Storing and picking optimization group intelligent management and control method and device based on neural network algorithm

Abstract

The invention discloses a storage and sorting optimization group intelligent management and control method and device based on a neural network algorithm, and the method comprises the steps: obtaining data related to storage and sorting tasks, and carrying out the data preprocessing; mapping order demands and cargo position data, and constructing a corresponding traveling salesman TSP problem instance to describe a path planning problem in a warehouse storage and picking task; influence factors based on information entropy are introduced, an Attention mechanism and a BP neural network are combined to carry out feature extraction on traveling salesman TSP problem instance data, a multi-agent cooperation framework based on a neural network algorithm is constructed, an internal relation between data features is shared based on a particle swarm algorithm to optimize a storage and picking path, an optimal solution is achieved, and the storage and picking efficiency is improved. And obtaining a storing and picking path scheme. According to the method, a group intelligent management and control method can be realized, data are shared by a plurality of agents, and an accurate and efficient result is achieved in storage and sorting tasks of a warehouse.

Description

Technical Field

[0001] The present invention relates to the field of warehousing and logistics technology, and specifically to a storage and sorting optimization group intelligent management and control method and device based on a neural network algorithm, which combines the neural network algorithm with the group intelligent management and optimization technology to improve the efficiency and accuracy of warehouse storage and sorting operations through intelligent means. Background Art

[0002] With the rapid development of e-commerce, the diversification of consumer demand and the sharp increase in order volume, the warehousing and logistics industry is facing unprecedented severe challenges. In this context, how to build an efficient, intelligent and accurate warehousing management system to achieve rapid storage, accurate sorting and safe distribution of large-scale goods has become a core issue that the entire industry needs to solve. Traditional manual or simple automated management methods can no longer meet the needs of modern warehousing and logistics. There is an urgent need for a management and control method that can automatically learn, make intelligent decisions and optimize storage and picking paths. Existing methods have the following problems: (1) Efficiency bottleneck. High reliance on manual labor: Many warehousing and logistics centers still rely heavily on manual labor to store and sort goods, which is particularly inefficient when processing massive orders. Manual operations are not only slow, but also prone to fatigue, which leads to increased error rates. Low penetration rate of automated equipment: Although automation technology can significantly improve efficiency, its high cost limits its popularity in small and medium-sized warehouses. Even in large warehouses, efficiency improvements may be limited due to untimely equipment updates or unreasonable configurations.

[0003] (2) Low accuracy. Human error: During the manual sorting process, due to lack of concentration, unfamiliarity with the goods, etc., it is easy to misplace, miss or send the wrong goods. These problems not only affect the customer experience, but may also lead to subsequent problems such as returns and complaints. Technical limitations: Although technologies such as RFID and barcodes can improve recognition accuracy, in actual applications, recognition failures or errors may occur due to reasons such as label damage, reading device failure or signal interference.

[0004] (3) Poor flexibility. Fixed layout: Traditional warehouse layouts are often fixed and difficult to flexibly adjust according to changes in market demand and product characteristics. This leads to low warehouse space utilization and difficulty in meeting rapidly changing order needs. Rigid processes: Sorting processes are often fixed and cumbersome, making it difficult to quickly respond to emergencies or special needs. For example, during peak order periods or product promotions, fixed sorting processes may cause congestion and delays.

[0005] (4) Lack of intelligent decision-making support. Insufficient data utilization: Many warehousing and logistics centers do not do enough in data collection, or even if they collect data, they fail to use it effectively. This leads to a lack of scientific basis and accurate decision-making support in inventory management, demand forecasting, etc. Delayed decision-making process: Due to the lack of support from intelligent decision-making systems, warehousing and logistics centers are often slow to respond to market changes. For example, when a product is hot, it may not be replenished in time, and when a product is unsalable, it may have too much inventory backlog.

[0006] In summary, the existing methods of large-scale commodity storage and sorting have significant shortcomings such as efficiency bottlenecks, insufficient accuracy, high costs, poor flexibility, poor environmental adaptability, and lack of intelligent decision support when facing the challenges brought by the rapid development of e-commerce. These shortcomings not only affect the overall operational efficiency and service quality of the warehousing and logistics industry, but also restrict its further development and innovation. Summary of the invention

[0007] Aiming at the problems of low efficiency, poor flexibility and few intelligent decision-making in existing warehouse picking tasks, the present invention proposes an intelligent management and control method and device for picking optimization groups based on neural network algorithm, aiming to improve the efficiency of warehouse picking tasks, optimization of picking paths and intelligent coverage.

