on top of cargo A and cargo B, then there are two edges in the graph: one from A to C and one from B to C. This graph structure represents the force transmission relationships within the entire cargo stack.

[0024] Based on this graph model, the load balancing degree is calculated for each non-bottom-level item i. With support stability Two quantitative metrics: Load balancing. This can be obtained by calculating the distance between the horizontal projection point of the combined center of gravity of all supporting goods above it and the geometric center of the top surface of goods i, and then normalizing the result. =max(0, 1 - offset distance / (0.5 × minimum side length of cargo i's top surface)). Support stability It is the ratio of the actual contact area between the bottom surface of cargo i and the supporting surface below to its total bottom area.

[0025] The overall stability assessment value for each cargo is calculated by weighted summation. Load balancing weights are then set. The value is 0.4, which supports the stability weight. It is 0.6. For example, the assessed value of goods C is... The removal threshold is set to 0.4, because... Below this threshold, cargo C is identified as unstable cargo. The operator selects one or more cargoes, such as cargo C, with the lowest stability assessment value, removes them from the current loading scheme, and adds them to a list to be reinserted, thereby disrupting the current structure.

[0026] In one embodiment, constructing a support relationship diagram among goods and evaluating the load balance and support stability of each goods includes: Construct a directed acyclic graph with the container unit floor as the root node and each cargo as a child node. When cargo A is located above cargo B and the two are in physical contact, construct a directed edge from B to A. The load balance is defined as the ratio of the effective support area obtained by the bottom surface of the cargo to the total area of ​​the bottom surface of the cargo. The support stability is defined as the normalized value of the minimum distance from the projection point of the cargo's center of gravity onto the horizontal plane to the boundary of the support polygon below it, wherein the support polygon is formed by combining the top surface contours of the lower layer of cargo supporting the cargo.

[0027] Specifically, for constructing the directed support relationship graph, the algorithm traverses each cargo i in the loading scheme and identifies all supports immediately below it, including other cargoes or container unit floor plates. If the top height of cargo j is equal to the bottom height of cargo i, and their vertical projections on the horizontal plane (XY plane) have a non-zero overlapping area, then a physical support relationship is determined to exist, and a directed edge from node j to node i is created in the graph.

[0028] For load balancing The calculation first involves calculating the total geometric overlap area between the bottom surface of cargo i and all its parent nodes, i.e., the top surface of cargo j that directly supports it, which is denoted as the effective support area. Obtain the total area of ​​the bottom surface of cargo i itself. Load balancing is the ratio of the two, that is... The closer the value is to 1, the less the bottom of the cargo is suspended and the more evenly the force is distributed.

[0029] For support stability The calculation first involves performing a geometric union operation on the top surface projections of all supporting cargo i, forming a supporting polygon region. Obtain the physical center of gravity coordinates of cargo i, and project them onto the plane containing the base to obtain the center of gravity projection point P. If point P falls outside the supporting polygon area, the cargo is in an overturned state. If the value is 0, then calculate the shortest Euclidean distance from P to the boundary of the supporting polygon region. Half the length of the shorter side of the bottom surface of cargo i is selected as the normalization reference value. ,but .

[0030] In an optional embodiment, the repair operator evaluates all candidate loading positions for the cargo to be inserted and selects the optimal position for insertion based on a comprehensive evaluation of at least four dimensions: impact on the overall center of gravity of the container unit, stability of the contact surface with adjacent cargo, degree of aggregation of cargo at the same destination port, and degree of improvement on the structural stability of the local area where the cargo is located. This includes: For each candidate loading location, scores are evaluated across at least four dimensions, including: impact on the overall center of gravity of the container unit. Stability of the contact surface with adjacent goods The degree of aggregation of goods destined for the same port within the planned neighboring area. And the degree of improvement in the structural stability of the local area where the goods are located. ; The comprehensive evaluation value is obtained by weighted summation of the scores of the at least four dimensions. And select the position with the highest comprehensive evaluation value for insertion.

