Order batching method and device based on improved K-means algorithm

Abstract

The invention relates to the technical field of clustering analysis, in particular to an order batching method and equipment based on an improved K-means algorithm, and the method comprises the steps: firstly, enabling orders to be mapped into high-dimensional feature vectors through one-hot coding based on an EIQ analysis framework, and constructing a low-dimensional feature subset through two-stage feature selection, the dimension disaster is relieved; the feature representativeness is improved; secondly, aiming at the sensitivity of the traditional K-means to an initial center and the limitation of the Euclidean distance in high-dimensional sparse data, on one hand, a roulette strategy is used to optimize the selection of a clustering center and reduce the local optimal risk, and on the other hand, a cosine distance is used to replace the Euclidean distance and the SKU overlapping degree between orders is measured by direction similarity; therefore, the orders with similar requirements can be more accurately aggregated to the same batch.

Description

Technical Field

[0001] The present invention relates to the technical field of cluster analysis, and in particular to an order batching method and device based on an improved K-means algorithm. Background Art

[0002] In China, with the deepening of pharmaceutical policy reforms such as tiered diagnosis and treatment, zero-markup drug sales, and national centralized drug procurement, pharmaceutical distribution companies are facing a business restructuring characterized by an increase in small-lot orders and a decrease in full-package orders. The traditional person-to-person picking model, due to its high error rate and high labor costs, has become unsuitable for these new demands, prompting the rapid development of goods-to-person picking systems. Multi-level shuttle systems, a representative example of the goods-to-person model, offer high-density storage and efficient inbound and outbound operations. These systems consist of multi-level storage racks, shuttles, elevators, conveyor lines, and a control system. Bins are removed from storage via a coordinated process of elevators and shuttles, and subsequent order picking occurs at the picking table. However, due to limited physical space at the picking table, dedicated bin locations cannot be allocated for each order, necessitating the use of batched orders to achieve bin reuse. Centralized order processing can lead to congestion in conveyor lines, while decentralized processing reduces equipment utilization due to the cross-aisle distribution of SKUs. Therefore, developing an efficient order batching algorithm has become a key link in optimizing the operation of the multi-layer shuttle system. A reasonable order batching method can significantly improve the overall efficiency of the system by optimizing SKU demand overlap and reducing the number of repeated outbound shipments of material boxes.

[0003] The traditional K-means algorithm is highly sensitive to the initial cluster centers. Randomly selecting initial centers can lead to local optima, resulting in unstable clustering results. These challenges are particularly acute in the pharmaceutical logistics sector. With the increasing variety of orders and the expansion of the number of SKUs, traditional batching methods are increasingly difficult to address. Therefore, a new algorithm that integrates feature selection and clustering optimization is urgently needed to address this issue and meet the specific needs of the pharmaceutical logistics industry. Summary of the Invention

[0004] In view of this, the object of the present invention is to provide an order batching method and device based on an improved K-means algorithm to overcome the problems existing in the current prior art.

[0005] To achieve the above objectives, the present invention adopts the following technical solutions: On the one hand, the present application provides an order batching method based on an improved K-means algorithm, comprising: Get order information; Extract features from the order information using one-hot encoding; The extracted features are selected by IK frequency and Pearson correlation coefficient to generate a low-dimensional feature set; Selecting cluster centers in the low-dimensional feature set by using a roulette wheel strategy; Constructing a low-dimensional feature vector of the order based on the low-dimensional feature set, and calculating the cosine distance between the low-dimensional feature vectors; The orders are clustered and batched according to the cluster center and the cosine distance using an improved K-means order batching algorithm.

[0006] Furthermore, in the above method, extracting features from the order information by one-hot encoding includes: Extract the full set of SKU features from the order information, extracting all different SKUs; One-hot encode each order according to all different SKUs to generate the feature vector of the corresponding order.

[0007] Furthermore, the method described above, wherein the extracted features are selected by using the IK frequency and the Pearson correlation coefficient to generate a low-dimensional feature subset, includes: Calculate the order frequency of each SKU in the order information, and sort the SKUs from high to low according to the order frequency; According to the sorting results, a preset number of top-ranked SKUs are selected as feature subsets to construct feature subvectors and order sample vectors for the order; Calculate the Pearson correlation coefficient between the features in the order sample vector using the Pearson correlation coefficient calculation formula. If the correlation coefficient is less than a preset threshold, the feature with the lower order frequency is recorded as a feature to be removed from the two features involved in the calculation. The features to be removed are removed from the feature sub-vector of the order to construct a low-dimensional feature set of the order.

