Machine management method of enterprise warehouse cargo management system based on artificial intelligence

Abstract

The invention discloses an enterprise warehouse cargo management system based on artificial intelligence, and the system comprises a cargo pick-and-place information storage module which is used for storing the storage position data of cargos and the previous pick-and-place path data of the cargos; the AGV comprehensive regulation and control module is used for controlling the running path and the running real-time state of the AGV; and the cargo storage position and pick-and-place path selection module is used for optimizing the storage position and the pick-and-place path of the cargo. According to the method, the defects in the prior art can be overcome, the adaptability of the ant colony algorithm to multi-target optimization is improved by optimizing the target data fusion method and the ant colony optimization process, and efficient management of the whole storage process from transportation to storage is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of warehousing and logistics optimization, and in particular to a machine management method for an enterprise warehouse cargo management system based on artificial intelligence. Background Art

[0002] With the widespread adoption of the Internet of Things and automated control technologies, AGVs are now widely used in large corporate warehouses to transport goods. AGVs achieve precise positioning through technologies such as laser navigation and magnetic stripe navigation, enabling 24-hour uninterrupted operation. The coordinated operation of multiple AGVs can effectively improve response speed. The ant colony algorithm, a path optimization algorithm, is often used for adaptive navigation of AGVs. However, traditional ant colony algorithms have limited multi-objective optimization capabilities, so optimization of AGV navigation routes is mostly limited to path length, failing to achieve unified scheduling of path utilization and cargo storage locations. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a machine management method for an enterprise warehouse cargo management system based on artificial intelligence, which can solve the shortcomings of the existing technology, realize the process of optimizing target data fusion method and ant colony optimization, improve the adaptability of ant colony algorithm for multi-objective optimization, and realize efficient management of the entire warehousing process from transportation to storage.

[0004] In order to solve the above technical problems, the technical solutions adopted by the present invention are as follows.

[0005] An enterprise warehouse cargo management system based on artificial intelligence, including:

[0006] The cargo access information storage module is used to store the cargo storage location data and the cargo access path data;

[0007] AGV transport vehicle comprehensive control module, used to control the AGV transport vehicle's running path and real-time running status;

[0008] The cargo storage location and pick-up path selection module is used to optimize the storage location and pick-up path of the cargo.

[0009] A management method for the above-mentioned enterprise warehouse cargo management system based on artificial intelligence includes the following steps:

[0010] The cargo storage location and pick-up / placement path selection module reads the cargo storage location data and the cargo pick-up / placement path data of the cargo through the cargo pick-up / placement information storage module, reads the current AGV transport vehicle operation path data through the AGV transport vehicle comprehensive control module, synthesizes the read data into a path data group with characteristic nodes, and inputs the path data group and the current cargo pick-up / placement demand into the ant colony algorithm model to obtain the optimal path planning and / or the optimal cargo storage location;

[0011] The AGV transport vehicle integrated control module transports the goods according to the obtained optimal path planning and / or optimal storage location of the goods, and the goods pick-up and placement information storage module updates the storage location data of the goods and the data of the previous pick-up and placement paths of the goods.

[0012] Preferably, the synthesis of the path data set with characteristic nodes comprises the following steps:

[0013] For each combination of "pickup port-cargo storage location", a corresponding path set including all path combinations is generated; the average time for picking up and placing goods corresponding to the cargo storage location at the end of each path in the path set is converted into the first feature point of the path, and the average passing time of each path segment on each path is converted into the corresponding second feature point.

[0014] Preferably, the transformation of the first feature point includes the following steps:

[0015] The path data and the average time of picking up and putting goods are synchronized in the time dimension, and a timestamp mapping relationship φ:t1→t2 is established, where t1 is the time when the path data is collected, and t2 is the average time when the path data is collected. Then, the path data and the average time when the goods are picked up and put are aligned in the heterogeneous data dimension, and the processed path data and the average time when the goods are picked up and put are projected into the unified dimensional space to construct a joint data configuration F=φ(f1, f2)·W=(F1, F2), where f1 is the path data, f2 is the average time when the goods are picked up and put, and W is the mapping. The mapping matrix F1 is the constructed path data configuration, and F2 is the constructed time data configuration. The sliding window is used to traverse the joint data configuration F to obtain the feature data set P of the joint data configuration F. The correction function λ(φ) synchronized with the timestamp mapping relationship is calculated according to the feature data sets of different joint data configurations F, so that the Pλ(φ) of different data configurations F are linearly correlated. Fλ(φ) is used as the calculation object for inverse mapping calculation to obtain the path data with the first feature point, where the inverse mapping result of the time data configuration F2 is the first feature point data.

