Intelligent logistics warehousing method and system based on WMS
Through image acquisition and cargo identification algorithms, combined with WMS data, a three-dimensional grid model and cargo location decision model are built, which solves the problem of traditional WMS relying on manual experience, realizes intelligent warehousing management, and improves the storage and access efficiency and accuracy of the warehouse .
Patent Information
- Application Number
- CN202411094501.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-10
- Publication Date
- 2024-11-26
- Estimated Expiration
- Not applicable · inactive patent
Smart Images

Figure CN119018515A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of intelligent warehousing technology, and in particular to an intelligent logistics warehousing method and system based on WMS. Background Art
[0002] With the rapid development of e-commerce and the continuous upgrading of the logistics industry, the traditional warehouse management model can no longer meet the needs of modern logistics business. In order to improve the efficiency of warehouse operations, reduce labor costs, and realize the intelligence and automation of warehouse management, intelligent logistics warehousing technology based on WMS (Warehouse Management System) came into being.
[0003] Existing WMS mainly manages and optimizes the warehouse operation process through information technology, including warehousing and outbound delivery. However, traditional WMS still has some limitations in practical applications. The scheduling and optimization of warehousing operations mainly rely on manual experience and lack the support of intelligent algorithms, making it difficult to achieve global optimization.
[0004] To sum up, in order to overcome the above limitations, there is an urgent need for an intelligent logistics warehousing method based on WMS. By introducing advanced technologies such as the Internet of Things and artificial intelligence, the intelligent and automated warehousing operations can be realized, and the efficiency and accuracy of warehousing management can be improved. The present invention can solve the problems in the prior art. Summary of the invention
[0005] The embodiments of the present invention provide a WMS-based intelligent logistics warehousing method and system, which can solve the problems in the prior art.
[0006] According to a first aspect of the embodiments of the present invention,
[0007] Provided is a WMS-based intelligent logistics warehousing method, including:
[0008] Acquire image information and weight information of incoming goods through image acquisition equipment and weighing equipment, and determine comprehensive goods data of the incoming goods based on the image information and weight information and through a goods recognition algorithm in combination with goods feature data in the WMS;
[0009] Based on the warehouse data obtained from the WMS, a three-dimensional warehouse grid model is constructed to determine the decision constraints. Combined with the comprehensive data of the incoming goods, a cargo location decision model is constructed. By searching for the optimal cargo location allocation strategy, the optimal incoming cargo location is determined, and the incoming goods are stored in the optimal incoming cargo location.
[0010] Obtain outbound orders from WMS, generate picking tasks, determine picking locations based on outbound cargo data in the picking tasks, and build a picking scheduling optimization algorithm based on the current position of the carrier robot, with the goal of minimizing the idle time of the carrier robot and minimizing the picking movement time. Generate the optimal picking strategy, determine outbound control instructions, and control the carrier robot to perform outbound actions.
[0011] In an optional embodiment,
[0012] The image information and weight information of the incoming goods are obtained by the image acquisition device and the weighing device. Based on the image information and the weight information, the goods identification algorithm is used in combination with the goods feature data in the WMS to determine the comprehensive goods data of the incoming goods, including:
[0013] graying, denoising and image enhancement preprocessing are performed on the image information to determine a standard cargo image;
[0014] Based on a preset feature point detection threshold and a preset scale space parameter, feature extraction is performed on the standard cargo image to obtain a set of initial feature points and corresponding scale information, and a corresponding feature descriptor is calculated;
[0015] Extracting local histogram features and local gradient features from the local area around each of the initial feature points, determining local image features, calculating local complexity, and combining the feature density corresponding to the initial feature points, updating the feature point detection threshold and scale space parameters, combining the local image features with the initial feature points, and recalculating feature descriptors;
[0016] Iterate in sequence until a preset number of iterations is reached, determine valid feature points according to the feature descriptor and a preset descriptor threshold, and extract the features of the cargo image;
[0017] Matching the cargo image features with the cargo feature data in the WMS, calculating the matching degree, selecting the cargo with the highest matching degree as the matching result, and obtaining the corresponding cargo coding information;
[0018] The weight information is compared and verified with the standard weight in the cargo coding information. When the absolute value of the difference between the weight information and the standard weight is less than a preset weight error threshold, the comprehensive cargo data is confirmed.
[0019] In an optional embodiment,
[0020] Based on the warehouse data obtained from the WMS, a three-dimensional warehouse grid model is constructed, decision constraints are determined, and a cargo location decision model is constructed in combination with the comprehensive data of the incoming goods. The optimal incoming cargo location is determined by searching for the optimal cargo location allocation strategy, and the incoming goods are stored in the optimal incoming cargo location, including:
[0021] Based on the warehouse data obtained from the WMS, the cargo location layout, cargo location attributes and cargo location constraints are extracted, and based on the cargo location layout, a three-dimensional grid is determined. In combination with the cargo location attributes and the cargo location constraints, the warehouse is abstracted into a warehouse three-dimensional grid model, the distance matrix between cargo locations is determined, and decision constraints are constructed;
[0022] Based on the location and quantity of the goods, the corresponding cargo space allocation is determined, and the state space is constructed. According to the allocation action of allocating the incoming goods to the corresponding cargo spaces, the action space is constructed. According to the cargo space attributes and decision constraints, the transition probability matrix between the states in the state space is determined. According to the cargo space allocation efficiency and the cargo space utilization rate, a reward function is generated, and the evaluation quality of each state in the state space is determined. With the goal of maximizing the cumulative reward of the evaluation quality, a cargo space decision model is constructed;
[0023] Use a multi-level search algorithm to solve the cargo location decision model, determine the candidate action sequence through global search, perform local optimization search on the candidate action sequence, determine multiple groups of cargo location allocation strategies, calculate the cumulative rewards of the cargo location allocation strategies, select the cargo location allocation strategy corresponding to the maximum value of the cumulative reward, and determine the optimal storage cargo location.
[0024] In an optional embodiment,
[0025] Determining candidate action sequences through global search includes:
[0026] Determine a root node of a global search tree based on a current state in the state space of the warehouse;
[0027] Starting from the root node, perform the iteration operation:
[0028] Calculate the node search value based on the node search value formula, and select the node with the highest node search value as the benchmark child node;
[0029] If the reference child node is not a terminal node, randomly generating a plurality of extended child nodes based on the transition probability matrix, and adding the extended child nodes to the global search tree;
[0030] Based on each extended child node, starting from the corresponding extended child node, based on the reward function, a simulation operation is performed to evaluate the cumulative reward value of each extended child node;
[0031] Back-propagating the accumulated reward value obtained by the simulation operation to the root node, and updating the statistical information of each child node in the global search tree, wherein the statistical information includes the number of visits and the average reward value;
[0032] Repeatedly generate child nodes and update the statistical information until the preset maximum depth of the global search tree is reached, start from the root node, traverse all child nodes to generate corresponding action sequences, and determine the candidate action sequence according to the preset number of candidate sequences.
[0033] In an optional embodiment,
[0034] Based on the node search value formula, the node search value is calculated, and the node with the highest node search value is selected as the benchmark child node including:
[0035] The node search value formula is as follows:
[0036]
[0037] Among them, o represents the action number, P o represents the node search value of action o, represents the average reward value of action o, C(N) represents the exploration item, N represents the total number of trials, and n o represents the number of times action o is selected, C 0 represents the initial value of the exploration term, and α represents the decay rate.