[0008] To achieve the above objectives, the present invention adopts the following technical solutions: In the first aspect, the present invention provides an intelligent management and control method for a storage and picking optimization group based on a neural network algorithm, the method comprising the following steps: (1) Collect warehouse data, obtain data related to storage and sorting tasks, and perform data preprocessing; (2) Map order demand and cargo location data and construct a corresponding traveling salesman problem instance to describe the path planning problem in warehouse picking tasks; (3) The influencing factor based on information entropy is introduced and combined with the Attention mechanism and BP neural network to extract features of the traveling salesman problem instance data, and the optimization solution is performed based on the particle swarm algorithm to obtain the storage and picking path solution.

[0009] Furthermore, the data related to the storage and sorting tasks include storage and sorting path data, warehouse layout and cargo data, historical data and environmental parameter data.

[0010] Furthermore, data preprocessing specifically includes using isolation forests to detect and process outliers, performing data cleaning and normalization, and converting the data into a format suitable for neural network processing.

[0011] Furthermore, the specific steps for constructing a traveling salesman TSP problem instance are as follows: (2.1) Number or mark each storage location in the warehouse; (2.2) Calculate the moving distance or cost between storage nodes, which serves as the "weight" or "edge" in the traveling salesman problem (TSP).

[0012] (2.3) Based on the obtained “weights” or “edges”, a graph is constructed, where each node represents a storage location and each edge represents the cost of moving between two storage locations.

[0013] (2.4) Set the starting and ending points of the TSP problem according to the actual needs of warehouse operations; (2.5) Integrate the nodes, edges, starting points, and ending points obtained in the above steps into a complete TSP problem instance.

[0014] Furthermore, based on the traveling salesman TSP problem example, the path planning problem in the warehouse picking task is described, that is, how to traverse all storage nodes once and only once with the minimum total cost or total moving distance, and finally return to the starting point or the end point.

[0015] Furthermore, the specific process of feature extraction in step (3) is as follows: (1) Preliminary data processing and feature extraction based on BP neural network; (2) Introduce an impact factor based on information entropy to count the frequency of occurrence of features in each dimension, assign an external parameter to each feature to reflect the importance of the feature, calculate the information entropy of the features in each dimension to evaluate the amount of information it contains and its importance in the overall data, and assign an impact factor to each feature based on information entropy for subsequent feature weighting processing; (3) Fusion of the Attention mechanism layer and the BP neural network. Apply the weights generated by the Attention mechanism layer to the output of the previous layer to obtain the weighted feature representation; (4) Train the model and adjust the weights and biases in the network through the back-propagation algorithm to minimize the output error. The trained network outputs a feature vector weighted by the Attention mechanism to guide the optimization of the picking path.

[0016] Furthermore, the number of occurrences of a certain dimension data on all shelves in the warehouse is defined as the global influence of the dimension data, and its influence factor is expressed by the formula:

[0017] Among them I G represents the global impact factor, n represents the total number of warehouse shelves, i represents the data of the i-th dimension, H iIndicates the number of times the data of the i-th dimension appears on all shelves. Apply the influencing factor of each dimension to the corresponding feature, and the formula is as follows:

[0018] in is the feature of the i-th dimension, represents the global impact factor of the feature, Indicates features to which global influence information is added.

[0019] Furthermore, based on the particle swarm algorithm, the optimization solution is obtained, and the process of obtaining the storage and picking path solution is as follows: (1) Setting the parameters of the particle swarm algorithm, including: the size of the particle swarm, that is, the number of agents, the initial position of each particle, representing a possible picking path solution, the initial speed, representing the direction and amplitude of the path adjustment, as well as the maximum number of iterations, inertia weight, individual learning factor, and social learning factor; (2) Randomly generate the initial position and velocity of each particle in the solution space; the position of each particle represents the sequence of the picking order, and the velocity reflects the direction and intensity of the sequence adjustment; for each particle, calculate its corresponding fitness value according to the path plan it represents; in the picking optimization problem, the fitness value is calculated based on the total length of the path, time cost or energy consumption index, and the goal is to minimize these indicators; (3) After obtaining the optimal solution, the particle speed is updated according to the particle’s current speed, the difference between the individual optimal solution and the current position, the difference between the global optimal solution and the current position, and the inertia weight; the particle’s position is adjusted according to the updated speed, that is, the storage and picking path plan is adjusted.

[0020] In a second aspect, the present invention also provides an intelligent management and control device for a storage and picking optimization group based on a neural network algorithm, comprising a memory and one or more processors, wherein the memory stores executable code, and when the processor executes the executable code, the intelligent management and control method for a storage and picking optimization group based on a neural network algorithm is implemented.