[0008] Furthermore, in the above method, the selecting of cluster centers from the low-dimensional feature set by using a roulette wheel strategy includes: Randomly select a sample point in the low-dimensional feature set as the first cluster center; Randomly select a sample that has not been selected as the center in the low-dimensional feature set, and calculate the minimum value of the distance between the sample and all currently selected centers; Calculate the probability of the next cluster center being selected based on the minimum value of the center distance through the probability formula; Selecting the next cluster center from the samples not selected as the center of the low-dimensional feature set according to the probability that the next cluster center is selected; The probability of selecting the cluster center is recalculated, and new cluster centers are selected again until a preset number of cluster centers are selected.

[0009] Furthermore, the method described above, wherein constructing the low-dimensional feature vector of the order based on the low-dimensional feature set and calculating the cosine distance between the low-dimensional feature vectors, includes: Constructing a low-dimensional feature vector of the order according to the low-dimensional feature set; Calculate the cosine similarity between the low-dimensional feature vectors using a cosine similarity calculation formula; The cosine distance between the low-dimensional feature vectors is calculated according to the cosine similarity and cosine distance calculation formula.

[0010] Furthermore, in the above method, the Pearson correlation coefficient calculation formula is:

[0011] in, is the Pearson correlation coefficient between features, Features The average value over N orders, Features The average value over N orders, is the order sample vector The i-th feature of is the order sample vector The i-th feature of .

[0012] Furthermore, in the above method, the probability formula is:

[0013] Among them, P(x=C next ) is the probability of the next cluster center being selected, D(X) 2 is the minimum distance between the sample and all currently selected centers, is the sum of the minimum distance squares of all sample points z in the data set.

[0014] Furthermore, in the above method, the cosine similarity calculation formula is:

[0015] Among them, cos(X,Y) is the cosine similarity between the low-dimensional feature vector X and the low-dimensional feature vector Y, is the Euclidean norm of the low-dimensional feature vector X, is the Euclidean norm of the low-dimensional feature vector Y.

[0016] Furthermore, in the above method, the cosine distance calculation formula is:

[0017] in, is the cosine distance between the low-dimensional feature vector X and the low-dimensional feature vector Y.

[0018] On the other hand, the present application provides an order batching device based on an improved K-means algorithm, comprising a processor and a memory, wherein the processor is connected to the memory: The processor is configured to call and execute the program stored in the memory; The memory is used to store the program, and the program is at least used to execute any one of the above order batching methods based on the improved K-means algorithm.

[0019] The beneficial effects of the present invention are: This application first obtains order information, extracts features from the order information through one-hot encoding, selects the extracted features through IK frequency and Pearson correlation coefficient, generates a low-dimensional feature set, then selects cluster centers from the low-dimensional feature set through a roulette wheel strategy, constructs low-dimensional feature vectors of the order based on the low-dimensional feature set, calculates the cosine distance between the low-dimensional feature vectors, and finally clusters and batches the orders based on the cluster centers and cosine distance using an improved K-means order batching algorithm. In this application, firstly, based on the EIQ analysis framework, one-hot encoding is used to map orders into high-dimensional feature vectors, and a low-dimensional feature subset is constructed through two-stage feature selection to alleviate the "curse of dimensionality" and improve feature representativeness; secondly, in view of the limitations of traditional K-means that are sensitive to initial centers and Euclidean distance in high-dimensional sparse data, a roulette wheel strategy is used to optimize the selection of cluster centers and reduce the risk of local optimality. On the other hand, cosine distance is used instead of Euclidean distance, and the SKU overlap between orders is measured by directional similarity, so as to more accurately aggregate orders with similar demands into the same batch. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. 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 paying any creative work.

[0021] Figure 1 This is a flowchart provided by an embodiment of an order batching method based on an improved K-means algorithm of the present invention; Figure 2 This is a structural diagram provided by an embodiment of an order batching device based on an improved K-means algorithm of the present invention. DETAILED DESCRIPTION

[0022] To make the objectives, technical solutions, and advantages of the present invention more apparent, the technical solutions of the present invention will be described in detail below. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other implementations obtained by those of ordinary skill in the art without inventive effort are within the scope of protection of the present invention.

[0023] Figure 1 This is a flowchart of an embodiment of an order batching method based on an improved K-means algorithm provided by the present invention. Figure 1 , this embodiment may include the following steps: S1. Obtain order information; S2, extract features of order information through one-hot encoding; S3, select the extracted features by IK frequency and Pearson correlation coefficient to generate a low-dimensional feature set; S4, select cluster centers from low-dimensional feature sets through roulette strategy; S5. Construct a low-dimensional feature vector of the order based on the low-dimensional feature set, and calculate the cosine distance between the low-dimensional feature vectors; S6. Cluster and batch the orders using the improved K-means order batching algorithm based on the cluster center and cosine distance.