[0038] In an optional embodiment,
[0039] Obtain outbound orders from WMS, generate picking tasks, determine picking locations based on outbound cargo data in the picking tasks, and build a picking scheduling optimization algorithm based on the current position of the carrier robot with the goal of minimizing the idle time of the carrier robot and minimizing the picking movement time. Generate the optimal picking strategy, including:
[0040] Based on the outbound cargo data in the picking task, all picking locations are determined, the target point is determined, and based on the current position of the carrier robot, the starting point is determined; minimizing the idle time of the carrier robot is taken as the first goal, minimizing the picking movement time is taken as the second goal, the first goal and the second goal are dynamically weighted to determine the comprehensive objective function;
[0041] Based on the target point, starting point and comprehensive objective function, a picking scheduling optimization algorithm is constructed;
[0042] The picking scheduling optimization algorithm includes:
[0043] According to the preset number of water drops, a water drop group is randomly determined, wherein each water drop corresponds to a picking strategy, and based on the comprehensive objective function, the strategy fitness of each water drop is calculated; based on the strategy fitness and the distance between water drops in the solution space, the water drop gravity and the direction of the water drop gravity are determined, wherein the water drop gravity is inversely proportional to the distance between water drops, and the direction of the water drop gravity is from the water drop with a smaller strategy fitness to the water drop with a larger strategy fitness;
[0044] Based on the water droplet gravity and the water droplet gravity direction, updating the water droplet position in the solution space, and determining the updated water droplet;
[0045] Based on the update droplet, a neighborhood droplet is generated within a preset neighborhood exploration radius, the neighborhood strategy fitness corresponding to the neighborhood droplet and the update strategy fitness corresponding to the update droplet are calculated, and the neighborhood strategy fitness and the update strategy fitness are compared. If the neighborhood strategy fitness is greater than the update strategy fitness, the update droplet is replaced with the neighborhood droplet;
[0046] Based on the preset initial evaporation amount and the number of iterations, the evaporation amount of the water drop is calculated, and according to the evaporation amount of the water drop, the strategy fitness corresponding to the water drop is reduced. When the strategy fitness is less than the preset strategy fitness threshold, the water drop is removed from the water drop group, and the same number of supplementary water droplets as the removed water droplets are randomly generated and added to the water drop group, and the total number of water droplets in the water drop group remains unchanged;
[0047] The next round of iteration begins until the number of iterations reaches a preset upper limit of the number of iterations, and the optimal droplet corresponding to the maximum strategy fitness is extracted, and the optimal droplet corresponds to the optimal picking strategy.
[0048] In an optional embodiment,
[0049] Taking minimizing the idle time of the transport robot as the first goal and minimizing the picking movement time as the second goal, the first goal and the second goal are dynamically weighted to determine the comprehensive objective function including:
[0050] The first objective function f1 corresponding to the first objective:
[0051]
[0052] Where I represents the total number of picking tasks, i represents the picking task index, J represents the total number of carrier robots, j represents the carrier robot index, and ci j represents the cost of moving from the starting point to the target point, xi j Indicates whether to assign picking task i to carrier robot j, yi j indicates whether the target point j has been selected for picking task i, λ represents the penalty coefficient, si represents the number of goods for picking task i, and Aj represents the maximum load capacity of the carrier robot j;
[0053] The second objective function f2 corresponding to the second objective is:
[0054]
[0055] Among them, dijk represents the time when the carrier robot j moves to the picking cargo location k when performing the picking task i, β represents the height penalty coefficient, hk represents the height of the picking cargo location k, and zijk represents whether the carrier robot j chooses the picking cargo location k for picking when performing the picking task i;
[0056] The comprehensive objective function F is formulated as follows:
[0057] F = w1 × f1 + w2 × f2;
[0058] Among them, ω1 represents the weight of the first objective function, and ω2 represents the weight of the second objective function.
[0059] In an optional embodiment,
[0060] Also includes:
[0061] The water drop gravity is as follows:
[0062]
[0063] Among them, H pq represents the gravitational force between water droplets p and q, G represents the preset gravitational constant, and F p represents the strategic fitness of the comprehensive objective function calculation of droplet p, F q represents the strategic fitness of the comprehensive objective function calculation of the water droplet q, r pq represents the distance between water drop p and water drop q, ε represents the parameter to avoid division by zero, and δ represents the water drop gravity attenuation factor;
[0064] The formula for calculating the evaporation of water droplets is as follows:
[0065]
[0066] Among them, E(t) represents the function of the evaporation of water droplets changing with the number of iterations, E 0 represents the initial evaporation amount, t represents the number of iterations, T represents the upper limit of the number of iterations, θ represents the evaporation rate control parameter, θ>0, μ represents the evaporation periodicity parameter, μ>0.
[0067] According to a second aspect of the embodiments of the present invention,
[0068] Provide a WMS-based intelligent logistics warehousing system, including:
[0069] The first unit is used to obtain image information and weight information of incoming goods through an image acquisition device and a weighing device, and determine the comprehensive data of the incoming goods based on the image information and the weight information through a goods recognition algorithm and in combination with the goods feature data in the WMS;
[0070] The second unit is used to construct a three-dimensional warehouse grid model based on the warehouse data obtained from the WMS, determine the decision constraints, and build a cargo location decision model in combination with the comprehensive data of the incoming goods, determine the optimal incoming cargo location by searching for the optimal cargo location allocation strategy, and store the incoming goods in the optimal incoming cargo location;
[0071] The third unit is used to obtain the outbound order from the WMS, generate a picking task, determine the picking location based on the outbound cargo data in the picking task, and build a picking scheduling optimization algorithm based on the current position of the carrier robot to minimize the idle time of the carrier robot and minimize the picking movement time, generate the optimal picking strategy, determine the outbound control instructions, and control the carrier robot to perform the outbound action.
[0072] According to a third aspect of the embodiments of the present invention,
[0073] An electronic device is provided, comprising:
[0074] processor;
[0075] a memory for storing processor-executable instructions;
[0076] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.
[0077] In the embodiment of the present invention, through image acquisition equipment and cargo recognition algorithms, automatic identification and classification of incoming cargo can be achieved, reducing manual intervention and errors; by combining image information, weight information and cargo feature data in WMS, the accuracy and completeness of incoming cargo information can be improved, and data inconsistency and errors caused by manual operations can be avoided; by constructing a three-dimensional warehouse grid model and a cargo location decision model, the internal space of the warehouse can be accurately described, the cargo storage location can be reasonably planned, the warehouse space can be maximized, and the storage density and storage efficiency can be improved; by determining the optimal incoming cargo location and formulating the optimal cargo location allocation strategy, the cargo access path can be shortened, the transportation time can be reduced, the cargo access efficiency can be improved, and the operating cost can be reduced; the picking scheduling optimization algorithm can comprehensively consider information such as outbound cargo data, picking tasks and the position of the carrier robot, and generate the optimal picking strategy, so that the path planning of the carrier robot is more reasonable when performing the outbound action, and the picking movement time is minimized, thereby improving the overall operation efficiency of the warehouse. BRIEF DESCRIPTION OF THE DRAWINGS
[0078] Figure 1 The figure is a flow chart of an intelligent logistics warehousing method based on WMS according to an embodiment of the present invention;
[0079] Figure 2 The figure is a schematic diagram of the structure of an intelligent logistics warehousing system based on WMS according to an embodiment of the present invention. DETAILED DESCRIPTION
[0080] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0081] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.
[0082] Figure 1 FIG. 1 is a flow chart of an intelligent logistics warehousing method based on WMS according to an embodiment of the present invention. Figure 1 As shown, the method includes:
[0083] S101. Obtain image information and weight information of incoming goods through image acquisition equipment and weighing equipment, and determine the comprehensive data of the incoming goods based on the image information and weight information through a goods recognition algorithm and in combination with the goods feature data in the WMS;
[0084] The image acquisition device specifically refers to a device used to obtain image information of incoming goods, usually a camera or other image sensor that can take photos or videos of incoming goods;
[0085] The weighing device specifically refers to a device used to measure the weight of goods entering the warehouse. Preferably, a weighing sensor installed at the entrance of the warehouse can accurately measure the weight of the goods;
[0086] The cargo identification algorithm specifically refers to an algorithm for identifying incoming cargo, which identifies and classifies cargo by analyzing image information and weight information in combination with a pre-trained model;
[0087] The cargo characteristic data in the WMS specifically refers to various characteristic data related to the cargo stored in the warehouse management system (WMS), including but not limited to cargo name, model, appearance, packaging style, etc.;
[0088] The comprehensive cargo data specifically refers to the complete data of the incoming cargo obtained through processing and analysis of the image information, weight information and cargo feature data in the WMS, including comprehensive data of information such as the name, weight, quantity, specification, batch number, production date, etc. of the cargo.