[0021] In a third aspect, the present invention further provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the intelligent management and control method of a storage and picking optimization group based on a neural network algorithm.

[0022] Beneficial effects of the present invention: The present invention combines a neural network with a particle swarm algorithm, and can fully utilize the advantages of both in feature extraction and optimization search. The neural network is responsible for extracting useful feature information from complex data, while the particle swarm algorithm performs efficient path search and optimization based on this feature information. This combination not only ensures the accuracy of feature extraction, but also improves the efficiency of path optimization. In addition, combined with the influencing factor mechanism, the frequency of occurrence of features in each dimension is counted, and an external parameter is assigned to each feature to reflect the importance of the feature in the overall data. This not only retains the information of the original features, but also emphasizes the role of important features. Finally, these weighted features are used as model inputs for further processing and analysis to extract key features that are useful for warehouse management decisions, providing strong support for optimizing storage and picking paths and improving warehouse efficiency.

[0023] In warehouse storage and picking operations, factors such as cargo location and picking tasks may change over time. The present invention can adjust feature extraction and path optimization strategies in real time according to the latest data to adapt to the dynamic changes in the warehouse environment. This real-time and dynamic adaptability is the key to improving warehouse operation efficiency and response speed; it can effectively improve the efficiency of warehouse storage and picking tasks, save storage and picking time, and lay the foundation for intelligence. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0025] Figure 1 It is a flow chart of the intelligent management and control method of storage and picking optimization group based on neural network algorithm of the present invention.

[0026] Figure 2 It is a schematic diagram of the storage and sorting conversion TSP of the present invention.

[0027] Figure 3 It is a schematic diagram of feature extraction of the present invention.

[0028] Figure 4 It is a schematic diagram of the Attention mechanism of the present invention.

[0029] Figure 5 It is a schematic diagram of the BPNN structure of the present invention.

[0030] Figure 6 It is the flow chart of the particle swarm algorithm of the present invention.

[0031] Figure 7It is a structural diagram of an intelligent management and control device for storage and picking optimization groups based on a neural network algorithm of the present invention. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical solutions and advantages of the present invention more intuitive and clear, the present invention is further described in detail below with reference to the accompanying drawings.

[0033] Feature extraction and particle swarm optimization of storage and picking tasks are the key technologies in this method. The purpose of this invention is to automatically extract key features from warehouse operation data and construct efficient feature engineering by utilizing advanced machine learning techniques, especially neural network algorithms. By constructing a multi-agent collaboration framework based on a neural network algorithm and combining the particle swarm algorithm to deeply explore the intrinsic connections between data features, the intelligent optimization of the storage and picking path is achieved to achieve the goal of the shortest path and the highest efficiency. This method aims to solve the problems of unreasonable path planning and low efficiency in traditional warehouse management. Through group intelligent management and control, data sharing and collaborative work between multiple agents are realized to ensure accurate and efficient results in warehouse storage and sorting tasks, thereby improving the overall operational efficiency and customer satisfaction of the warehouse.

[0034] This embodiment provides a method for intelligent management and control of storage and picking optimization groups based on a neural network algorithm. Figure 1 As shown, the specific steps include: (1) Warehouse data collection steps: Use infrared sensors, ultrasonic sensors, and environmental sensors to monitor the movement and location changes of goods and the internal environment data of the warehouse in real time. At the same time, RFID tags and readers can also be used to attach RFID tags to each product or shelf, and RFID readers can be arranged in the warehouse. When the goods pass through the reader, the tag information is automatically read to quickly and accurately obtain the location, type, quantity and other information of the goods. Use the Internet of Things (IoT) platform to connect various sensors, RFID readers, cameras and other equipment in the warehouse, collect environmental data such as warehouse layout and equipment status, and store historical data, and realize unified data management and analysis based on the warehouse management system (WMS). Warehouse data collection is a key step in the entire method. It is crucial to build a real-time location map of goods and track the status of goods. Environmental sensors such as temperature, humidity, and light are used to monitor the internal environmental conditions of the warehouse to ensure that the storage conditions meet the requirements of the goods. These data can be used to optimize the storage layout and improve the quality of goods preservation. Weight sensors are used on shelves or handling equipment to measure the weight of goods to help determine the handling path and select appropriate handling tools. After collecting real-time data, the Internet of Things (IoT) platform is used to connect various sensor devices in the warehouse to achieve unified data management and analysis. Its powerful data processing and analysis capabilities help to extract valuable information from massive data. All data related to storage and sorting tasks in the warehouse include but are not limited to: Storage and picking path data: collects storage location information, picking path, picking time, etc. of all goods in the warehouse.

