Sorting center intelligent management method and system based on cloud platform

Abstract

The invention relates to the technical field of logistics automation and information processing, and discloses a sorting center intelligent management method and system based on a cloud platform. The method comprises the following steps: acquiring an image and input information of an abnormal package, and extracting a preliminary attribute; if the preliminary attribute is incomplete, analyzing the high-definition image, calculating a damage index and generating a repair queue; according to the repair queue, matching from a historical feature library to obtain inferred physical attributes; the input information is analyzed, all known attributes are synthesized, and a final destination is determined; generating an initial path according to the final destination and the real-time working condition; the preliminary path is optimized by fusing historical congestion records; and carrying out load balancing based on the optimized path, generating a final allocation scheme, and issuing and executing. According to the method, a complete process from multi-dimensional information intelligent completion to dynamic closed-loop scheduling is constructed, so that the technical problem of low abnormal package processing efficiency in the prior art is solved.

Description

Technical Field

[0001] This invention relates to the field of logistics automation and information processing technology, and in particular to an intelligent management method and system for sorting centers based on a cloud platform. Background Technology

[0002] As the core hub of the modern supply chain, the operational efficiency and accuracy of logistics sorting centers are crucial to meeting the ever-growing demands of e-commerce. With the explosive growth in parcel volume and the increasing complexity of business models, how to efficiently and intelligently handle various dynamic changes in the sorting process has become a key challenge restricting the industry's development.

[0003] In existing technologies, sorting centers mostly employ static rule-based scheduling strategies and labor-intensive anomaly handling processes. For example, sorting routes are typically pre-defined, and when abnormal packages with damaged labels or incorrect information occur, manual intervention is generally relied upon for identification and handling. However, this static and manually driven model is increasingly showing its limitations in the face of today's large-scale, dynamic logistics scenarios: on the one hand, manual handling of abnormal packages is not only slow and error-prone, but also easily becomes a core bottleneck of the entire sorting line during peak business periods; on the other hand, the fixed route rules cannot be dynamically adjusted according to real-time congestion or equipment load, resulting in a lack of adaptability to changes on-site, thus causing resource waste and decreased efficiency.

[0004] In summary, existing technologies suffer from low efficiency in handling abnormal packages. Summary of the Invention

[0005] This invention provides a cloud-based intelligent management method and system for sorting centers to solve the technical problem of low efficiency in handling abnormal packages in existing technologies.

[0006] Firstly, in order to solve the above-mentioned technical problems, the present invention provides a cloud platform-based intelligent management method for sorting centers, comprising:

[0007] Acquire and verify the image data and entered information of the abnormal package to obtain preliminary attribute information;

[0008] If the preliminary attribute information is incomplete, a high-resolution scan image of the abnormal package is obtained, the damage index is calculated and sorted, and a repair priority queue is generated.

[0009] Based on the repair priority queue, similar package features are matched from a pre-established historical database of abnormal package features to obtain inferential physical attribute information;

[0010] The entered information is parsed to obtain a set of candidate destinations. The optimal destination probability is calculated by combining the preliminary attribute information and the inferred physical attribute information, and the final destination of the package is determined.

[0011] Based on the final package destination, real-time device status and path occupancy data are obtained to generate a preliminary allocation path;

[0012] Peak-hour route congestion records are extracted from a pre-established historical database of sorting route congestion, and the preliminary allocation route is optimized to obtain an optimized allocation route.

[0013] Based on the optimized allocation path, load balancing analysis is performed on the equipment, a final allocation plan is generated, and the plan is sent out for execution.

[0014] In one optional implementation, the step of acquiring and verifying the image data and recorded information of the abnormal package to obtain preliminary attribute information includes:

[0015] Real-time acquisition of image data of the abnormal package and the corresponding recorded information;

[0016] The image data is denoised and enhanced to obtain optimized image data;

[0017] Extract label content from the optimized image data to generate a label text set;

[0018] Calculate the matching degree between the tag text set and the entered information, and integrate the tag text set and the entered information based on the matching degree to generate a preliminary attribute set;

[0019] The initial attribute set is compared with the package records in the pre-established cloud platform database to verify consistency, and the initial attribute information is extracted.