[0089] In this embodiment, automatic identification and classification of incoming goods are achieved through image acquisition equipment and cargo recognition algorithms, reducing manual intervention and errors; by combining image information, weight information and cargo feature data in WMS, the accuracy and completeness of incoming cargo information are improved, avoiding data inconsistencies and errors caused by manual operations; the integrated and automated data acquisition process greatly improves the efficiency of warehousing operations, reduces labor costs and time costs, and improves the efficiency and quality of warehouse logistics management.
[0090] In an optional embodiment, the image information and weight information of the incoming goods are obtained by an image acquisition device and a weighing device, and based on the image information and the weight information, a goods recognition algorithm is used in combination with the goods feature data in the WMS to determine the comprehensive goods data of the incoming goods, including:
[0091] graying, denoising and image enhancement preprocessing are performed on the image information to determine a standard cargo image;
[0092] Based on a preset feature point detection threshold and a preset scale space parameter, feature extraction is performed on the standard cargo image to obtain a set of initial feature points and corresponding scale information, and a corresponding feature descriptor is calculated;
[0093] Extracting local histogram features and local gradient features from the local area around each of the initial feature points, determining local image features, calculating local complexity, and combining the feature density corresponding to the initial feature points, updating the feature point detection threshold and scale space parameters, combining the local image features with the initial feature points, and recalculating feature descriptors;
[0094] Iterate in sequence until a preset number of iterations is reached, determine effective feature points according to the feature descriptor and a preset descriptor threshold, and extract the features of the cargo image;
[0095] Matching the cargo image features with the cargo feature data in the WMS, calculating the matching degree, selecting the cargo with the highest matching degree as the matching result, and obtaining the corresponding cargo coding information;
[0096] The weight information is compared and verified with the standard weight in the cargo coding information. When the absolute value of the difference between the weight information and the standard weight is less than a preset weight error threshold, the comprehensive cargo data is confirmed.
[0097] The standard cargo image specifically refers to a standardized cargo image for feature extraction obtained after preprocessing, such as grayscale, denoising and image enhancement, as a reference for feature matching;
[0098] The feature point detection threshold specifically refers to the threshold parameter used in the feature point detection algorithm, which controls the number and quality of feature point detection, and is usually a gray value or a probability value; the scale space parameter specifically refers to the parameter used in the feature detection algorithm to detect feature points at different scales, including parameters such as scale range and scale step size;
[0099] The feature descriptor specifically refers to a vector matrix that describes the local features of an image, and is usually used to represent the local structure and features of an image;
[0100] The local histogram feature specifically refers to the statistical distribution characteristics of the grayscale values of pixels in the local area around the feature point, which can reflect the texture and appearance characteristics of the local image; the local gradient feature specifically refers to the gradient information calculated in the local area around the feature point, which is used to describe the detail features such as the edge and texture of the image;
[0101] The local complexity specifically refers to the complexity of the local image around the feature point, which is used to measure the richness and variability of the local structure of the image; the feature density specifically refers to the distribution density of the feature points on the image, which is used to evaluate the feature richness of the image and the complexity of the local structure.
[0102] The obtained cargo image information is grayed out, the color image is converted into a gray image, and then the gray image is denoised, and the noise interference in the image is removed by using the Gaussian filtering method. Then, the denoised image is enhanced, and the contrast and clarity of the image are improved by using the contrast adjustment technology. After preprocessing, a standardized cargo image is obtained, which is recorded as a standard cargo image;
[0103] According to the characteristics and recognition requirements of the cargo image, the feature point detection threshold and scale space parameters are preset. The feature point detection threshold is used to control the quantity and quality of feature points. A higher threshold will screen out more stable and significant feature points. The scale space parameters are used to define the scale space of the image, including the number of scale layers, scale factor, etc., which determine the performance of feature points at different scales;
[0104] Preferably, the Gaussian pyramid method is used to represent the standard cargo image at multiple scales, with the original image as the bottom layer of the scale space, and then through Gaussian filtering and downsampling, smaller and smaller images are generated layer by layer to form a scale pyramid, and each scale layer corresponds to a different image resolution and detail level; at each scale layer in the scale space, based on the SIFT feature extraction algorithm, key feature points are detected, and potential feature points are determined by finding local extreme points in the image, especially corner points, spots, edge points, etc., and stable and significant feature points are screened out by comparing the response value of the feature point with the feature point detection threshold. For each detected feature point, the scale level of the feature point is determined, and the scale information of the feature point is obtained by tracking the position of the feature point in the scale space;
[0105] For each initial feature point, the image features of the surrounding area are extracted to generate a feature descriptor, preferably a SIFT descriptor. By calculating the direction and amplitude histogram of the pixel gradient in the neighborhood of the feature point, a 128-dimensional feature vector is generated to represent the image features of the area around the feature point, which is rotationally invariant and scale invariant.
[0106] After the feature descriptor is generated, the histogram features of the local area are calculated, the distribution of the grayscale values of the pixels in the local area is counted, and the gradient features of the local area are calculated. The distribution of the direction and size of the pixel gradient in the local area is extracted, and the histogram features and gradient features are combined to obtain the local image features of the feature point; then, based on the local image features, the complexity of the local area is calculated to indicate the richness of the image texture and details in the local area, and the feature point detection threshold and scale space parameters are dynamically updated in combination with the feature density information of the initial feature points, that is, the distribution density of the feature points in the image; the local image features are combined with the corresponding initial feature points to form a complete feature point representation, and the feature descriptor of each feature point is recalculated to fuse the local image features and scale information;
[0107] The feature point extraction and optimization are performed iteratively until the preset number of iterations is reached. In each iteration, the selection and extraction of feature points are continuously optimized according to the updated feature point detection threshold and scale space parameters. After the iteration, the feature descriptor is compared with the preset descriptor threshold to screen out valid feature points, and finally the feature representation of the cargo image is obtained, including a set of optimized feature points and their corresponding feature descriptors.
[0108] The extracted cargo image features are matched with the cargo feature data pre-stored in the WMS. For each candidate cargo, the similarity between its feature data and the image features is calculated to obtain the matching degree. The cargo with the highest matching degree is selected as the matching result, and the coding information corresponding to the cargo is obtained. The actual weight information of the cargo is compared and verified with the standard weight in the matched cargo coding information. The absolute value of the difference between the actual weight and the standard weight is calculated to determine whether it is less than the preset weight error threshold. If the absolute value of the difference is less than the threshold, it is considered that the weight verification has passed, and the comprehensive data of the cargo, including the cargo coding and weight information, is confirmed.
[0109] In this embodiment, a standard cargo image is obtained through preprocessing steps such as grayscale, denoising and image enhancement, which can improve the quality and clarity of the image and serve as a benchmark for feature matching, which is conducive to the accuracy of subsequent feature extraction and matching. A multi-scale feature point detection algorithm is used, combined with local image information such as grayscale histogram and gradient features, to extract stable and significant feature points, and calculate the corresponding feature descriptor, which has scale invariance and rotation invariance and can accurately describe the local structure and features of the image. According to information such as the distribution density and local complexity of the feature points, the feature point detection threshold and scale space parameters are dynamically adjusted to optimize the selection and extraction of feature points and enhance the robustness and adaptability of the algorithm. The extracted image features are matched with the cargo feature data in the WMS, and the cargo with the highest matching degree is selected as the matching result, and weight verification is performed to ensure the accuracy of matching and the authenticity of weight, and improve the reliability and accuracy of cargo identification.