[0016] Preferably, the transformation of the second feature point includes the following steps:

[0017] Increase the data dimension of the path data with the first feature point, fill the average passing time of each path segment into the newly added data dimension, and solve the optimal data matching configuration in the parameter constraint space according to the taboo search algorithm. Reconstruct the path data according to the obtained optimal data matching configuration to obtain the path data with the second feature point.

[0018] Preferably, the ant colony algorithm model calculates the optimal path planning and / or the optimal storage location of the goods including the following steps:

[0019] Initialize the ant colony and use the path data set with characteristic nodes as the ant colony candidate path set. The ant colony selects the path according to the probability of Randomly select a path, where τ is the pheromone concentration, η is the heuristic factor, α is the pheromone weight factor, β is the heuristic factor weight, and allowed k is the set of nodes currently accessible to ant k; after the ant colony completes a path, it follows τ ij (t+1)=(1-ρ)·[τ ij (t)+Δτ ij ] Update the pheromone concentration on each path, ρ is the pheromone volatility coefficient, Δτ ij It is the pheromone increment after the ants pass through the path. When the maximum number of iterations is reached or the optimal path is stable, the optimal combination of path data with characteristic nodes is output.

[0020] Preferably, the pheromone concentration τ is updated twice according to the first characteristic point and the second characteristic point.

[0021] After each update of the pheromone concentration τ, the first feature point data and the second feature point data on the path are updated. If the second feature point data increases, the pheromone concentration of this path segment is reduced. If the second feature point data decreases, the pheromone concentration of this path segment is increased. A correspondence between the first feature point and the pheromone concentration range is established. If the pheromone concentration on the current path exceeds the pheromone concentration range corresponding to the first feature point, the pheromone concentration on the current path is corrected, and the maximum correction amplitude does not exceed 10% of the current pheromone concentration.

[0022] As a preferred method, a dynamic masking training strategy is introduced to randomly mask part of the path according to the masking probability during the iteration process. The masking probability P s =ρ / τ, ρ is the path time coefficient, Where k is the proportional coefficient, T1 is the first characteristic point data, and T2 is the second characteristic point; the number of shielded paths accounts for 3% to 6% of the total number of paths.

[0023] The beneficial effect of adopting the above technical solution is that: the present invention improves the adaptability of the ant colony algorithm for multi-objective optimization by optimizing the integration method of warehousing logistics data and the iterative process of the ant colony algorithm, and realizes efficient management of the entire warehousing process from transportation to storage. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a schematic diagram of a specific embodiment of the present invention. DETAILED DESCRIPTION

[0025] The ant colony algorithm (ACO) demonstrates unique advantages in path optimization by simulating the pheromone positive feedback mechanism in the foraging behavior of ant colonies. However, in the field of warehousing and logistics, path optimization is no longer a single distance minimization objective. Instead, it becomes a comprehensive decision-making task that requires dynamically balancing multi-dimensional variables and adapting to real-time environmental changes. This makes the limitations of the traditional ACO increasingly prominent in multivariable scenarios. Because the pheromone update mechanism of the traditional ACO is based on the assumption of a static environment, when encountering dynamic multi-parameter target inputs, the traditional ACO simplifies the multiple objectives into a single objective function through linear weighting. This processing method not only fails to accurately reflect the nonlinear relationship between the various indicators, but can also lead to suboptimal solutions due to biased weight settings. When the constraints exhibit dynamic coupling characteristics, the single-dimensional pheromone representation system cannot effectively encode complex constraint relationships and is prone to invalid solutions.

[0026] In response to the above problems, the present invention proposes a new enterprise warehouse cargo management process based on artificial intelligence.

[0027] Reference Figure 1 The present invention provides an enterprise warehouse cargo management system based on artificial intelligence, including:

[0028] The cargo pick-up and placement information storage module 1 is used to store cargo storage location data and cargo pick-up and placement path data;

[0029] AGV transport vehicle comprehensive control module 2, used to control the AGV transport vehicle's running path and real-time running status;

[0030] The cargo storage location and pick-up / drop-off path selection module 3 is used to optimize the cargo storage location and pick-up / drop-off path.