[0031] The core of the repair operator is a multi-objective decision model used to find the optimal placement location for a piece of cargo to be inserted. Suppose cargo G is removed by the destruction operator and now needs to be re-inserted. All physically identifiable candidate spaces within the container that can accommodate cargo G are searched, denoted as position 1 and position 2. For each candidate position, a score is assigned across four dimensions, with all scores normalized to between 0 and 1.

[0032] The evaluation is conducted using positions 1 and 2 as examples. The new overall center of gravity coordinates of the entire container unit are calculated after cargo G is placed at position pos. The degree of deviation of this new center of gravity from the ideal center of gravity of the container unit is evaluated, typically the deviation from a point on the vertical line of the bottom center. A smaller deviation indicates a higher center of gravity. The higher the (pos) score, the better. One calculation method is... (pos)=1 / (1+α (Horizontal offset distance), where α is an adjustment coefficient, for example, the value range is [0.02, 0.1]. For example, position 1 is at the bottom of the container, and putting cargo G in will significantly lower the overall center of gravity, resulting in a high score, such as 0.9; position 2 is on the upper layer of the cargo stack, which will raise the center of gravity, resulting in a low score, such as 0.2.

[0033] Calculate the total contact area between the bottom and sides of cargo G at position pos and the existing cargo and container walls. A larger contact area generally indicates a more stable placement. The higher the (pos) score, the more normalized the score can be to the ratio of the contact area to the maximum possible contact area of ​​the goods G. For example, the bottom surface of position 1 is a flat container floor, with stable contact and a high score, such as 1.0; the bottom surface of position 2 is two small packaging boxes, with a small contact area and a low score, such as 0.4.

[0034] Define a predetermined neighborhood centered at location pos, such as a cube with sides of 2 meters. Count the quantity of other goods within this neighborhood that have the same destination port as goods G. The larger the quantity, the better. The higher the (pos) score, the more normalized the score can be to the ratio of the actual quantity to the total number of goods in the neighborhood or a preset maximum value. For example, if the destination port is Shanghai, and the area around position 1 is full of goods going to Singapore, the clustering is poor and the score is low, such as 0.1; while position 2 has exactly three goods going to Shanghai, the clustering is good and the score is high, such as 0.8.

[0035] The evaluation assesses the improvement in stacking stability of goods within a local area centered on position pos after placing goods G. This can be calculated by evaluating the average increase in the overall stability assessment value of all relevant goods within this local area before and after placement, including goods G itself and the goods directly supported and supported by it. A greater increase indicates better stability. The higher the (pos) score, the better. For example, placing goods G at position 1 has no impact on the stability of surrounding goods, resulting in a neutral score of 0.5; placing goods G at position 2 can effectively stabilize a previously wobbly goods next to it, improving local stability, resulting in a higher score, such as 0.9.

[0036] After scoring all dimensions, a comprehensive evaluation value is calculated based on preset weights. The weights for the impact on the overall center of gravity of the container unit are assumed. Weight of the stability of the contact surface with adjacent goods The degree of aggregation of goods destined for the same port within the predetermined neighborhood area. The weight of the degree of improvement in the structural stability of the local area where the goods are located The sum of the weights is 1. The overall evaluation value for position 1 is... The overall evaluation value for position 2 is... Since position 1 has a higher overall evaluation value, the repair operator decides to insert cargo G into position 1.

[0037] Optionally, the dimension also includes the fit of the aircraft's overall center of gravity. Specifically, the candidate position coordinates of the cargo to be inserted in the container are mapped to the absolute lever arm value relative to the aircraft fuselage reference plane. The additional torque generated by the insertion operation is calculated in combination with the cargo weight. The current estimated total weight and total torque of the entire aircraft are updated to solve for the temporary center of gravity position of the aircraft after the cargo is inserted. The absolute deviation between the temporary center of gravity and the airline's preset ideal center of gravity target value is calculated, and the deviation is mapped to a normalized score through an inverse proportional function or a Gaussian function. This ensures that the loading position where the aircraft's center of gravity converges to the ideal target receives a higher fit score, while the position where the center of gravity exceeds the safety envelope boundary receives a score of zero or negative infinity.