[0035] Other warehouse data: including warehouse layout, shelf location, cargo weight, volume, picking frequency, etc.

[0036] Historical data: past storage and sorting records, operation time, error rate, etc., for subsequent data analysis and optimization.

[0037] Environmental parameters: environmental conditions such as temperature, humidity, light, etc. in the warehouse, as well as other factors that may affect work efficiency.

[0038] By acquiring high-quality data, a reliable foundation is provided for subsequent warehouse storage and sorting tasks.

[0039] (2) Warehouse data preprocessing step: This step aims to clean and normalize the collected raw warehouse data to eliminate noise, fill in missing values, and convert the data into a format suitable for neural network processing. In warehouse data, outliers are often significantly different from normal data, which means they are more easily isolated in the data space. Specifically, during the data cleaning stage, check the outliers and duplicate values ​​in the warehouse data. Due to the influence of the environment and other factors that affect work efficiency, there will be outliers due to human error causing the picking path to be too long or multiple trips on one path. These noise data will affect the accuracy of the model prediction; This step has a great impact on the accuracy of subsequent feature extraction, model training and final results. Therefore, the present invention uses Isolation Forest (iForest) to effectively detect and remove outliers. The Isolation Forest method simulates the isolation process of data by constructing multiple isolated trees and calculates the path length of each data point. Data points with shorter path lengths are considered outliers. This method is particularly effective for processing high-dimensional data and complex distributed data, and can more accurately identify outliers in warehouse data. It isolates outliers based on a tree structure. In the data preprocessing stage, the use of isolation forests can help identify and remove those points that are obviously deviated from the normal data distribution, i.e., outliers, and then take measures to delete, replace or correct them. At the same time, there will be duplicate recorded data. After data cleaning, the data is converted and uniformly input. For string type data, it is converted into numerical data, and according to the data distribution, for data of different dimensions, the Min-Max normalization method is used for standardization, so that the data falls on the same scale to align the data, which is convenient for neural network processing. This process can be implemented using the IsolationForest method in scikit-learn. In addition, for data of different dimensions, the normalizer MinMaxScaler in sklearn.preprocessing can be used to align the data and map the data of different dimensions to the interval range of [0, 1]. Assume that h(x) is the path length of sample x in the isolated tree, is the average path length (for a given number of samples and trees), then the anomaly score The formula can be defined as:

[0040] Where n is the total number of samples, c(n) is the average path length for a given n, and is a constant. For a given n, c(n) = 2 / ln(n-1) + γ - 2 / ln(2) (where γ is Euler's constant, approximately equal to 0.57722).

[0041] Since neural networks are sensitive to the scale of input data, the data needs to be normalized after processing outliers, scaling the feature values ​​to a small specific interval (usually [0, 1]) to reduce the impact of dimensional differences between different features on model training. For each feature X, find its maximum value max(X) and minimum value min(X). Apply the normalization formula to each feature value x:

[0042] where x is the original eigenvalue, is the normalized eigenvalue.

[0043] (3) TSP (Traveling Salesman) Problem Transformation Steps: Figure 2 As shown in Figure 1, this step abstracts the storage and sorting tasks of the warehouse into a TSP problem (traveling salesman problem), which includes determining the location of each storage / sorting area and mapping them to cities in the TSP problem. Convert "the traveling salesman needs to find the shortest path to visit each city once and return to the starting point" to "find the shortest path to traverse all storage / sorting points and return to the starting point". Calculate the distance between storage / sorting points, build a distance matrix, find the shortest path, so that all designated cargo points are visited from the starting point and return to the starting point, the path length is the shortest, and use the optimization algorithm to solve the optimal storage and picking path. This usually involves mapping order requirements with cargo locations and constructing corresponding TSP problem instances. In the TSP problem, each "city" represents a storage location or sorting point in the warehouse. Set the specific parameters of the TSP problem according to the actual operational needs of the warehouse. For example, set the entrance or specific area of ​​the warehouse, and whether to allow a longer stay at the sorting point. The specific contents are as follows: 1) Determine the task requirements and constraints. First, clarify the warehouse picking task, including the cargo points that need to be visited (such as shelves, storage locations), the required quantity at each point (for sorting tasks), the starting point and the end point (usually the starting point and the end point are the same, such as the entrance / exit of the warehouse). Constraints: Consider the physical limitations within the warehouse, such as aisle width, the movement speed of cargo handling equipment, the safe distance between goods, etc., which will affect the choice of path.