[0031] Based on an intelligent warehousing system, this system uses a network of RFID readers and electronic tags to create a digital warehouse map centered on cargo location coordinates. Each shelf unit is equipped with a pressure sensor and infrared detection device to monitor the storage and access status of goods in real time. The data is pre-processed by edge computing nodes and stored in a distributed database.

[0032] The cargo storage location and pick-up and placement path selection module 3 reads the cargo storage location data and the cargo pick-up and placement path data of the cargo through the cargo pick-up and placement information storage module 1, reads the current AGV transport vehicle operation path data through the AGV transport vehicle comprehensive control module 2, synthesizes the read data into a path data group with feature nodes, and inputs the path data group and the current cargo pick-up and placement requirements into the ant colony algorithm model to obtain the optimal path planning and / or the optimal cargo storage location.

[0033] To mitigate the challenges faced by the ant colony algorithm when dealing with dynamic changes in multiple parameters, this invention converts the average cargo pickup and dropoff time corresponding to the cargo storage location at the end of each path in the path set into the first characteristic point of that path, and converts the average transit time of each path segment on each path into the corresponding second characteristic point. Each "cargo pickup port-cargo storage location" combination corresponds to a path set that includes all path combinations. Path data with these two types of characteristic points preserves the original path data structure to the greatest extent possible, thereby reducing the adaptation workload for the ant colony algorithm.

[0034] By collaboratively modeling route planning and cargo pickup and placement times, we obtained the first characteristic point. The route data and the average cargo pickup and placement times were synchronized in the temporal dimension, establishing a timestamp mapping relationship φ:t1→t2, where t1 is the route data collection time and t2 is the average cargo pickup and placement time corresponding to the route data collection time. The heterogeneous data dimensions of the route data and the average cargo pickup and placement time were then aligned and projected into a unified dimensional space, constructing a joint data configuration F = φ(f1, f2)·W = (F1, F2), where f1 is the route data, f2 is the average cargo pickup and placement time, and W is the mapping matrix. F1 is the constructed route data configuration, and F2 is the constructed time data configuration. The mapping matrix W is solved using a constrained Frobenius norm optimization. The route data and the average cargo pickup and placement time were tensor-normalized, and an improved Hilbert-Huang transform was used to eliminate phase offsets caused by sampling frequency differences. A dynamic time warping algorithm was used to establish the correspondence between unequally spaced data points. A variable-width Hanning window is used to traverse the joint data configuration F to obtain its feature dataset P. A dynamic adjustment mechanism for the window overlap ratio is used to balance data continuity and computational load. A correction function λ(φ) is calculated based on the feature datasets of different joint data configurations F, synchronized with the timestamp mapping relationship. This ensures that Pλ(φ) for different data configurations F is linearly correlated. Using Fλ(φ) as the computational object, a Moore-Penrose pseudo-inverse operator with a Tikhonov regularization term is constructed for inverse mapping, resulting in path data with the first feature point. The inverse mapping result of the time data configuration F2 is the first feature point data.

[0035] The data dimension of the path data with the first characteristic point is increased. Time series data of vehicle traffic is extracted for each path segment. After eliminating abnormal fluctuations using a sliding average filter algorithm, the calculated average transit time is used as a new spatiotemporal coupling characteristic dimension. The average transit time for each path segment is then populated into this new data dimension. A tabu search algorithm is then used to find the optimal data matching configuration within the parameter constraint space. A dynamic tabu table management mechanism (including a short-term memory tabu strategy and a long-term frequency penalty function) effectively avoids local optimality. The neighborhood solution generation module uses a spatial interpolation method based on Delaunay triangulation to ensure that parameter adjustments comply with geographic spatial constraints. The path data is reconstructed based on the obtained optimal data matching configuration, generating path data with the second characteristic point. Multi-resolution fusion technology is used in the data reconstruction phase to map the optimal matching configuration back to the original path data space. Path segment junctions are first smoothed using cubic spline interpolation to eliminate geometric abrupt changes caused by parameter adjustments. The first characteristic points (including key locations such as road intersections and speed change points) are then extracted using an improved Douglas-Peucker algorithm. The energy function of the feature point is established by combining the spatiotemporal coupling characteristics, and the second feature point with dual time-space attributes is generated after iterative optimization using the gradient descent method.