[0024] It can be understood that this embodiment first obtains order information, extracts features from the order information through one-hot encoding, selects the extracted features using IK frequency and Pearson correlation coefficient to generate a low-dimensional feature set, then selects cluster centers from the low-dimensional feature set using a roulette wheel strategy, constructs low-dimensional feature vectors of the orders based on the low-dimensional feature set, calculates the cosine distance between the low-dimensional feature vectors, and finally clusters and batches the orders using an improved K-means order batching algorithm based on the cluster centers and cosine distance. In this embodiment, first, based on the EIQ analysis framework, one-hot encoding is used to map orders into high-dimensional feature vectors, and low-dimensional feature subsets are constructed through two-stage feature selection to alleviate the "curse of dimensionality" and improve feature representativeness. Secondly, in view of the traditional K-means sensitivity to initial centers and the limitations of Euclidean distance in high-dimensional sparse data, a roulette wheel strategy is used to optimize the selection of cluster centers and reduce the risk of local optimality. On the other hand, cosine distance is used instead of Euclidean distance, and directional similarity is used to measure the SKU overlap between orders, thereby more accurately aggregating orders with similar requirements into the same batch.

[0025] Preferably, step S2 includes: Extract the full set of SKU features from the order information and extract all different SKUs; One-hot encode each order according to all different SKUs to generate the feature vector of the corresponding order.

[0026] It is understandable that based on the high-dimensional, sparse, and discrete characteristics of pharmaceutical orders, one-hot encoding is very suitable for mapping each order into a 0 / 1 vector across several possible SKU dimensions.

[0027] Order collection , and record all the different SKUs involved as Before any screening is done, there may theoretically be thousands or even tens of thousands of SKUs, and further "feature selection" is needed on this basis.

[0028] For any order , if it contains , then the vector In the The dimension is 1, otherwise it is 0. After encoding, the feature vector of each order can be expressed as

[0029] at this time It may be very large, so it is necessary to use the feature selection process to reduce the dimension and avoid the "curse of dimensionality".

[0030] Preferably, step S3 includes: Calculate the order frequency of each SKU in the order information and sort the SKUs from high to low based on the order frequency; According to the sorting results, a preset number of top-ranked SKUs are selected as feature subsets to construct the feature sub-vector and order sample vector of the order; The Pearson correlation coefficient between features in the order sample vector is calculated using the Pearson correlation coefficient calculation formula. If the correlation coefficient is less than the preset threshold, the feature with the lower order frequency is marked as the feature to be removed. Remove the features to be removed from the feature sub-vector of the order and construct a low-dimensional feature set of the order.

[0031] It is understandable that to improve clustering performance, it is necessary to filter existing one-hot features, remove unimportant or redundant features, and form a smaller and better low-dimensional feature set. This specifically involves two steps: important feature analysis and feature correlation analysis.

[0032] Count the order frequency of each SKU , that is, how many orders does the SKU appear in. Sort by size in descending order:

[0033] Setting feature extraction threshold , that is, select the top n% SKUs as the feature subset and construct the feature sub-vector of the order.

[0034] when When n is small, the feature subsets are too small and it is difficult to accurately describe the differences between orders; when n is large, the “dimensionality curse” and noise interference increase; The changes tend to be stable and the clustering effect is ideal value or interval.

[0035] Even if the former Even for a large number of SKUs, some features may still have almost the same "explanation" for the same batch of orders, showing a high degree of redundancy. To further eliminate these redundant features, the Pearson correlation coefficient is used to measure the linear correlation between features.

[0036] For any two features and , the corresponding order sample vector can be recorded as:

[0037] in If the order Contains features but , otherwise 0.

[0038] Calculate the Pearson correlation coefficient between features:

[0039] in Features The average value over N orders.

[0040] Set the feature removal threshold m, if , indicating that the two features are almost highly positively correlated in the order data. You can choose to retain the feature with higher IK frequency and remove the other feature with lower IK frequency. Repeat until the correlation coefficients between all retained features do not exceed .

[0041] Through the above-mentioned important feature analysis and inter-feature correlation analysis, a more representative and non-redundant set of feature subsets can be selected from massive SKUs, laying an efficient foundation for subsequent clustering processing.