[0110] S102. Based on the warehouse data obtained from the WMS, a three-dimensional warehouse grid model is constructed, decision constraints are determined, and a storage location decision model is constructed in combination with the comprehensive data of the incoming goods. The optimal incoming storage location is determined by searching for the optimal storage location allocation strategy, and the incoming goods are stored in the optimal incoming storage location;
[0111] The warehouse three-dimensional grid model specifically refers to a three-dimensional coordinate system established inside the warehouse, which is used to describe the spatial relationship and layout of different locations in the warehouse. It is usually represented by a three-dimensional coordinate system, including three dimensions of length, width and height. Each coordinate point represents a specific location in the warehouse, and can be used to represent the location and spatial distribution of various elements in the warehouse, such as shelves, cargo spaces, and aisles;
[0112] The decision constraints specifically refer to a series of constraints that must be met when determining the optimal storage location, including but not limited to cargo size, weight, shelf capacity, channel accessibility, safety, etc. Constraints are usually determined by factors such as the physical structure of the warehouse, safety regulations, and equipment performance, and are used to limit the storage location and method of goods to ensure safe and efficient warehouse operations;
[0113] The cargo location decision model specifically refers to determining the optimal warehouse cargo location allocation strategy based on the warehouse's three-dimensional grid model and decision constraints through decision algorithms and optimization methods. The cargo location decision model takes into account factors such as cargo attributes, warehouse structure, and operation requirements, with the goal of maximizing warehouse space utilization and cargo storage and retrieval efficiency. By searching and evaluating different cargo location allocation plans, the most suitable warehouse location is found to achieve effective storage and management of cargo in the warehouse.
[0114] In this embodiment, by constructing a three-dimensional warehouse grid model and a cargo location decision model, the internal space of the warehouse can be accurately described, the cargo storage location can be reasonably planned, the warehouse space can be maximized, and the storage density and storage efficiency can be improved; by determining the optimal storage cargo location and formulating the optimal cargo location allocation strategy, the cargo access path can be shortened, the handling time can be reduced, the cargo access efficiency can be improved, and the operating cost can be reduced; by combining warehouse data and comprehensive cargo data, comprehensive decision constraints can be established, and various restrictions in the cargo storage process can be accurately controlled to ensure the safety and stability of the operation process; by using advanced decision algorithms and optimization methods, the optimal cargo location allocation strategy can be intelligently searched, and dynamic adjustments can be made according to different cargo attributes and warehouse conditions to improve cargo location utilization and the overall warehouse operation efficiency; through automated cargo location allocation and storage processes, errors and uncertainties in human operations can be reduced, the accuracy and stability of warehouse operations can be improved, and the risk of cargo damage and loss can be reduced.
[0115] In an optional embodiment, based on the warehouse data obtained from the WMS, a three-dimensional warehouse grid model is constructed, decision constraints are determined, and a cargo location decision model is constructed in combination with the comprehensive data of the incoming goods. The optimal incoming cargo location is determined by searching for the optimal cargo location allocation strategy, and the incoming goods are stored in the optimal incoming cargo location, which includes:
[0116] Based on the warehouse data obtained from the WMS, the cargo location layout, cargo location attributes and cargo location constraints are extracted, and based on the cargo location layout, a three-dimensional grid is determined. In combination with the cargo location attributes and the cargo location constraints, the warehouse is abstracted into a warehouse three-dimensional grid model, the distance matrix between cargo locations is determined, and decision constraints are constructed;
[0117] Based on the location and quantity of the goods, the corresponding cargo space allocation is determined, and the state space is constructed. According to the allocation action of allocating the incoming goods to the corresponding cargo spaces, the action space is constructed. According to the cargo space attributes and decision constraints, the transition probability matrix between the states in the state space is determined. According to the cargo space allocation efficiency and the cargo space utilization rate, a reward function is generated, and the evaluation quality of each state in the state space is determined. With the goal of maximizing the cumulative reward of the evaluation quality, a cargo space decision model is constructed;
[0118] Use a multi-level search algorithm to solve the cargo location decision model, determine the candidate action sequence through global search, perform local optimization search on the candidate action sequence, determine multiple groups of cargo location allocation strategies, calculate the cumulative rewards of the cargo location allocation strategies, select the cargo location allocation strategy corresponding to the maximum value of the cumulative reward, and determine the optimal storage cargo location.
[0119] Relevant warehouse data is obtained through WMS, including information such as storage location layout, storage location attributes and storage location constraints. The storage location layout represents the spatial distribution and arrangement of storage locations in the warehouse. Storage location attributes include the type, capacity, load-bearing capacity and other characteristics of storage locations. Storage location constraints reflect the relationships and restrictions between storage locations, such as the storage requirements for different types of goods.
[0120] According to the extracted cargo space layout information, the warehouse space is divided into three-dimensional grids, each grid unit represents a cargo space or space unit, and the warehouse is abstractly modeled by combining cargo space attributes and cargo space constraints; each cargo space is mapped to the corresponding position in the three-dimensional grid model, and the cargo space attributes and constraints are associated with the corresponding grid units. According to the spatial relationship between cargo spaces in the three-dimensional grid model, the distance matrix between cargo spaces is calculated; according to cargo space attributes and cargo space constraints, decision constraints are constructed, which reflect the rules and restrictions that cargo space allocation needs to follow, such as the compatibility requirements of goods, the capacity restrictions of cargo spaces, and the loading and unloading sequence of goods.
[0121] According to the location and quantity information of the goods to be stored, the current cargo space allocation is determined. The cargo space allocation indicates the storage location and occupancy of the goods in the warehouse, which provides a basis for constructing the state space and action space; the cargo space allocation is abstracted into the state space, and each state represents a possible cargo space allocation scheme, including the specific location and quantity distribution of the goods in the warehouse. According to the goal of the cargo space decision, the action space is constructed, and each action represents the operation of allocating the incoming goods to a specific cargo space. The action space contains all possible cargo space allocation actions; according to the cargo space attributes and decision constraints, by analyzing factors such as the compatibility and capacity limit of the cargo space, the transition probability matrix between the states in the state space is determined. The state transition probability represents the possibility of transferring from one cargo space allocation state to another; according to the cargo space allocation efficiency and cargo space utilization rate, a reward function is generated to evaluate the quality of each state and measure the advantages and disadvantages of the cargo space allocation scheme. Efficient cargo space allocation and high cargo space utilization rate will obtain higher reward values; the reward function is used to evaluate each state in the state space and calculate the quality value of each state. The quality value of the state reflects the quality of the cargo space allocation plan. Taking the maximization of the cumulative reward of the evaluation quality as the optimization goal, a cargo space decision model is constructed, in which the cumulative reward represents the total reward value obtained in a series of decision steps. By maximizing the cumulative reward, the overall optimal cargo space allocation strategy can be obtained.
[0122] Use a multi-level search algorithm to solve the cargo location decision model. The multi-level search algorithm combines global search and local optimization strategies. Perform a global search in the state space to generate candidate cargo location allocation action sequences. The global search considers the entire state space and can optionally use random sampling and other methods to quickly locate potential decision paths. Perform a local optimization search on the candidate action sequences generated by the global search. For each candidate solution, perform a refined search in the state space near the candidate solution, fine-tune and improve the cargo location allocation strategy, and determine multiple groups of excellent cargo location allocation strategies through global search and local optimization. Each group of strategies corresponds to a decision path, which is a combination of a series of cargo location allocation actions.
[0123] For each group of cargo allocation strategies, calculate the total rewards obtained in each decision step when executing the cargo allocation strategy. The higher the cumulative reward value, the better the overall quality. Compare the cumulative reward values of multiple groups of cargo allocation strategies, and select the strategy with the largest cumulative reward as the optimal strategy. The cargo allocation plan corresponding to the optimal strategy will be used as the execution plan for the warehousing decision, and select which cargo location the corresponding goods need to be stored in.
[0124] In this embodiment, by constructing a three-dimensional grid model and a cargo space decision model based on warehouse data in WMS, the intelligent and automated warehouse management is realized, and the optimal cargo space allocation plan is automatically generated according to factors such as cargo attributes and space restrictions, thereby improving the efficiency and accuracy of warehouse management; by dynamically allocating cargo spaces, the warehouse space is reasonably utilized according to actual conditions, avoiding waste and congestion of cargo spaces, maximizing the utilization rate of warehouse cargo spaces, and improving the storage capacity and efficiency of the warehouse; by solving the cargo space decision model through a multi-level search algorithm, a combination of global search and local optimization is realized, and potential decision paths can be quickly located and refined searches can be performed, thereby improving the optimization effect and decision quality of the cargo space allocation strategy; by evaluating the cumulative reward values of different cargo space allocation strategies, the strategy with the largest cumulative reward is selected as the optimal strategy, thereby maximizing and optimizing the warehousing efficiency and improving the overall efficiency and competitiveness of warehouse management.