[0038] In an optional embodiment, the step of determining whether to accept the neighborhood loading scheme based on simulated annealing criteria includes: Calculate the difference in objective function values ​​between the newly generated neighborhood loading scheme and the current loading scheme. The objective function is a comprehensive evaluation function, where a smaller value indicates a better solution. This function integrates multiple key optimization objectives into a single scalar value through a weighted summation, for example, calculated using the following formula: .

[0039] in, , , , The preset weighting coefficients for each sub-item; the loading rate is the ratio of the total volume of loaded cargo to the volume of the container unit; the center of gravity offset penalty is a function of the distance between the horizontal projection of the total center of gravity of the container unit and the center of the bottom surface; the average stability score is the stability assessment value of all cargo. The mean value; the destination port dispersion penalty is used to quantify the dispersion of goods at the same destination port. Specifically, all goods in the current loading scheme are grouped according to the destination port code, and the geometric center coordinates of each destination port group in the container unit space, i.e., the centroid of the cluster, are calculated. All goods are traversed, and the Euclidean distance between the center point of each goods and the centroid of its group is calculated. The distance values ​​are then summed. Usually, a normalization factor, such as the total number of goods multiplied by the maximum diagonal length of the container unit, is used to process this sum to obtain the penalty value. The larger the value, the more dispersed the distribution of goods to the same destination, and the smaller the value, the more clustered the stacking.

[0040] If the difference is less than zero, the new scheme has a smaller overall objective function value, indicating better overall performance, and the new scheme is unconditionally accepted as the current scheme. If the difference is not less than zero, meaning the new scheme has worse or equal overall performance, an acceptance probability is calculated based on the simulated annealing criterion. Where T is the current temperature and exp is the natural exponent. Then, a uniformly distributed random number is generated within the interval [0,1). If the random number is less than... If the worse option is not found, the worse option will still be accepted; otherwise, the original option will be rejected and retained. The neighborhood loading scheme refers to a new candidate loading configuration that changes only locally, based on the current loading scheme during the iteration process of the optimization algorithm. Specifically, this involves using a destruction operator to remove some specific goods and then using a repair operator to reinsert these goods into new positions according to the optimization strategy.

[0041] In an optional embodiment, the step of determining whether to accept the neighborhood loading scheme based on simulated annealing criteria includes: A global objective function is constructed, which is a weighted sum of three indicators: the space utilization rate of the container unit, the distance deviation between the overall center of gravity and the geometric center of the loading scheme, and the average stability assessment value of all cargoes. Calculate the difference between the global objective function values ​​of the neighboring loading scheme and the current loading scheme; If the difference indicates that the neighboring loading scheme is better than the current loading scheme, then the neighboring loading scheme is accepted; if the difference indicates that the neighboring loading scheme is worse than the current loading scheme, then the acceptance probability is calculated based on the ratio of the difference to the current temperature, and a random decision is made on whether to accept based on the acceptance probability.

[0042] Specifically, the global objective function is defined as a comprehensive score, which aims to maximize the loading efficiency. It is composed of three normalized and weighted sums: the first part is the space utilization rate, which is the proportion of the total volume of loaded cargo to the effective volume of the container unit; the second part is the center of gravity deviation score, which is negatively correlated with the Euclidean distance from the overall physical center of gravity of the loading scheme to the geometric center of the container unit floor, i.e., the closer the distance, the higher the score; the third part is the average stability, which is the arithmetic mean of the stability assessment values ​​of all loaded cargo in the current scheme.

[0043] During the iteration process, the comprehensive score of the newly generated neighborhood loading scheme and the comprehensive score of the current loading scheme are calculated, and the difference between the two is determined. If the difference is positive, it indicates that the neighborhood scheme is better than the current scheme, and the neighborhood loading scheme is unconditionally accepted as the new current solution. If the difference is negative, it indicates that the quality of the neighborhood scheme has decreased. In this case, the scheme is not directly discarded, but an acceptance probability is calculated. The probability is determined by an exponential function with the natural constant as the base, and the exponent is the difference divided by the temperature parameter of the current simulated annealing algorithm. The calculated acceptance probability is higher if and only if the decrease in loading quality is smaller, or the current temperature parameter is higher. A random number between zero and one is generated. If the random number is less than the previously calculated acceptance probability, the worse neighborhood scheme is accepted to escape local optima; otherwise, the neighborhood scheme is rejected, and the current loading scheme is retained.