[0042] Preferably, step S4 includes: randomly selecting a sample point in the low-dimensional feature set as the first cluster center; Randomly select a sample that has not been selected as the center in the low-dimensional feature set, and calculate the minimum distance between the sample and all currently selected centers; According to the minimum value of the center distance, the probability of the next cluster center being selected is calculated through the probability formula; According to the probability of the next cluster center being selected, the next cluster center is selected from the samples that are not selected as the center in the low-dimensional feature set; The probability of selecting the cluster center is recalculated, and new cluster centers are selected again until a preset number of cluster centers are selected.

[0043] It is understandable that after the selection of feature subsets is completed, each order (sample) will be represented as a relatively low-dimensional 0 / 1 feature vector:

[0044] in Much smaller than the original dimension At this point, the improved K-means can be used for clustering batches.

[0045] In order to reduce the algorithm's sensitivity to the selection of initial centers and avoid falling into local optimality, the roulette wheel method is used to select cluster centers.

[0046] First, randomly select a sample point As the first center; then, for any sample that is not selected as the center , calculate the minimum distance between the sample and all currently selected centers, and record it as , let the probability of the next center being selected be:

[0047] And use the Roulette Wheel method to draw ; Finally, iterate until the Initial cluster centers.

[0048] Preferably, step S5 includes: Construct a low-dimensional feature vector of the order based on the low-dimensional feature set; Calculate the cosine similarity between low-dimensional feature vectors using the cosine similarity calculation formula; The cosine distance between low-dimensional feature vectors is calculated according to the cosine similarity and cosine distance calculation formulas.

[0049] It's understandable that Euclidean distance measures the absolute distance between points in space, while cosine similarity measures the angle between spatial vectors. The numerical result of the former is affected by the specific values ​​on each coordinate axis, while the latter focuses on directional differences. Given that the feature subsets in this paper are 0 / 1 vectors, emphasizing the differences between different features, using cosine similarity as a similarity distance calculation is more suitable for this paper. Cosine distance, by focusing on directional differences between vectors, transforms the similarity perspective into a distance metric. Therefore, this paper selects cosine distance as the distance metric.

[0050] For eigenvectors of two orders: and , Euclidean distance It can be expressed as:

[0051] Cosine similarity can be expressed as:

[0052] in and is the Euclidean norm of the vector, respectively:

[0053] The value range of cosine similarity is [-1, 1]. The closer it is to 1, the more similar the two vectors are, and the closer it is to -1, the less similar the two vectors are.

[0054] Then the cosine distance can be expressed as

[0055] The cosine distance has a value range of [0, 2]. When the angle between two vectors is 0 (that is, they are exactly the same), the cosine distance is 0; when the angle between them is 180° (that is, they are completely opposite), the cosine distance is 2.

[0056] In some optional embodiments, the pseudo code of the improved K-means clustering algorithm is shown in Table 1.

[0057] Table 1 Pseudocode of the improved K-means clustering algorithm

[0058] The present invention also provides an order batching device based on an improved K-means algorithm, which is used to implement the above method embodiment. Figure 2 This is a structural diagram of an embodiment of an order batching device based on an improved K-means algorithm provided by the present invention. Figure 2As shown, the order batching device based on the improved K-means algorithm of this embodiment includes a processor 21 and a memory 22, wherein the processor 21 is connected to the memory 22. The processor 21 is used to call and execute the program stored in the memory 22; the memory 22 is used to store the program, which is used to at least execute the order batching method based on the improved K-means algorithm in the above embodiment.

[0059] The specific implementation scheme of the order batching device based on the improved K-means algorithm provided in the embodiments of the present application can refer to the implementation scheme of the order batching method based on the improved K-means algorithm in any of the above embodiments, and will not be repeated here.

[0060] It can be understood that the same or similar parts of the above embodiments can be referenced to each other, and the contents not described in detail in some embodiments can refer to the same or similar contents in other embodiments.

[0061] It should be noted that, in the description of the present invention, the terms "first", "second", etc. are used for descriptive purposes only and should not be understood as indicating or implying relative importance. In addition, in the description of the present invention, unless otherwise specified, the meaning of "plurality" is at least two.

[0062] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code comprising one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may be performed out of the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention pertain.

[0063] It should be understood that various components of the present invention may be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods may be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof may be used: a discrete logic circuit having logic gate circuits for implementing logic functions on data signals, an application-specific integrated circuit having suitable combinational logic gate circuits, a programmable gate array (PGA), a field-programmable gate array (FPGA), etc.