[0125] In an optional embodiment, determining the candidate action sequence through global search includes:
[0126] Determine a root node of a global search tree based on a current state in the state space of the warehouse;
[0127] Starting from the root node, perform the iteration operation:
[0128] Calculate the node search value based on the node search value formula, and select the node with the highest node search value as the benchmark child node;
[0129] If the reference child node is not a terminal node, randomly generating a plurality of extended child nodes based on the transition probability matrix, and adding the extended child nodes to the global search tree;
[0130] Based on each extended child node, starting from the corresponding extended child node, based on the reward function, a simulation operation is performed to evaluate the cumulative reward value of each extended child node;
[0131] Back-propagating the accumulated reward value obtained by the simulation operation to the root node, and updating the statistical information of each child node in the global search tree, wherein the statistical information includes the number of visits and the average reward value;
[0132] Repeatedly generate child nodes and update the statistical information until the preset maximum depth of the global search tree is reached, start from the root node, traverse all child nodes to generate corresponding action sequences, and determine the candidate action sequence according to the preset number of candidate sequences.
[0133] The node search value specifically refers to an indicator for evaluating the quality of nodes in the global search tree, which comprehensively considers the number of visits and the average reward value of the node, and is used to guide the search process in a more promising direction.
[0134] The transition probability matrix represents the transition probability between states in the state space. In the cargo location decision model, the transition probability matrix reflects the possibility of transitioning from one cargo location allocation state to another. The calculation of the transition probability takes into account factors such as cargo location attributes, cargo location constraints, and compatibility of goods.
[0135] The simulation operation specifically refers to simulating the process of goods warehousing and storage based on a specific cargo space allocation strategy in the cargo space decision model. Through the simulation operation, the pros and cons of the cargo space allocation plan corresponding to each decision node are evaluated. The simulation operation takes into account factors such as the actual flow direction of the goods, the occupancy of the cargo space, and the transportation distance. The cumulative reward value of each decision node is calculated through the reward function. The simulation operation provides a method for evaluating the quality of decisions without actual execution.
[0136] According to the current storage allocation status of the warehouse, determine the root node of the global search tree in the state space. The root node represents the current decision starting point and is the starting position of the subsequent search. Starting from the root node, perform the following iterative operations:
[0137] Use the node search value formula to calculate the search value of the current node. The search value comprehensively considers the number of node visits and the average reward value, and is used to evaluate the quality of the node. According to the calculated node search value, select the node with the highest search value as the benchmark subnode, and the benchmark subnode represents the most promising search direction at present. If the benchmark subnode is not a terminal node, that is, it is not the final cargo space allocation state, then according to the transition probability matrix, randomly generate multiple extended subnodes, and the extended subnode represents the new state that may be reached after taking different decision actions under the current cargo space allocation state. Add the generated extended subnode to the global search tree as a child of the benchmark subnode, and expand the search tree in width and depth. For each extended subnode, start the simulation operation, based on the reward function, simulate the execution of the cargo space allocation strategy corresponding to the extended subnode, and evaluate its cumulative reward value. Back-propagate the cumulative reward value obtained by the simulation operation to the root node, and update the statistical information of each subnode in the global search tree. The statistical information includes the number of node visits and the average reward value. The number of visits records the number of times the subnode has been explored, and the average reward value reflects the historical performance of the subnode.
[0138] Repeat the iteration, continuously generate and evaluate child nodes, and update statistics until the preset maximum depth of the global search tree is reached. The maximum depth is used to limit the time and space complexity of the search;
[0139] When the maximum depth of the global search tree is reached, starting from the root node, all child nodes are traversed to generate corresponding action sequences. Each action sequence represents a decision path from the root node to the child node, which contains a series of cargo location allocation actions. According to the preset number of candidate sequences, the first few sequences with higher cumulative reward values in the generated action sequences are selected as candidate action sequences. These candidate action sequences represent decision paths that perform better in the global search process and will serve as the starting point for subsequent local optimization searches.
[0140] In this embodiment, by using a global search tree and a node search value formula, the search process can be effectively guided, the most promising search direction can be quickly determined, and a feasible cargo space allocation plan can be generated; by generating multiple extended child nodes according to the transition probability matrix and expanding the tree during the search process, a comprehensive exploration can be performed on the breadth and depth of the search space, and the cargo space allocation plan can be more comprehensively evaluated; by simulating each child node and back-propagating the accumulated reward value to the root node, the statistical information of each child node in the global search tree, including the number of visits and the average reward value, can be dynamically updated to reflect the historical exploration performance of each node; by comprehensively considering the number of node visits and the average reward value to calculate the node search value, the search efficiency and the quality of the generated cargo space allocation plan can be taken into account during the search process, and the search speed and search quality can be better balanced.
[0141] In an optional embodiment, the node search value is calculated based on the node search value formula, and the node with the highest node search value is selected as the reference child node, including:
[0142] The node search value formula is as follows:
[0143]
[0144] Among them, o represents the action number, P o represents the node search value of action o, represents the average reward value of action o, C(N) represents the exploration item, N represents the total number of trials, and n o represents the number of times action o is selected, C 0 represents the initial value of the exploration term, and α represents the decay rate.
[0145] The node search value is calculated based on the average reward value and exploration item of the node. The exploration item is a value that gradually decreases as the total number of trials increases. It can balance the relationship between exploration and utilization. The total number of trials is used to measure the progress of the search. The number of times an action is selected reflects the degree of exploration of the action. The initial value and decay rate of the exploration item affect the speed of change of the exploration item.
[0146] In this embodiment, the exploration term in the formula gradually decreases as the total number of trials increases, which can balance the relationship between exploration and utilization. Initially, the exploration term is large, prompting the system to explore more unknown action spaces. As the number of trials increases, the exploration term gradually decreases, and the system gradually tends to utilize known information to improve search efficiency. The node search value comprehensively considers the average reward value and the exploration term of the action, which can effectively guide the search process, give priority to actions with potential high rewards and less exploration, and find high-quality solutions more quickly.
[0147] S103. Obtain the outbound order from the WMS, generate a picking task, determine the picking location based on the outbound cargo data in the picking task, and build a picking scheduling optimization algorithm based on the current position of the carrier robot, with the goal of minimizing the idle time of the carrier robot and minimizing the picking movement time, generate the optimal picking strategy, determine the outbound control instructions, and control the carrier robot to perform the outbound action.
[0148] In this embodiment, the picking location is dynamically determined based on the current position of the carrier robot and the outbound cargo data of the picking task, so that the carrier robot can minimize the idle time during the picking task, thereby improving the transportation efficiency and warehouse throughput; the picking scheduling optimization algorithm can comprehensively consider the outbound cargo data, picking tasks, and the position of the carrier robot, and generate the optimal picking strategy, so that the path planning of the carrier robot is more reasonable when performing the outbound action, and the picking movement time is minimized, thereby improving the overall operation efficiency of the warehouse; the algorithm can dynamically generate the optimal picking strategy and outbound control instructions based on the real-time picking tasks and the position information of the carrier robot, realize real-time scheduling optimization of the warehouse outbound process, and enable the warehouse management system to respond to different outbound needs and situation changes more flexibly and intelligently.