[0044] In an optional embodiment, the historical performance score of the synchronization update operator includes: Let the currently executing destruction operator be... Repair operator ; After iteration, if the generated neighborhood loading scheme is better than the current best scheme, a historical optimization performance score is added to the currently executed operator; if the generated neighborhood loading scheme is a new scheme, a historical exploration performance score is added to the currently executed operator. At the end of each iteration, for all operators o in the operator set O, both scores are multiplied by a preset decay coefficient, which is less than 1 and greater than 0.

[0045] The scoring update mechanism is a credit allocation model based on iterative feedback, with preset reward scores and decay coefficients. For example, the first score... Set to 15, representing the reward for finding a new optimal solution; the second score. Set to 5, representing the reward for discovering an unseen solution; decay coefficient. Set to 0.98 to gradually forget historical performance. Assume that in the current iteration, the algorithm selects and executes the destruction operator. and repair operator .

[0046] After this iteration, the generated neighborhood loading schemes are evaluated, and three scenarios may occur. The objective function value of the new scheme is superior to all schemes discovered so far, becoming the new globally optimal scheme. In this case, the operator combination exhibits excellent optimization capabilities, thus rewarding both its optimization performance score and exploration performance score. and Each will be increased by 15 points. and Each of these solutions will also be increased by 5 points, because the new optimal solution must also be a new solution.

[0047] While the new solution is not superior to the globally optimal solution, it is superior to the current solution and represents a novel solution not previously explored by the algorithm. This indicates that the operator combination possesses good exploratory capabilities. Therefore, only these solutions are given additional exploratory performance scores. and Each will be increased by 5 points, such as Figure 3 As shown, the performance score is optimized. It remains unchanged.

[0048] The new solution is worse than the current solution and is rejected by the acceptance criterion of simulated annealing, or it is an old solution that has already been visited. In this case, the operator combination makes no reliable contribution, and neither of its two scores increases. Regardless of which of the above situations occurs, in the step of this iteration, the two scores of all operators in the operator set O, whether or not they are used, will be multiplied by a decay factor of 0.98.

[0049] S3 outputs the load allocation scheme when the preset termination conditions are met.

[0050] In an optional embodiment, a loading scheme is output when a preset termination condition is met, including: The total number of iterations has reached the set maximum number of iterations; Alternatively, the simulated annealing algorithm can reduce the current temperature to a preset termination temperature; Alternatively, the objective function value of the optimal loading scheme does not improve within a preset number of consecutive iterations.

[0051] A strategy of multiple parallel termination conditions is adopted. The entire optimization process stops and the currently found optimal load configuration is output as soon as any one of the conditions is met. This approach ensures that the algorithm can provide results within a finite time and terminate promptly when the search stalls, avoiding waste of computational resources.

[0052] For example, a fixed upper limit of 20,000 iterations can be set. The algorithm internally uses an iteration counter, starting from 1 and incrementing by 1 after each iteration. When the counter reaches 20,000, the algorithm will terminate regardless of the search results. This provides a definite upper bound on the overall running time of the algorithm, facilitating planning and deployment in practical applications.

[0053] The core parameter of the simulated annealing algorithm is the current temperature T, which gradually decreases with each iteration. For example, the initial temperature is set to 5000, and after each iteration, the temperature is multiplied by a cooling factor of 0.999, with a termination temperature threshold set to 0.05. At the beginning of each iteration, the algorithm checks if the current temperature T is below 0.05. Once it falls below this threshold, the system is considered sufficiently cooled, the solution is stable, and the reward for continuing the search is extremely low, so the algorithm terminates.

[0054] The algorithm continuously tracks the global optimum and the objective function value. Simultaneously, it sets a counter for the number of iterations without improvement and an upper limit threshold, such as 5000 iterations. If the algorithm finds a new, better solution in a given iteration, this counter is reset to zero. If the global optimum remains unchanged for 5000 consecutive iterations, the algorithm determines that the search has entered a stationary region or a local optimum, making further iterations unlikely to yield a breakthrough. Therefore, it triggers the termination condition and ends the computation.