[0020] In one optional implementation, if the preliminary attribute information is incomplete, obtaining a high-resolution scan image of the abnormal package, calculating and sorting the damage index, and generating a repair priority queue includes:

[0021] The preliminary attribute information is checked for completeness. If the check result is incomplete, the high-resolution scanned image is obtained.

[0022] The high-resolution scanned image is processed using a pre-trained convolutional neural network model to obtain the coordinates of the damaged pixels on the packaged surface;

[0023] The area ratio of the damaged pixels is calculated, and the area ratio is mapped using a preset mapping rule to obtain the damage index;

[0024] The abnormal packages are sorted from high to low according to the damage index to obtain the repair priority queue;

[0025] The training process of the convolutional neural network model is as follows:

[0026] Obtain a training sample dataset containing high-resolution scanned images and corresponding labels of the actual damaged pixel locations;

[0027] Construct a convolutional neural network model and input the training sample dataset into the model to calculate the loss between the predicted damaged pixel coordinates and the actual damaged pixel location labels;

[0028] Based on the loss, the model parameters are iteratively adjusted using the backpropagation algorithm until the loss is less than a preset loss threshold, thus obtaining a trained convolutional neural network model.

[0029] In one optional implementation, the step of matching similar package features from a pre-established historical database of abnormal package features to obtain inferential physical attribute information according to the repair priority queue includes:

[0030] The size and material of the current package are extracted from the preliminary attribute information and the entered information to obtain its physical characteristics;

[0031] Based on the repair priority queue and in conjunction with the physical characteristics, match historical package cases with similar characteristics in the abnormal package feature history database;

[0032] Based on the attribute records of the historical package cases, the missing attributes of the current package are supplemented to obtain the inferred physical attribute information.

[0033] In one optional implementation, parsing the entered information to obtain a set of candidate destinations, and combining the preliminary attribute information with the inferred physical attribute information to calculate the optimal destination probability and determine the final package destination, includes:

[0034] The semantic structure of the entered information is analyzed, and keywords related to the address are extracted to obtain the candidate destination set;

[0035] For each candidate destination in the candidate destination set, at least one business feature is extracted from the address, estimated timeliness, or path cost corresponding to the candidate destination, and a multi-dimensional feature vector is constructed by combining the preliminary attribute information and the inferred physical attribute information.

[0036] The multidimensional feature vector is compared with the feature vector of historical successful delivery cases in the pre-established historical successful delivery database to calculate the similarity and find the k nearest neighbor historical cases for each candidate destination.

[0037] The frequency of each destination in the k nearest neighbor historical cases is counted, and the frequency of occurrence is used as the basic score. Combined with the preset sorting rules, the candidate destination with the highest score is determined as the final package destination.

[0038] In one optional implementation, the step of obtaining device status and path occupancy data in real time based on the final package destination and generating a preliminary allocation route includes:

[0039] Obtain the preset path topology model of the sorting center, as well as the real-time status of the equipment and the path occupancy data;

[0040] Based on the path topology model, and combined with the device status and the path occupancy data, path optimization calculation is performed to obtain an initial path scheme;

[0041] The initial path scheme is subjected to path conflict detection and adjustment to obtain the preliminary allocation path.

[0042] In one optional implementation, the step of extracting peak-hour route congestion records from a pre-established historical database of sorting route congestion and optimizing the initial allocation route to obtain an optimized allocation route includes:

[0043] Filter and extract peak-hour route congestion records from the historical congestion database of the sorting route;

[0044] Identify the set of congested road segments in the preliminary dispatch route that overlap with the peak-hour route congestion records;

[0045] For the set of congested road segments, based on preset path adjustment rules and allocation weights, path replanning calculations are performed to obtain the optimized allocation path.

[0046] In one optional implementation, the step of performing load balancing analysis on the devices based on the optimized allocation path, generating a final allocation plan, and issuing it for execution includes:

[0047] Obtain the rated capacity and real-time operating data of each piece of equipment, and calculate the load rate of each piece of equipment;

[0048] The statistical coefficient of variation is calculated based on the load rate to obtain a quantitative assessment result of load balance.

[0049] Based on the quantitative evaluation results, task optimization and redistribution calculations are performed to obtain an optimized task allocation scheme;

[0050] The optimized task allocation scheme was simulated and verified to obtain the final allocation scheme that passed the verification.