[0149] In an optional embodiment, an outbound order is obtained from the WMS, a picking task is generated, and the picking location is determined based on the outbound cargo data in the picking task. In combination with the current position of the carrier robot, a picking scheduling optimization algorithm is constructed with the goal of minimizing the idle time of the carrier robot and minimizing the picking movement time. The generation of the optimal picking strategy includes:
[0150] Based on the outbound cargo data in the picking task, all picking locations are determined, the target point is determined, and based on the current position of the carrier robot, the starting point is determined; minimizing the idle time of the carrier robot is taken as the first goal, minimizing the picking movement time is taken as the second goal, the first goal and the second goal are dynamically weighted to determine the comprehensive objective function;
[0151] Based on the target point, starting point and comprehensive objective function, a picking scheduling optimization algorithm is constructed;
[0152] The picking scheduling optimization algorithm includes:
[0153] According to the preset number of water drops, a water drop group is randomly determined, wherein each water drop corresponds to a picking strategy, and based on the comprehensive objective function, the strategy fitness of each water drop is calculated; based on the strategy fitness and the distance between water drops in the solution space, the water drop gravity and the direction of the water drop gravity are determined, wherein the water drop gravity is inversely proportional to the distance between water drops, and the direction of the water drop gravity is from the water drop with a smaller strategy fitness to the water drop with a larger strategy fitness;
[0154] Based on the water droplet gravity and the water droplet gravity direction, updating the water droplet position in the solution space, and determining the updated water droplet;
[0155] Based on the update droplet, a neighborhood droplet is generated within a preset neighborhood exploration radius, the neighborhood strategy fitness corresponding to the neighborhood droplet and the update strategy fitness corresponding to the update droplet are calculated, and the neighborhood strategy fitness and the update strategy fitness are compared. If the neighborhood strategy fitness is greater than the update strategy fitness, the update droplet is replaced with the neighborhood droplet;
[0156] Based on the preset initial evaporation amount and the number of iterations, the evaporation amount of the water drop is calculated, and according to the evaporation amount of the water drop, the strategy fitness corresponding to the water drop is reduced. When the strategy fitness is less than the preset strategy fitness threshold, the water drop is removed from the water drop group, and the same number of supplementary water droplets as the removed water droplets are randomly generated and added to the water drop group, and the total number of water droplets in the water drop group remains unchanged;
[0157] The next round of iteration begins until the number of iterations reaches a preset upper limit of the number of iterations, and the optimal droplet corresponding to the maximum strategy fitness is extracted, and the optimal droplet corresponds to the optimal picking strategy.
[0158] The water droplet specifically refers to a possible picking strategy in the picking scheduling optimization algorithm. Each water droplet corresponds to a solution vector, representing a specific picking path and sequence. The water droplet moves and updates in the solution space, and gradually approaches the optimal solution through interaction and iterative optimization.
[0159] The water droplet gravity specifically refers to the mutual influence and attraction relationship between different picking strategies. In the algorithm, the size of the water droplet gravity is inversely proportional to the distance between the water droplets, that is, the closer the distance between the water droplets, the greater the gravity. At the same time, the direction of the water droplet gravity is from the water droplet with smaller strategy fitness to the water droplet with larger strategy fitness. Through the effect of gravity, the water droplets will move in the direction of a better picking strategy to achieve optimized search.
[0160] The evaporation amount represents the decay rate of the water drop strategy fitness during the iteration process. The algorithm introduces an evaporation mechanism to eliminate poorly performing picking strategies by gradually reducing the strategy fitness of the water drop. The evaporation amount is related to the initial evaporation amount and the number of iterations, and increases with the number of iterations. When the strategy fitness of a water drop is less than the preset threshold, the corresponding water drop will be removed from the water drop group and replaced by a newly generated water drop to maintain the total number of the water drop group unchanged.
[0161] According to the outbound cargo data in the picking task, all the cargo locations that need to be picked are determined. At the same time, the starting point of the picking path is determined according to the current position of the carrier robot, and the target point of the picking path, that is, the end point, is determined according to the position of the picking cargo location. Minimizing the idle time of the carrier robot is the first goal, and minimizing the picking movement time is the second goal. The two goals are dynamically weighted to construct a comprehensive objective function. The comprehensive objective function represents the quality of the picking strategy and is the optimization goal of the picking scheduling optimization algorithm.
[0162] According to the preset number of water drops, the initial water drop group is randomly generated. Each water drop represents a possible picking strategy and corresponds to a solution vector. The strategy fitness of each water drop is calculated, that is, the evaluation value of the picking strategy under the comprehensive objective function.
[0163] Start the calculation and iterate:
[0164] Based on the distance between the water drops and the strategy fitness, the magnitude and direction of the gravitational force between the water drops are calculated. The magnitude of the gravitational force is inversely proportional to the distance between the water drops, and the direction of the gravitational force is from the water drop with a smaller strategy fitness to the water drop with a larger strategy fitness. Based on the calculated gravitational force and direction of the water drop, the position of each water drop in the solution space is updated to obtain the updated water drop. The updated water drop represents the improvement and optimization of the picking strategy.
[0165] For each updated droplet, a neighborhood droplet is generated within the preset neighborhood exploration radius. The neighborhood droplet represents a new strategy obtained by local adjustment and optimization based on the current picking strategy. The fitness of the neighborhood strategy corresponding to the neighborhood droplet and the fitness of the update strategy corresponding to the update droplet are calculated and compared. If the fitness of the neighborhood strategy is higher, the neighborhood droplet is used to replace the update droplet.
[0166] Based on the preset initial evaporation amount and the current number of iterations, the evaporation amount of the water droplets is calculated, and the strategy fitness of each water droplet is reduced according to the evaporation amount. When the strategy fitness of a water droplet is less than a preset threshold, the water droplet is removed from the water droplet group, and the same number of supplementary water droplets as the removed water droplets are randomly generated and added to the water droplet group, so that the total number of water droplet groups remains unchanged;
[0167] Repeat multiple rounds of iterative optimization. Each round of iteration will update the water drop position, explore the neighborhood, evaporate and update the water drop. The iterative process continues until the preset upper limit of the number of iterations is reached.
[0168] When the number of iterations reaches the upper limit, the droplet with the largest strategy fitness is extracted from the final droplet group as the optimal droplet. The picking strategy corresponding to the optimal droplet is the optimal picking strategy obtained by the optimization algorithm, including the optimal picking path and picking order.
[0169] In this embodiment, by randomly generating an initial group of water droplets and iteratively updating them in the solution space, the algorithm can perform a global search in the solution space, explore different picking strategies, and effectively find the optimal picking path and picking order; in each round of iteration, the algorithm will perform a local neighborhood search on the updated water droplets to explore the local optimization space of the current picking strategy, so that the algorithm can further refine and optimize the picking strategy on the basis of the global search, and improve the quality and stability of the solution; by dynamically calculating the evaporation amount of water droplets and removing low-fitness water droplets, the algorithm can maintain the diversity and adaptability of the water droplet group, prevent the algorithm from falling into the local optimal solution, and at the same time maintain the search efficiency and robustness of the algorithm.
[0170] In an optional embodiment, minimizing the idle time of the carrier robot is the first goal, minimizing the picking movement time is the second goal, and the first goal and the second goal are dynamically weighted to determine the comprehensive objective function, including:
[0171] The first objective function f1 corresponding to the first objective:
[0172]
[0173] Where I represents the total number of picking tasks, i represents the picking task index, J represents the total number of carrier robots, j represents the carrier robot index, and c ij represents the cost of moving from the starting point to the target point, x ij Indicates whether to assign picking task i to carrier robot j, y ij indicates whether the target point j has been selected for picking task i, λ represents the penalty coefficient, si represents the number of goods for picking task i, and A j represents the maximum load capacity of the carrier robot j;
[0174] The second objective function f2 corresponding to the second objective is:
[0175]
[0176] Among them, dijk represents the time when the carrier robot j moves to the picking cargo location k when performing the picking task i, β represents the height penalty coefficient, hk represents the height of the picking cargo location k, and zijk represents whether the carrier robot j chooses the picking cargo location k for picking when performing the picking task i;
[0177] The comprehensive objective function F is formulated as follows:
[0178] F = w1 × f1 + w2 × f2;
[0179] Among them, ω1 represents the weight of the first objective function, and ω2 represents the weight of the second objective function.
[0180] The first objective function is to minimize the idle time of the carrier robot, which consists of two parts. The first part represents the cost of moving from the picking location to the starting point of the carrier robot, and takes into account whether the picking task has been assigned to the carrier robot and whether the target point has been selected for the picking task; the second part represents that for each carrier robot, if the picking goods it carries exceed the maximum load capacity, it needs to be punished;
[0181] The second objective function aims to minimize the picking movement time, taking into account the time it takes for each carrier robot to move to the picking location when performing the picking task, and combining the height information of the picking location and whether the location is selected for picking. Among them, β represents the height penalty coefficient, which is used to control the degree of penalty when considering the height difference of the picking location.