[0036] Next, we use the ant colony algorithm to optimize the path that combines the two types of feature data. The path data set with characteristic nodes is used as the ant colony candidate path set, and the ant colony selects the path based on the probability of Randomly select a path, where τ is the pheromone concentration, η is the heuristic factor, α is the pheromone weight factor, β is the heuristic factor weight, and allowed k is the set of nodes currently accessible to ant k; after the ant colony completes a path, it follows τ ij (t+1)=(1-ρ)·[τ ij (t)+Δτ ij ] Update the pheromone concentration on each path, ρ is the pheromone volatility coefficient, Δτ ij It is the pheromone increment after the ants pass through the path. When the maximum number of iterations is reached or the optimal path is stable, the optimal combination of path data with characteristic nodes is output.

[0037] After each update of the pheromone concentration τ, the first feature point data and the second feature point data on the path are updated. If the second feature point data increases, the pheromone concentration of this path segment is reduced. If the second feature point data decreases, the pheromone concentration of this path segment is increased. A correspondence between the first feature point and the pheromone concentration range is established. If the pheromone concentration on the current path exceeds the pheromone concentration range corresponding to the first feature point, the pheromone concentration on the current path is corrected, and the maximum correction amplitude does not exceed 10% of the current pheromone concentration.

[0038] A dynamic masking training strategy is introduced to randomly mask some paths according to the masking probability during the iteration process. In this optimization model, the implementation of the dynamic masking training strategy requires the combination of multi-source data fusion and adaptive parameter adjustment mechanism. The core parameter of this strategy is the masking probability P. s =ρ / τ, ρ is the path time coefficient, Where k is a proportional coefficient dynamically adjusted based on feature point data, which is updated in real time by the Kalman filter to adapt to the model requirements of different training stages. T1 is the first feature point data, and T2 is the second feature point. The number of shielded paths accounts for 3% to 6% of the total number of paths. The masking operation is performed using the Monte Carlo random sampling method. In each round of iteration, the dynamic calculated P s The value generates a shielding matrix that follows a Bernoulli distribution. When the cumulative shielding rate exceeds the threshold, the reverse compensation mechanism is automatically triggered.

[0039] This strategy is deeply coupled with the residual network architecture during the model training phase, and a dual-channel feature extraction module is designed to process the original path data and the masked reconstruction data respectively. Force the model to learn stable feature representation under masking perturbations, where f(·) represents the feature extractor, The path data after masking.

[0040] Finally, the AGV transporter integrated control module 2 transports the goods according to the obtained optimal path planning and / or optimal storage location of the goods, and the goods pick-up and placement information storage module 1 updates the storage location data of the goods and the data of the previous pick-up and placement paths of the goods.

[0041] This invention conducts groundbreaking research in dynamic heterogeneous data fusion, coordinated control of time-varying parameters, and multimodal inverse mapping verification. After deployment at Yunda's Nanjing distribution center in Jiangsu Province, a 72-hour stress test showed a 23% increase in overall system logistics efficiency and a 93.1% comprehensive equipment utilization rate.

[0042] In the description of the present invention, it should be understood that the terms "longitudinal", "transverse", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.

[0043] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. An enterprise warehouse cargo management system based on artificial intelligence, characterized by: include, A goods pick-up and placement information storage module (1) is used to store the storage location data of the goods and the data of the previous pick-up and placement paths of the goods; AGV transport vehicle comprehensive control module (2), used to control the running path and real-time running status of the AGV transport vehicle; The cargo storage location and pick-up and placement path selection module (3) is used to optimize the cargo storage location and pick-up and placement path.

2. A management method for an enterprise warehouse cargo management system based on artificial intelligence according to claim 1, characterized in that The following steps are involved: The cargo storage location and pick-up / placement path selection module (3) reads cargo storage location data and previous cargo pick-up / placement path data through the cargo pick-up / placement information storage module (1), reads current AGV transport vehicle operation path data through the AGV transport vehicle integrated control module (2), synthesizes the read data into a path data group with characteristic nodes, and inputs the path data group and current cargo pick-up / placement requirements into an ant colony algorithm model to obtain an optimal path planning and / or an optimal cargo storage location; The AGV transporter integrated control module (2) transports the goods according to the obtained optimal path planning and / or optimal storage location of the goods, and the goods pick-up and placement information storage module (1) updates the storage location data of the goods and the data of the previous pick-up and placement paths of the goods.