[0051] The final allocation plan is converted into equipment control commands and sent to the controller for execution.

[0052] Secondly, the present invention provides a cloud-based intelligent management system for sorting centers, comprising:

[0053] The data acquisition module is used to acquire and verify the image data and input information of the abnormal package to obtain preliminary attribute information;

[0054] The damage assessment module is used to acquire high-resolution scan images of the abnormal package if the preliminary attribute information is incomplete, calculate and sort the damage index, and generate a repair priority queue.

[0055] The attribute completion module is used to match similar package features from a pre-established historical database of abnormal package features according to the repair priority queue, and obtain inferential physical attribute information.

[0056] The destination decision module is used to parse the input information, obtain a set of candidate destinations, and combine the preliminary attribute information with the inferred physical attribute information to calculate the optimal destination probability and determine the final package destination.

[0057] The preliminary route generation module is used to generate a preliminary allocation route based on the final package destination, by acquiring device status and route occupancy data in real time.

[0058] The route optimization module is used to extract peak-hour route congestion records from a pre-established historical database of sorting route congestion, and optimize the initial allocation route to obtain an optimized allocation route.

[0059] The load balancing and scheduling module is used to perform load balancing analysis on the devices based on the optimized allocation path, generate the final allocation plan, and send it out for execution.

[0060] Thirdly, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the cloud-based intelligent management method for sorting centers as described in any one of the above.

[0061] Fourthly, the present invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the cloud-based intelligent management method for sorting centers described in any one of the above-mentioned methods.

[0062] Compared with the prior art, the present invention has the following beneficial effects:

[0063] (1) This invention acquires package images and text information. When the information is incomplete, it employs a lightweight convolutional neural network, collaborative filtering, natural language processing, and the K-nearest neighbor algorithm to perform multi-layered, progressive information completion and destination decision-making. This combination of technologies transforms the traditional, slow, and error-prone process that relies on manual identification into an automated, intelligent decision-making process based on multi-dimensional data analysis. This solves the problem of existing technologies lacking effective means of identifying and handling abnormal packages, significantly improving the processing efficiency and decision-making accuracy of abnormal packages.

[0064] (2) This invention establishes a two-layer path planning mechanism that combines real-time response with historical experience optimization by acquiring real-time equipment status and path occupancy data of the sorting center and integrating peak-hour congestion records from the historical database. This mechanism changes the existing technology's reliance on fixed, static path rules, solves the problem that path planning cannot adapt to dynamic changes in working conditions, effectively avoids sorting congestion, and improves the overall throughput and operational smoothness of the sorting center.

[0065] (3) This invention, after determining the path, further analyzes the load rate of each device on the path, and when imbalance is detected, optimizes and redistributes tasks in combination with constraints such as maintenance and production capacity, and finally sends the simulation-verified solution to the controller for execution. This forms a complete closed-loop control system from information processing and path planning to load balancing execution, which solves the problem of existing scheduling schemes being disconnected from the execution capabilities of physical equipment, which may lead to single-point overload or resource idleness, and ensures the stability and resource utilization of the entire system. Attached Figure Description

[0066] Figure 1 This is a schematic diagram of a cloud-based intelligent management method for sorting centers provided in the first embodiment of the present invention;

[0067] Figure 2 This is a schematic diagram of the structure of a cloud-based intelligent management system for sorting centers provided in the second embodiment of the present invention. Detailed Implementation

[0068] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0069] Reference Figure 1 The first embodiment of the present invention provides a cloud platform-based intelligent management method for sorting centers, comprising the following steps:

[0070] S11, acquire and verify the image data and input information of the abnormal package to obtain preliminary attribute information;

[0071] S12, if the preliminary attribute information is incomplete, then obtain a high-resolution scan image of the abnormal package, calculate and sort the damage index, and generate a repair priority queue.

[0072] S13, according to the repair priority queue, match similar package features from the pre-established abnormal package feature historical database to obtain inferential physical attribute information;

[0073] S14, parse the entered information to obtain a set of candidate destinations, and combine the preliminary attribute information with the inferred physical attribute information to calculate the optimal destination probability and determine the final package destination;

[0074] S15, Based on the final package destination, obtain the device status and path occupancy data in real time, and generate a preliminary allocation path;

[0075] S16, Extract peak-hour route congestion records from the pre-established historical database of sorting route congestion, optimize the preliminary allocation route, and obtain the optimized allocation route.