[0182] According to the formula, the first objective function is to minimize the idle time of the carrier robot, taking into account the cost of task allocation and the constraints of the quantity of goods; the second objective function is to minimize the picking movement time, taking into account the movement cost and height penalty of the carrier robot in performing the picking task; by weighted combination of the two objective functions, minimizing the idle time of the carrier robot and minimizing the picking movement time are taken as two goals, and a comprehensive objective function is dynamically weighted to construct a comprehensive objective function, which comprehensively considers the trade-off between the idle time and the picking movement time, so that the algorithm can optimize the two goals at the same time and achieve a balanced picking scheduling result.
[0183] In an optional embodiment, it also includes:
[0184] The water drop gravity is as follows:
[0185]
[0186] Among them, H pq represents the gravitational force between water droplets p and q, G represents the preset gravitational constant, and F p represents the strategic fitness of the comprehensive objective function calculation of droplet p, F q represents the strategic fitness of the comprehensive objective function calculation of the water droplet q, r pq represents the distance between water drop p and water drop q, ε represents the parameter to avoid division by zero, and δ represents the water drop gravity attenuation factor;
[0187] The formula for calculating the evaporation of water droplets is as follows:
[0188]
[0189] Among them, E(t) represents the function of the evaporation of water droplets changing with the number of iterations, E 0represents the initial evaporation amount, t represents the number of iterations, T represents the upper limit of the number of iterations, θ represents the evaporation rate control parameter, θ>0, μ represents the evaporation periodicity parameter, μ>0.
[0190] The water drop gravity formula describes the interaction force between two water droplets. The water drop gravity is inversely proportional to the distance between the two water droplets, that is, the closer the distance, the greater the gravity. At the same time, the magnitude of gravity is also affected by the difference between the comprehensive objective function values of the two water droplets and the distance. The greater the difference, the stronger the gravity, and the farther the distance, the weaker the gravity. The formula takes into account the mutual attraction and mutual repulsion between water droplets, and realizes the self-organized clustering between water droplets and the guidance of search direction by adjusting the gravity factor and attenuation factor.
[0191] The formula for water drop evaporation describes the gradual evaporation of water droplets during the iteration process. As the number of iterations increases, the evaporation of water droplets gradually increases, but the rate of increase will slow down over time until the iteration ends. The process is affected by the initial evaporation, evaporation rate parameters, and periodic parameters. By adjusting the parameters, the rate and law of water drop evaporation can be controlled, affecting the convergence speed of the algorithm and the degree of exploration of the search space.
[0192] According to the formula, the water droplet gravity formula forces water droplets to approach each other in the search space to form clusters, which helps to accelerate the search process and focus on the possible solution space; the water droplet evaporation formula introduces an evaporation mechanism that decreases over time, so that the algorithm can strike a balance between global search and local optimization. As the number of iterations increases, the increase in water droplet evaporation will slow down, allowing the algorithm to maintain a certain global search capability while converging to the local optimal solution; by adjusting the evaporation rate parameter in the water droplet evaporation formula, the convergence speed of the algorithm can be controlled. A larger evaporation rate parameter will speed up the evaporation of water droplets, thereby accelerating the convergence speed of the algorithm; conversely, a smaller evaporation rate parameter will slow down the evaporation of water droplets, which helps to explore the search space more deeply.
[0193] Figure 2 FIG. 1 is a schematic diagram of the structure of an intelligent logistics warehousing system based on WMS according to an embodiment of the present invention. Figure 2 As shown, the system comprises:
[0194] The first unit is used to obtain image information and weight information of incoming goods through an image acquisition device and a weighing device, and determine the comprehensive data of the incoming goods based on the image information and the weight information through a goods recognition algorithm and in combination with the goods feature data in the WMS;
[0195] The second unit is used to construct a three-dimensional warehouse grid model based on the warehouse data obtained from the WMS, determine the decision constraints, and build a cargo location decision model in combination with the comprehensive data of the incoming goods, determine the optimal incoming cargo location by searching for the optimal cargo location allocation strategy, and store the incoming goods in the optimal incoming cargo location;
[0196] The third unit is used to obtain the outbound order from the WMS, generate a picking task, determine the picking location based on the outbound cargo data in the picking task, and build a picking scheduling optimization algorithm based on the current position of the carrier robot to minimize the idle time of the carrier robot and minimize the picking movement time, generate the optimal picking strategy, determine the outbound control instructions, and control the carrier robot to perform the outbound action.
[0197] According to a third aspect of the embodiments of the present invention,
[0198] An electronic device is provided, comprising:
[0199] processor;
[0200] a memory for storing processor-executable instructions;
[0201] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.
[0202] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.
[0203] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. Intelligent logistics warehousing method based on WMS, It is characterized in that include: Acquire image information and weight information of incoming goods through image acquisition equipment and weighing equipment, and determine comprehensive goods data of the incoming goods based on the image information and weight information and through a goods recognition algorithm in combination with goods feature data in the WMS; Based on the warehouse data obtained from the WMS, a three-dimensional warehouse grid model is constructed to determine the decision constraints. Combined with the comprehensive data of the incoming goods, a cargo location decision model is constructed. By searching for the optimal cargo location allocation strategy, the optimal incoming cargo location is determined, and the incoming goods are stored in the optimal incoming cargo location. Obtain outbound orders from WMS, generate picking tasks, determine picking locations based on outbound cargo data in the picking tasks, and build a picking scheduling optimization algorithm based on the current position of the carrier robot, with the goal of minimizing the idle time of the carrier robot and minimizing the picking movement time. Generate the optimal picking strategy, determine outbound control instructions, and control the carrier robot to perform outbound actions.
2. The method according to claim 1, It is characterized in that The image information and weight information of the incoming goods are obtained by the image acquisition device and the weighing device. Based on the image information and the weight information, the goods identification algorithm is used in combination with the goods feature data in the WMS to determine the comprehensive goods data of the incoming goods, including: graying, denoising and image enhancement preprocessing are performed on the image information to determine a standard cargo image; Based on a preset feature point detection threshold and a preset scale space parameter, feature extraction is performed on the standard cargo image to obtain a set of initial feature points and corresponding scale information, and a corresponding feature descriptor is calculated; Extracting local histogram features and local gradient features from the local area around each of the initial feature points, determining local image features, calculating local complexity, and combining the feature density corresponding to the initial feature points, updating the feature point detection threshold and scale space parameters, combining the local image features with the initial feature points, and recalculating feature descriptors; Iterate in sequence until a preset number of iterations is reached, determine valid feature points according to the feature descriptor and a preset descriptor threshold, and extract the features of the cargo image; Matching the cargo image features with the cargo feature data in the WMS, calculating the matching degree, selecting the cargo with the highest matching degree as the matching result, and obtaining the corresponding cargo coding information; The weight information is compared and verified with the standard weight in the cargo coding information. When the absolute value of the difference between the weight information and the standard weight is less than a preset weight error threshold, the comprehensive cargo data is confirmed.
3. The method according to claim 1, It is characterized in that Based on the warehouse data obtained from the WMS, a three-dimensional warehouse grid model is constructed, decision constraints are determined, and a cargo location decision model is constructed in combination with the comprehensive data of the incoming goods. The optimal incoming cargo location is determined by searching for the optimal cargo location allocation strategy, and the incoming goods are stored in the optimal incoming cargo location, including: Based on the warehouse data obtained from the WMS, the cargo location layout, cargo location attributes and cargo location constraints are extracted, and based on the cargo location layout, a three-dimensional grid is determined. In combination with the cargo location attributes and the cargo location constraints, the warehouse is abstracted into a warehouse three-dimensional grid model, the distance matrix between cargo locations is determined, and decision constraints are constructed; Based on the location and quantity of the goods, the corresponding cargo space allocation is determined, and the state space is constructed. According to the allocation action of allocating the incoming goods to the corresponding cargo spaces, the action space is constructed. According to the cargo space attributes and decision constraints, the transition probability matrix between the states in the state space is determined. According to the cargo space allocation efficiency and the cargo space utilization rate, a reward function is generated, and the evaluation quality of each state in the state space is determined. With the goal of maximizing the cumulative reward of the evaluation quality, a cargo space decision model is constructed; Use a multi-level search algorithm to solve the cargo location decision model, determine the candidate action sequence through global search, perform local optimization search on the candidate action sequence, determine multiple groups of cargo location allocation strategies, calculate the cumulative rewards of the cargo location allocation strategies, select the cargo location allocation strategy corresponding to the maximum value of the cumulative reward, and determine the optimal storage cargo location.