[0044] 2) Construct a set of cargo points. Consider each cargo point that needs to be visited as a city in the TSP problem, and construct a set containing all cargo points. Each cargo point has a unique identifier (such as coordinates, shelf number, etc.) and corresponding demand information (for sorting tasks).

[0045] 3) Calculate the distance matrix. Based on the actual location of the cargo points in the warehouse, calculate the distance or cost of movement between these nodes (storage locations). This can be done by measuring the straight-line distance between two points (in two-dimensional or three-dimensional space) or using a more complex path planning algorithm (taking into account obstacles and passages in the warehouse). These distances or costs serve as "weights" or "edges" in the TSP problem. In the graph constructed in this way, each node represents a storage location and each edge represents the cost of movement between two storage locations.

[0046] 4) Set the parameters of the TSP problem. The objective function of the TSP problem is usually to find the shortest path that starts from the starting point, visits each cargo point exactly once, and finally returns to the starting point, with the shortest total length of the path. Set other parameters of the TSP problem, such as the order of visiting cargo points (if there are special requirements), the starting point and end point of the path, etc.

[0047] 5) Conversion and verification. Convert the warehouse picking task into an instance of the TSP problem, including a set of cargo points, a distance matrix, and TSP problem parameters. This example describes the path planning problem in the warehouse picking task, that is, how to traverse all storage locations (nodes) once and only once with the minimum total cost (total moving distance) and finally return to the starting point or the end point. Verify whether the converted TSP problem instance meets the requirements and constraints of the actual picking task.

[0048] According to the layout of the warehouse, calculate the distance or cost between any two nodes (i.e., key locations). This distance can be physical distance, time cost, energy consumption, or other metrics. The constructed distance matrix D is an n×n matrix, where n represents the number of samples and D[i][j] represents the cost from node i to node j. The storage and sorting tasks of the warehouse are modeled as a TSP problem, that is, finding the shortest path that traverses all nodes (key locations) exactly once and returns to the starting point. This path represents the optimal storage and picking sequence, which can minimize the total cost (such as total walking distance, total time, etc.). After the transformation, the TSP problem can be formalized as a minimization problem, the goal of which is to minimize the total cost C, and the formula is:

[0049] here, is a sequence of length n, representing the order in which nodes are visited (i.e., the path). Represents the i-th node in the path.

[0050] 6) Prepare input data. The transformed TSP problem instance (mainly the distance matrix and the possible order requirements for cargo point visits) is used as input data to prepare for the subsequent feature extraction step.

[0051] (4) Feature extraction steps: Figure 3 As shown in FIG. 1 , in this step, the present invention uses an influence factor based on information entropy and combines the Attention mechanism with the BP neural network to extract key features from the preprocessed data to enhance the model's ability to capture key information. Information entropy is an indicator used to measure the amount of information contained in a message. In the field of information, the higher the information entropy, the more information it contains. In the first step, as Figure 5 As shown in the figure, according to the task requirements and data characteristics, a BP neural network structure including an input layer, several hidden layers containing a certain number of neurons and an output layer is designed and constructed for preliminary data processing and feature extraction. Through multi-layer nonlinear transformation, the BP neural network can automatically learn the complex feature representation in the data, and introduce nonlinearity through activation functions (such as ReLU, Sigmoid, etc.), thereby enhancing the network's expression ability; the hidden layer extracts high-level features in the data through nonlinear transformation, and the output layer outputs the corresponding results according to the task objectives (such as predicting the optimal path, evaluating storage efficiency, etc.). In the second step, the impact factor based on information entropy is introduced to calculate the information entropy of the features of each dimension to evaluate the amount of information and importance it contains. Based on information entropy, an impact factor is assigned to each feature for subsequent feature weighting processing. Information entropy is a basic concept of information theory. In this task, it is used to describe the criticality of data in each dimension and improve the utilization efficiency of multi-dimensional data. The impact factor mechanism counts the frequency of occurrence of features in each dimension and assigns an external parameter to each feature to reflect the importance of the feature in the overall data. This not only retains the information of the original features, but also emphasizes the role of important features. Finally, these weighted features are used as model input for further processing and analysis to extract key features that are useful for warehouse management decisions, providing strong support for optimizing storage and picking paths and improving warehouse efficiency. Figure 4As shown in Figure 1, the Attention mechanism layer is fused with the BP neural network. The weights generated by the Attention mechanism layer are applied to the output of the previous layer (such as between the input layer and the first hidden layer) to obtain a weighted feature representation. The importance of the input features is dynamically adjusted by calculating the Attention weights of the input data. In this way, the network can automatically learn the features that are more important for the optimization of the picking path during the training process. The Attention mechanism allows the model to dynamically adjust the weights of different parts when processing data, so that the model can pay more attention to the features that have an important impact on the optimization of the picking path. For example, the location, type, quantity of goods and their relative position relationship may be key factors affecting the path selection. The fourth step is to train the model using the training data. During the training process, the weights and biases in the network are adjusted through the back propagation algorithm to minimize the output error. After training and optimization, the network will be able to output feature vectors weighted by the Attention mechanism. These feature vectors will be used as inputs in subsequent steps to guide the optimization of the picking path.