3. The management method of enterprise warehouse cargo management system based on artificial intelligence according to claim 2 is characterized in that The following steps are involved: The synthesis of the path data set with characteristic nodes includes the following steps: For each "pickup port - cargo storage location" combination, a corresponding path set including all path combinations is generated; the average cargo pickup and placement time corresponding to the cargo storage location at the end of each path in the path set is converted into the first feature point of the path, and the average transit time of each path segment on each path is converted into the corresponding second feature point.

4. The management method of the enterprise warehouse cargo management system based on artificial intelligence according to claim 3 is characterized by: The transformation of the first feature point includes the following steps: The path data and the average time of picking up and putting goods are synchronized in the time dimension, and a timestamp mapping relationship φ:t1→t2 is established, where t1 is the time when the path data is collected, and t2 is the average time when the path data is collected. Then, the path data and the average time when the goods are picked up and put are aligned in the heterogeneous data dimension, and the processed path data and the average time when the goods are picked up and put are projected into the unified dimensional space to construct a joint data configuration F=φ(f1, f2)·W=(F1, F2), where f1 is the path data, f2 is the average time when the goods are picked up and put, and W is the mapping. The mapping matrix F1 is the constructed path data configuration, and F2 is the constructed time data configuration. The sliding window is used to traverse the joint data configuration F to obtain the feature data set P of the joint data configuration F. The correction function λ(φ) synchronized with the timestamp mapping relationship is calculated according to the feature data sets of different joint data configurations F, so that the Pλ(φ) of different data configurations F are linearly correlated. Fλ(φ) is used as the calculation object for inverse mapping calculation to obtain the path data with the first feature point, where the inverse mapping result of the time data configuration F2 is the first feature point data.

5. The management method of the enterprise warehouse cargo management system based on artificial intelligence according to claim 4 is characterized by: The transformation of the second feature point includes the following steps: Increase the data dimension of the path data with the first feature point, fill the average passing time of each path segment into the newly added data dimension, and solve the optimal data matching configuration in the parameter constraint space according to the taboo search algorithm. Reconstruct the path data according to the obtained optimal data matching configuration to obtain the path data with the second feature point.

6. The management method of the enterprise warehouse cargo management system based on artificial intelligence according to claim 5 is characterized by: The ant colony algorithm model calculates the optimal path planning and / or the optimal storage location of the goods, including the following steps: Initialize the ant colony and use the path data set with characteristic nodes as the ant colony candidate path set. The ant colony selects the path according to the probability of Randomly select a path, where τ is the pheromone concentration, η is the heuristic factor, α is the pheromone weight factor, β is the heuristic factor weight, and allowed k For ants k The set of currently accessible nodes; after the ant colony completes a path, it follows τ ij (t+1)=(1-ρ)·[τ ij (t)+Δτ ij ] Update the pheromone concentration on each path, ρ is the pheromone volatility coefficient, Δτ ij It is the pheromone increment after the ants pass through the path. When the maximum number of iterations is reached or the optimal path is stable, the optimal combination of path data with characteristic nodes is output.

7. The management method of the enterprise warehouse cargo management system based on artificial intelligence according to claim 6 is characterized by: The pheromone concentration τ is updated twice according to the first and second feature points. After each update of the pheromone concentration τ, the first feature point data and the second feature point data on the path are updated. If the second feature point data increases, the pheromone concentration of this path segment is reduced. If the second feature point data decreases, the pheromone concentration of this path segment is increased. A correspondence between the first feature point and the pheromone concentration range is established. If the pheromone concentration on the current path exceeds the pheromone concentration range corresponding to the first feature point, the pheromone concentration on the current path is corrected, and the maximum correction amplitude does not exceed 10% of the current pheromone concentration.

8. The management method of the enterprise warehouse cargo management system based on artificial intelligence according to claim 7 is characterized by: Introducing a dynamic masking training strategy, in the iterative process, some paths are randomly masked according to the masking probability. The masking probability P s =ρ / τ, ρ is the path time coefficient, Where k is the proportional coefficient, T1 is the first characteristic point data, and T2 is the second characteristic point; the number of shielded paths accounts for 3% to 6% of the total number of paths.

Images