[0055] In a second embodiment, the present invention also provides an intelligent cargo loading system, comprising the following modules: The generation module is used to generate an initial loading plan using a layered filling strategy, prioritizing the stacking of heavy cargo at the bottom of the container unit, and then filling the upper layer and remaining space with light cargo. An update module is used to iteratively optimize the initial loading scheme. This iterative optimization includes: selecting a destruction operator and a repair operator based on the current temperature using a simulated annealing algorithm, where operators with better historical exploration performance are prioritized during high-temperature phases, and operators with better historical optimization performance are prioritized during low-temperature phases; executing the selected destruction operator to remove some cargo, where the destruction operator constructs a support relationship graph between cargoes, evaluates the load balance and support stability of each cargo, and identifies and removes cargoes with stability evaluation values ​​below a preset threshold; executing the selected repair operator to re-insert the removed cargo, generating a neighborhood loading scheme, where the repair operator evaluates all candidate loading positions for the cargo to be inserted and selects the optimal position for insertion based on a comprehensive evaluation value; and deciding whether to accept the neighborhood loading scheme based on simulated annealing criteria, while simultaneously updating the historical performance scores of the operators. The output module is used to output the load allocation scheme when the preset termination conditions are met.

[0056] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referred to each other.

[0057] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A smart loading method for air cargo, characterized in that, Includes the following steps: An initial loading plan is generated by adopting a layered filling strategy, prioritizing the stacking of heavy cargo at the bottom of the container unit, and then filling the upper layer and remaining space with light cargo; The initial loading scheme is iteratively optimized, including: selecting a destruction operator and a repair operator based on the current temperature using a simulated annealing algorithm, wherein operators with better historical exploration performance are prioritized during high-temperature stages, and operators with better historical optimization performance are prioritized during low-temperature stages; executing the selected destruction operator to remove some cargo, wherein the destruction operator constructs a support relationship graph between cargoes, evaluates the load balance and support stability of each cargo, and identifies and removes cargoes with stability evaluation values ​​below a preset threshold; executing the selected repair operator to re-insert the removed cargo, generating a neighborhood loading scheme, wherein the repair operator evaluates all candidate loading positions for the cargo to be inserted and selects the optimal position for insertion based on a comprehensive evaluation value; and deciding whether to accept the neighborhood loading scheme based on the simulated annealing criterion, while simultaneously updating the historical performance scores of the operators. Output the load allocation scheme when the preset termination conditions are met.

2. The method according to claim 1, characterized in that, The method also includes pre-obtaining the three-dimensional dimensions, weight, and center of gravity of the cargo set to be loaded; The method of generating an initial loading plan using a layered filling strategy, which prioritizes stacking heavy cargo at the bottom of the container unit and then fills the upper layer and remaining space with light cargo, includes: The cargo to be loaded is divided into heavy cargo and light cargo according to the preset density threshold. The heavy goods are arranged in descending order of weight and stacked from the bottom of the container unit towards the center along a predetermined path. The lighter goods are arranged in descending order of volume and filled into the remaining space above the heavier goods.

3. The method according to claim 1, characterized in that, The current temperature-based destruction and repair operator based on the simulated annealing algorithm includes: Compare the current temperature with the preset temperature boundary point to distinguish between high and low temperature stages; For any operator o in the operator set O, the historical exploration performance score is: Historical optimization performance score: ; The probability of operator o being selected when in the high-temperature phase. ; When in the low-temperature phase, the probability of operator o being selected. ; Where j is the index of the operator in operator set O.

4. The method according to claim 2, characterized in that, The structural stability-based failure operator removes some cargo. This failure operator constructs a support relationship graph between cargoes, evaluates the load balance and support stability of each cargo, and identifies and removes cargoes with stability evaluation values ​​below a preset threshold, including: Construct a directed support relationship graph between goods; Calculate the load balance and support stability of each cargo i; The load balance and support stability are weighted and summed to obtain the stability assessment value for each cargo i. Remove goods whose stability assessment value is lower than the preset threshold.