[0052] In the present invention, the number of occurrences of a certain dimension data on all shelves in the warehouse is defined as the global influence of the dimension data, and its influence factor is expressed by the formula:

[0053] Among them I G represents the global impact factor, n represents the total number of warehouse shelves, i represents the data of the i-th dimension, H i Indicates the number of times the data of the i-th dimension appears on all shelves. Apply the influencing factor of each dimension to the corresponding feature, and the formula is as follows:

[0054] in is the feature of the i-th dimension, represents the global impact factor of the feature, Indicates features to which global influence information is added.

[0055] The Attention mechanism can help the model pay more attention to important information during feature extraction, thereby improving the performance of the model. The Attention mechanism is usually implemented by calculating a weight vector α, which represents the importance of different parts of the input data. When combined with a BP neural network, the Attention layer can be embedded into the network, and its output is used as the input of the subsequent layer. Assuming that the input data is X and the output of the Attention layer is A, the Attention mechanism can be expressed as: Attention

[0056] Where α is the attention weight vector, and ⊙ represents element-wise multiplication. The attention weight α is usually calculated as follows:

[0057] in, is a scoring function, is a learnable parameter, is the i-th part of the input data X, yes Ratings, represents the weight of the i-th part of the input data X.

[0058] BP neural network is trained through forward propagation and back propagation. In forward propagation, input data is transformed through multiple layers of neurons (including the Attention layer) to finally generate output. In back propagation, network parameters are updated through methods such as gradient descent based on the error between the output and the true value.

[0059] Assume that the network has L layers, and the output of the lth layer is , the activation function is σ, then the forward propagation can be expressed as:

[0060] in, and are the weight and bias of the lth layer respectively.

[0061] (5) Particle swarm algorithm steps: Figure 6As shown in the figure, this step optimizes the picking path based on the features extracted by the neural network. The particle swarm algorithm searches for the optimal solution by simulating the foraging behavior of bird flocks. The size and shape of the search space are determined according to the actual layout of the warehouse and the specific requirements of the picking task, and then a group of particles are randomly generated. Each particle represents a potential solution (i.e., a path). Their initial position and speed are randomly selected in the predefined search space, and they are continuously iterated and updated in this search space to approximate the optimal solution to find the optimal or approximately optimal picking path. First, the basic parameters of the particle swarm algorithm need to be set, including the size of the particle swarm (i.e., the number of agents), the initial position of each particle (representing a possible picking path solution), the initial speed (representing the direction and amplitude of the path adjustment), the maximum number of iterations, the inertia weight, the individual learning factor, and the social learning factor. Secondly, the initial position and speed of each particle are randomly generated in the solution space. The position of each particle can be a sequence representing the picking order, and the speed reflects the direction and intensity of the sequence adjustment. For each particle, the corresponding fitness value is calculated according to the path solution it represents. In the picking optimization problem, the fitness value is usually calculated based on the total length of the path, time cost, energy consumption or other related indicators, and the goal is to minimize these indicators. It can be obtained by taking two methods: individual optimal solution and global optimal solution. After obtaining the optimal solution, the particle speed is updated according to the particle's current speed, the difference between the individual optimal solution and the current position (cognitive part), the difference between the global optimal solution and the current position (social part), and the inertia weight. According to the updated speed, the position of the particle is adjusted, that is, the picking path plan is adjusted.

[0062] The specific calculation process is as follows: First, initialize the particle swarm. Let the size of the particle swarm be N, which indicates how many particles participate in the search. For each particle i (i=1,2,…,N), randomly initialize its position x i and speed v i . Position x i It can be represented as an encoding of a picking path (such as a list of cities), or as a continuous vector of parameters that indirectly determine the picking path. For each particle i, use its current position x i (or the picking path obtained by decoding its position) as input, and calculate the fitness function f(x i ).

[0063] The fitness function is usually inversely proportional to the total cost of the TSP problem (such as total walking distance, total time, etc.), that is, the lower the cost, the higher the fitness. For each particle i, compare its current fitness f(x i ) and its individual historical best position pbest i If f(x i ) is better, then update pbesti For x i Among all particles, find the particle with the best fitness and take its position as the global best position gbest.