4. The method according to claim 3, It is characterized in that Determining candidate action sequences through global search includes: Determine a root node of a global search tree based on a current state in the state space of the warehouse; Starting from the root node, perform the iteration operation: Calculate the node search value based on the node search value formula, and select the node with the highest node search value as the benchmark child node; If the reference child node is not a terminal node, randomly generating a plurality of extended child nodes based on the transition probability matrix, and adding the extended child nodes to the global search tree; Based on each extended child node, starting from the corresponding extended child node, based on the reward function, a simulation operation is performed to evaluate the cumulative reward value of each extended child node; Back-propagating the accumulated reward value obtained by the simulation operation to the root node, and updating the statistical information of each child node in the global search tree, wherein the statistical information includes the number of visits and the average reward value; Repeatedly generate child nodes and update the statistical information until the preset maximum depth of the global search tree is reached, start from the root node, traverse all child nodes to generate corresponding action sequences, and determine the candidate action sequence according to the preset number of candidate sequences.
5. The method according to claim 4, It is characterized in that Based on the node search value formula, the node search value is calculated, and the node with the highest node search value is selected as the benchmark child node including: The node search value formula is as follows: Among them, o represents the action number, P o represents the node search value of action o, represents the average reward value of action o, C(N) represents the exploration item, N represents the total number of trials, and n o represents the number of times action o is selected, C 0 represents the initial value of the exploration term, and α represents the decay rate.
6. The method according to claim 1, It is characterized in that Obtain outbound orders from WMS, generate picking tasks, determine picking locations based on outbound cargo data in the picking tasks, and build a picking scheduling optimization algorithm based on the current position of the carrier robot with the goal of minimizing the idle time of the carrier robot and minimizing the picking movement time. Generate the optimal picking strategy, including: Based on the outbound cargo data in the picking task, all picking locations are determined, the target point is determined, and based on the current position of the carrier robot, the starting point is determined; minimizing the idle time of the carrier robot is taken as the first goal, minimizing the picking movement time is taken as the second goal, the first goal and the second goal are dynamically weighted to determine the comprehensive objective function; Based on the target point, starting point and comprehensive objective function, a picking scheduling optimization algorithm is constructed; The picking scheduling optimization algorithm includes: According to the preset number of water drops, a water drop group is randomly determined, wherein each water drop corresponds to a picking strategy, and based on the comprehensive objective function, the strategy fitness of each water drop is calculated; based on the strategy fitness and the distance between water drops in the solution space, the water drop gravity and the direction of the water drop gravity are determined, wherein the water drop gravity is inversely proportional to the distance between water drops, and the direction of the water drop gravity is from the water drop with a smaller strategy fitness to the water drop with a larger strategy fitness; Based on the water droplet gravity and the water droplet gravity direction, updating the water droplet position in the solution space, and determining the updated water droplet; Based on the update droplet, a neighborhood droplet is generated within a preset neighborhood exploration radius, the neighborhood strategy fitness corresponding to the neighborhood droplet and the update strategy fitness corresponding to the update droplet are calculated, and the neighborhood strategy fitness and the update strategy fitness are compared. If the neighborhood strategy fitness is greater than the update strategy fitness, the update droplet is replaced with the neighborhood droplet; Based on the preset initial evaporation amount and the number of iterations, the evaporation amount of the water drop is calculated, and according to the evaporation amount of the water drop, the strategy fitness corresponding to the water drop is reduced. When the strategy fitness is less than the preset strategy fitness threshold, the water drop is removed from the water drop group, and the same number of supplementary water droplets as the removed water droplets are randomly generated and added to the water drop group, and the total number of water droplets in the water drop group remains unchanged; The next round of iteration begins until the number of iterations reaches a preset upper limit of the number of iterations, and the optimal droplet corresponding to the maximum strategy fitness is extracted, and the optimal droplet corresponds to the optimal picking strategy.
7. The method according to claim 6, It is characterized in that Taking minimizing the idle time of the transport robot as the first goal and minimizing the picking movement time as the second goal, the first goal and the second goal are dynamically weighted to determine the comprehensive objective function including: The first objective function f1 corresponding to the first objective: Where I represents the total number of picking tasks, i represents the picking task index, J represents the total number of carrier robots, j represents the carrier robot index, and c ij represents the cost of moving from the starting point to the target point, x ij Indicates whether to assign picking task i to carrier robot j, y ij Indicates whether the target point j has been selected for picking task i, λ represents the penalty coefficient, s i represents the quantity of goods for picking task i, A j represents the maximum load capacity of the carrier robot j; The second objective function f2 corresponding to the second objective is: Among them, d ijk represents the time it takes for the carrier robot j to move to the picking location k when performing the picking task i, β represents the height penalty coefficient, and h k represents the height of picking location k, z ijk Indicates whether the carrier robot j chooses the picking location k to pick when performing the picking task i; The comprehensive objective function F is formulated as follows: F=ω 1 ×f 1 +oh 2 ×f 2 ; Among them, ω1 represents the weight of the first objective function, and ω2 represents the weight of the second objective function.
8. The method according to claim 7, It is characterized in that Also includes: The water drop gravity is as follows: Among them, H pq represents the gravitational force between water droplets p and q, G represents the preset gravitational constant, and F p represents the strategic fitness of the comprehensive objective function calculation of droplet p, F q represents the strategic fitness of the comprehensive objective function calculation of the water droplet q, r pq represents the distance between water drop p and water drop q, ε represents the parameter to avoid division by zero, and δ represents the water drop gravity attenuation factor; The formula for calculating the evaporation of water droplets is as follows: Among them, E(t) represents the function of the evaporation of water droplets changing with the number of iterations, E 0 represents the initial evaporation amount, t represents the number of iterations, T represents the upper limit of the number of iterations, θ represents the evaporation rate control parameter, θ>0, μ represents the evaporation periodicity parameter, μ>0.
9. An intelligent logistics warehousing system based on WMS, used to implement the method described in any one of claims 1 to 8. It is characterized in that include: The first unit is used to obtain image information and weight information of incoming goods through an image acquisition device and a weighing device, and determine the comprehensive data of the incoming goods based on the image information and the weight information through a goods recognition algorithm and in combination with the goods feature data in the WMS; The second unit is used to construct a three-dimensional warehouse grid model based on the warehouse data obtained from the WMS, determine the decision constraints, and build a cargo location decision model in combination with the comprehensive data of the incoming goods, determine the optimal incoming cargo location by searching for the optimal cargo location allocation strategy, and store the incoming goods in the optimal incoming cargo location; The third unit is used to obtain the outbound order from the WMS, generate a picking task, determine the picking location based on the outbound cargo data in the picking task, and build a picking scheduling optimization algorithm based on the current position of the carrier robot to minimize the idle time of the carrier robot and minimize the picking movement time, generate the optimal picking strategy, determine the outbound control instructions, and control the carrier robot to perform the outbound action.
10. An electronic device, It is characterized in that include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 8.
Citation Information
Cited By
Control method and system for automatic feeding and assembling of plywood
CN119568673A
Intelligent WMS system with intelligent scheduling function
CN119850027A
Book packaging method and device based on readable fields, server and medium
CN120383115A
Three-dimensional storage goods allocation optimal distribution method based on graph neural network
CN121073347A
Inventory logistics management method and system based on Internet of Things
CN121599584A