[0064] Those skilled in the art will understand that all or part of the steps in the method of the above embodiment can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.

[0065] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing module, or each unit may exist physically separately, or two or more units may be integrated into a single module. The aforementioned integrated modules may be implemented in the form of hardware or in the form of software functional modules. If the integrated modules are implemented in the form of software functional modules and sold or used as independent products, they may also be stored in a computer-readable storage medium.

[0066] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.

[0067] Throughout this specification, reference to terms such as "one embodiment," "some embodiments," "examples," "specific examples," or "some examples" means that a specific feature, structure, material, or characteristic described in conjunction with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0068] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. An order batching method based on an improved K-means algorithm, characterized in that: include: Get order information; Extract features from the order information using one-hot encoding; The extracted features are selected by IK frequency and Pearson correlation coefficient to generate a low-dimensional feature set; Selecting cluster centers in the low-dimensional feature set by using a roulette wheel strategy; Constructing a low-dimensional feature vector of the order based on the low-dimensional feature set, and calculating the cosine distance between the low-dimensional feature vectors; The orders are clustered and batched according to the cluster center and the cosine distance using an improved K-means order batching algorithm.

2. The method according to claim 1, characterized in that The feature extraction of the order information by one-hot encoding includes: Extract the full set of SKU features from the order information, extracting all different SKUs; One-hot encode each order according to all different SKUs to generate the feature vector of the corresponding order.

3. The method according to claim 2, characterized in that The extracted features are selected by IK frequency and Pearson correlation coefficient to generate a low-dimensional feature subset, including: Calculate the order frequency of each SKU in the order information, and sort the SKUs from high to low according to the order frequency; According to the sorting results, a preset number of top-ranked SKUs are selected as feature subsets to construct feature subvectors and order sample vectors for the order; Calculate the Pearson correlation coefficient between the features in the order sample vector using the Pearson correlation coefficient calculation formula. If the correlation coefficient is less than a preset threshold, the feature with the lower order frequency is recorded as a feature to be removed from the two features involved in the calculation. The features to be removed are removed from the feature sub-vector of the order to construct a low-dimensional feature set of the order.

4. The method according to claim 3, characterized in that The selecting of cluster centers in the low-dimensional feature set by the roulette wheel strategy includes: Randomly select a sample point in the low-dimensional feature set as the first cluster center; Randomly select a sample that has not been selected as the center in the low-dimensional feature set, and calculate the minimum value of the distance between the sample and all currently selected centers; Calculate the probability of the next cluster center being selected based on the minimum value of the center distance through the probability formula; Selecting the next cluster center from the samples not selected as the center of the low-dimensional feature set according to the probability that the next cluster center is selected; The probability of selecting the cluster center is recalculated, and new cluster centers are selected again until a preset number of cluster centers are selected.

5. The method according to claim 4, characterized in that The step of constructing the low-dimensional feature vector of the order according to the low-dimensional feature set and calculating the cosine distance between the low-dimensional feature vectors includes: Constructing a low-dimensional feature vector of the order according to the low-dimensional feature set; Calculate the cosine similarity between the low-dimensional feature vectors using a cosine similarity calculation formula; The cosine distance between the low-dimensional feature vectors is calculated according to the cosine similarity and cosine distance calculation formula.

6. The method according to claim 5, characterized in that The Pearson correlation coefficient calculation formula is: in, is the Pearson correlation coefficient between features, Features The average value over N orders, Features The average value over N orders, is the order sample vector The i-th feature of is the order sample vector The i-th feature of .

7. The method according to claim 6, characterized in that The probability formula is: Among them, P(x=C next ) is the probability of the next cluster center being selected, D(X) 2 is the minimum distance between the sample and all currently selected centers, is the sum of the minimum distance squares of all sample points z in the data set.

8. The method according to claim 7, characterized in that The cosine similarity calculation formula is: Among them, cos(X,Y) is the cosine similarity between the low-dimensional feature vector X and the low-dimensional feature vector Y, is the Euclidean norm of the low-dimensional feature vector X, is the Euclidean norm of the low-dimensional feature vector Y.

9. The method according to claim 8, characterized in that The cosine distance calculation formula is: in, is the cosine distance between the low-dimensional feature vector X and the low-dimensional feature vector Y.

10. An order batching device based on an improved K-means algorithm, characterized in that: The device comprises a processor and a memory, wherein the processor is connected to the memory: The processor is configured to call and execute the program stored in the memory; The memory is used to store the program, and the program is at least used to execute the order batching method based on the improved K-means algorithm as described in any one of claims 1 to 9.

Images