5. The method according to claim 1, characterized in that, The repair operator evaluates all candidate loading positions for the cargo to be inserted and selects the optimal position for insertion based on the comprehensive evaluation value, including: For each candidate loading location, scores are evaluated in at least four dimensions, including: impact on the overall center of gravity of the container unit, stability of the contact surface with adjacent cargo, degree of aggregation with cargo destined for the same port within the predetermined neighborhood, and degree of improvement on the structural stability of the local area where the cargo is located. The scores of the at least four dimensions are weighted and summed to obtain a comprehensive evaluation value, and the position with the highest comprehensive evaluation value is selected for insertion.

6. The method according to claim 1, characterized in that, The process of determining whether to accept the neighborhood loading scheme based on the simulated annealing criterion and simultaneously updating the historical performance score of the operator includes: After iteration, if the generated neighborhood loading scheme is better than the current best scheme, add a historical optimization performance score to the currently executed operator; When the generated neighborhood loading scheme is a new scheme, add a historical exploration performance score to the currently executed operator; At the end of each iteration, for all operators o in the operator set O, both scores are multiplied by a preset decay coefficient, which is less than 1 and greater than 0.

7. The method according to claim 5, characterized in that, The preset termination conditions include: The total number of iterations has reached the set maximum number of iterations; Alternatively, the simulated annealing algorithm can reduce the current temperature to a preset termination temperature; Alternatively, the objective function value of the optimal loading scheme does not improve within a preset number of consecutive iterations.

8. The method according to claim 1, characterized in that, The construction of the support relationship diagram between goods and the evaluation of the load balance and support stability of each goods include: Construct a directed acyclic graph with the container unit floor as the root node and each cargo as a child node. When cargo A is located above cargo B and the two are in physical contact, construct a directed edge from B to A. The load balance is defined as the ratio of the effective support area obtained by the bottom surface of the cargo to the total area of ​​the bottom surface of the cargo. The support stability is defined as the normalized value of the minimum distance from the projection point of the cargo's center of gravity onto the horizontal plane to the boundary of the support polygon below it, wherein the support polygon is formed by combining the top surface contours of the lower layer of cargo supporting the cargo.

9. The method according to claim 1, characterized in that, The step of determining whether to accept the neighborhood loading scheme based on the simulated annealing criteria includes: A global objective function is constructed, which is a weighted sum of three indicators: the space utilization rate of the container unit, the distance deviation between the overall center of gravity and the geometric center of the loading scheme, and the average stability assessment value of all cargoes. Calculate the difference in global objective function values ​​between the neighboring loading scheme and the current loading scheme; If the difference indicates that the neighboring loading scheme is better than the current loading scheme, then the neighboring loading scheme is accepted; if the difference indicates that the neighboring loading scheme is worse than the current loading scheme, then the acceptance probability is calculated based on the ratio of the difference to the current temperature, and a random decision is made on whether to accept based on the acceptance probability.

10. An intelligent loading system for air cargo, characterized in that, The intelligent cargo loading method for implementing any one of claims 1-9 includes the following modules: The generation module is used to generate an initial loading plan using a layered filling strategy, prioritizing the stacking of heavy cargo at the bottom of the container unit, and then filling the upper layer and remaining space with light cargo. An update module is used to iteratively optimize the initial loading scheme. This iterative optimization includes: selecting a destruction operator and a repair operator based on the current temperature using a simulated annealing algorithm, where operators with better historical exploration performance are prioritized during high-temperature phases, and operators with better historical optimization performance are prioritized during low-temperature phases; executing the selected destruction operator to remove some cargo, where the destruction operator constructs a support relationship graph between cargoes, evaluates the load balance and support stability of each cargo, and identifies and removes cargoes with stability evaluation values ​​below a preset threshold; executing the selected repair operator to re-insert the removed cargo, generating a neighborhood loading scheme, where the repair operator evaluates all candidate loading positions for the cargo to be inserted and selects the optimal position for insertion based on a comprehensive evaluation value; and deciding whether to accept the neighborhood loading scheme based on simulated annealing criteria, while simultaneously updating the historical performance scores of the operators. The output module is used to output the load allocation scheme when the preset termination conditions are met.

Images