[0064] According to the update formula of the particle swarm algorithm, the speed and position of each particle are updated. The update of speed and position usually depends on the inertia weight w and the learning factor c. 1 and c 2 , random number r 1 and r 2 、Individual historical best position pbest i and the global best position gbest i . Position x i and speed v i The update formula is:

[0065] Where t represents the number of updates.

[0066] The present invention combines neural networks with particle swarm algorithms, and can make full use of the advantages of both in feature extraction and optimization search. The neural network is responsible for extracting useful feature information from complex data, while the particle swarm algorithm performs efficient path search and optimization based on this feature information. This combination not only ensures the accuracy of feature extraction, but also improves the efficiency of path optimization. In warehouse storage and picking operations, factors such as cargo location and picking tasks may change over time. The storage and picking optimization method based on neural networks and particle swarm algorithms can adjust feature extraction and path optimization strategies in real time according to the latest data to adapt to the dynamic changes in the warehouse environment. This real-time and dynamic adaptability is the key to improving warehouse operating efficiency and response speed.

[0067] Corresponding to the aforementioned embodiment of a method for intelligent management and control of a storage and picking optimization group based on a neural network algorithm, the present invention also provides an embodiment of a device for intelligent management and control of a storage and picking optimization group based on a neural network algorithm.

[0068] See also Figure 7 An embodiment of the present invention provides an intelligent management and control device for a storage and picking optimization group based on a neural network algorithm, comprising a memory and one or more processors, wherein the memory stores executable code, and when the processor executes the executable code, it is used to implement an intelligent management and control method for a storage and picking optimization group based on a neural network algorithm in the above embodiment.

[0069] An embodiment of a storage and picking optimization group intelligent management and control device based on a neural network algorithm provided by the present invention can be applied to any device with data processing capabilities, and the any device with data processing capabilities can be a device or apparatus such as a computer. The device embodiment can be implemented through software, or through hardware or a combination of software and hardware. Taking software implementation as an example, as a device in a logical sense, it is formed by the processor of any device with data processing capabilities in which it is located reading the corresponding computer program instructions in the non-volatile memory into the internal memory for execution. From a hardware perspective, if Figure 7 As shown in FIG. 1 , a hardware structure diagram of any device with data processing capability in which a storage and picking optimization group intelligent control device based on a neural network algorithm provided by the present invention is located, except Figure 7 In addition to the processor, memory, network interface, and non-volatile memory shown, any device with data processing capabilities in which the apparatus in the embodiments is located may also include other hardware, generally based on the actual functions of the device with data processing capabilities, which will not be described in detail.

[0070] The implementation process of the functions and effects of each unit in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, and will not be repeated here.

[0071] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of the present invention. Ordinary technicians in this field can understand and implement it without paying creative work.

[0072] An embodiment of the present invention further provides a computer-readable storage medium having a program stored thereon. When the program is executed by a processor, an intelligent management and control method for a storage and picking optimization group based on a neural network algorithm in the above embodiment is implemented.

[0073] The computer-readable storage medium may be an internal storage unit of any device with data processing capability described in any of the aforementioned embodiments, such as a hard disk or a memory. The computer-readable storage medium may also be an external storage device of any device with data processing capability, such as a plug-in hard disk, a smart media card (SMC), an SD card, a flash card, etc. equipped on the device. Furthermore, the computer-readable storage medium may also include both an internal storage unit and an external storage device of any device with data processing capability. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capability, and may also be used to temporarily store data that has been output or is to be output.

[0074] The present invention also provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the intelligent management and control method of a storage and picking optimization group based on a neural network algorithm.

[0075] The above embodiments are used to illustrate the present invention rather than to limit the present invention. Any modification and change made to the present invention within the spirit of the present invention and the protection scope of the claims shall fall within the protection scope of the present invention.

Claims

1. A method for intelligent management and control of storage and picking optimization groups based on a neural network algorithm, characterized in that: The method comprises the following steps: (1) Collect warehouse data, obtain data related to storage and sorting tasks, and perform data preprocessing; (2) Map order demand and cargo location data and construct a corresponding traveling salesman problem instance to describe the path planning problem in warehouse picking tasks; (3) The influencing factor based on information entropy is introduced and combined with the Attention mechanism and BP neural network to extract features of the traveling salesman problem instance data, and the optimization solution is performed based on the particle swarm algorithm to obtain the storage and picking path solution.

2. According to claim 1, a method for intelligent management and control of storage and picking optimization groups based on a neural network algorithm is characterized in that: The data related to the storage and sorting tasks include storage and sorting path data, warehouse layout and cargo data, historical data and environmental parameter data.

3. According to claim 1, a method for intelligent management and control of storage and picking optimization groups based on a neural network algorithm is characterized in that: Data preprocessing specifically includes using isolation forests to detect and process outliers, performing data cleaning and normalization, and converting the data into a format suitable for neural network processing.

4. According to claim 1, a method for intelligent management and control of storage and picking optimization groups based on a neural network algorithm is characterized in that: The specific steps to construct a traveling salesman TSP problem instance are as follows: (2.1) Number or mark each storage location in the warehouse; (2.2) Calculate the moving distance or cost between storage nodes as the "weight" or "edge" in the traveling salesman problem; (2.3) Construct a graph based on the obtained "weights" or "edges", where each node represents a storage location and each edge represents the cost of moving between two storage locations; (2.4) Set the starting and ending points of the TSP problem according to the actual needs of warehouse operations; (2.5) Integrate the nodes, edges, starting points, and ending points obtained in the above steps into a complete TSP problem instance.

5. According to claim 1, a method for intelligent management and control of storage and picking optimization groups based on a neural network algorithm is characterized in that: Based on the traveling salesman TSP problem example, the path planning problem in warehouse picking tasks is described, that is, how to traverse all storage nodes once and only once with the minimum total cost or total moving distance, and finally return to the starting point or the end point.

6. The method for intelligent management and control of storage and picking optimization groups based on a neural network algorithm according to claim 1 is characterized in that: The specific process of feature extraction in step (3) is as follows: (1) Preliminary data processing and feature extraction based on BP neural network; (2) Introduce an impact factor based on information entropy to count the frequency of occurrence of features in each dimension, assign an external parameter to each feature to reflect the importance of the feature, calculate the information entropy of the features in each dimension to evaluate the amount of information it contains and its importance in the overall data, and assign an impact factor to each feature based on information entropy for subsequent feature weighting processing; (3) Fusion of the Attention mechanism layer and the BP neural network; applying the weights generated by the Attention mechanism layer to the output of the previous layer to obtain a weighted feature representation; (4) Train the model and adjust the weights and biases in the network through the back-propagation algorithm to minimize the output error. The trained network outputs a feature vector weighted by the Attention mechanism to guide the optimization of the picking path.

7. The method for intelligent management and control of storage and picking optimization groups based on a neural network algorithm according to claim 6 is characterized in that: The number of times a certain dimension data appears on all shelves in the warehouse is defined as the global influence of the dimension data, and its influence factor is expressed by the formula: ； Among them I G represents the global impact factor, n represents the total number of warehouse shelves, i represents the data of the i-th dimension, H i Indicates the number of times the data of the i-th dimension appears on all shelves; the influencing factor of each dimension is applied to the corresponding feature, and the formula is as follows: ； in is the feature of the i-th dimension, represents the global impact factor of the feature, Indicates features to which global influence information is added.

8. The method for intelligent management and control of storage and picking optimization groups based on a neural network algorithm according to claim 1 is characterized in that: Based on the particle swarm algorithm, the optimization solution is obtained, and the process of obtaining the storage and picking path solution is as follows: (1) Setting the parameters of the particle swarm algorithm, including: the size of the particle swarm, that is, the number of agents, the initial position of each particle, representing a possible picking path solution, the initial speed, representing the direction and amplitude of the path adjustment, as well as the maximum number of iterations, inertia weight, individual learning factor, and social learning factor; (2) Randomly generate the initial position and velocity of each particle in the solution space; the position of each particle represents the sequence of the picking order, and the velocity reflects the direction and intensity of the sequence adjustment; for each particle, calculate its corresponding fitness value according to the path plan it represents; in the picking optimization problem, the fitness value is calculated based on the total length of the path, time cost or energy consumption index, and the goal is to minimize these indicators; (3) After obtaining the optimal solution, the particle speed is updated according to the particle’s current speed, the difference between the individual optimal solution and the current position, the difference between the global optimal solution and the current position, and the inertia weight; the particle’s position is adjusted according to the updated speed, that is, the storage and picking path plan is adjusted.

9. An intelligent control device for storage and picking optimization group based on neural network algorithm, comprising a memory and one or more processors, wherein the memory stores executable code, characterized in that: When the processor executes the executable code, an intelligent management and control method for a picking optimization group based on a neural network algorithm as described in any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by the processor, an intelligent management and control method for a storage and picking optimization group based on a neural network algorithm as described in any one of claims 1 to 8 is implemented.

Images