[0076] S17. Based on the optimized allocation path, perform load balancing analysis on the equipment, generate the final allocation plan, and send it out for execution.

[0077] In step S11, the image data and recorded information of the abnormal package are acquired and verified to obtain preliminary attribute information, including:

[0078] Real-time acquisition of image data of the abnormal package and the corresponding recorded information;

[0079] The image data is denoised and enhanced to obtain optimized image data;

[0080] Extract label content from the optimized image data to generate a label text set;

[0081] Calculate the matching degree between the tag text set and the entered information, and integrate the tag text set and the entered information based on the matching degree to generate a preliminary attribute set;

[0082] The initial attribute set is compared with the package records in the pre-established cloud platform database to verify consistency, and the initial attribute information is extracted.

[0083] In one implementation, this embodiment acquires image data and recorded information of abnormal packages through an image acquisition unit deployed on the sorting equipment and a data interface connected to a cloud platform. The recorded information is structured data containing multiple fields, while the image data is unstructured data to be processed.

[0084] For example, the image acquisition unit can be a high-resolution industrial camera, which acquires an image containing the package label. Simultaneously, the data interface extracts the associated information for the package from the cloud platform database. This information is: {Package Number: “12345678”, Recipient: “Zhang Wei”, Address: “XX Road, Pudong New Area, Shanghai”}.

[0085] It should be noted that since the acquired raw image data may contain noise caused by equipment vibration or uneven lighting, this embodiment processes the image data. This processing first applies Gaussian filtering, which calculates a weighted average value based on the distances of each pixel in the image to other pixels within a 3x3 neighborhood. This average value is then used to update the original value of the center pixel, thereby smoothing the image noise. Subsequently, histogram equalization is performed on the denoised image. By redistributing the pixel gray levels of the image, the dynamic range of the image is expanded, resulting in optimized image data.

[0086] It is worth noting that the system performs optical character recognition on the optimized image data. This operation detects the brightness and darkness patterns in the image and compares them with a character library, converting the pixel array in the image into computer-editable text characters, thereby generating a set of tag text in string form, such as "12345678 Zhang Wei".

[0087] In one implementation, the system performs a consistency assessment between the generated tag text set and the acquired input information. This assessment calculates the string similarity of corresponding fields in the two data sources (e.g., using the Levenstein distance algorithm) to obtain a quantitative matching score between 0 and 1. The system has a preset configurable matching threshold, such as 0.9. When the calculated matching score is higher than this threshold, the system performs an integration operation, which extracts the mutually verified fields from the tag text set and the input information to generate a preliminary attribute set.

[0088] In one implementation, a final consistency check is performed on the preliminary attribute set. Key fields such as the package number "12345678" from the preliminary attribute set are used as query conditions to search and compare the data in the cloud platform database's total package records. If they match, the system merges the information from the authoritative record with the preliminary attribute set to confirm and generate a highly reliable and complete set of preliminary attribute information. If they do not match, the corresponding fields in the preliminary attribute information are marked as "inconsistent," and an incomplete set of preliminary attribute information is generated.

[0089] In step S12, if the preliminary attribute information is incomplete, a high-resolution scan image of the abnormal package is obtained, the damage index is calculated and sorted, and a repair priority queue is generated, including:

[0090] The preliminary attribute information is checked for completeness. If the check result is incomplete, the high-resolution scanned image is obtained.

[0091] The high-resolution scanned image is processed using a pre-trained convolutional neural network model to obtain the coordinates of the damaged pixels on the packaged surface;

[0092] The area ratio of the damaged pixels is calculated, and the area ratio is mapped using a preset mapping rule to obtain the damage index;

[0093] The abnormal packages are sorted from high to low according to the damage index to obtain the repair priority queue;

[0094] The training process of the convolutional neural network model is as follows:

[0095] Obtain a training sample dataset containing high-resolution scanned images and corresponding labels of the actual damaged pixel locations;

[0096] Construct a convolutional neural network model and input the training sample dataset into the model to calculate the loss between the predicted damaged pixel coordinates and the actual damaged pixel location labels;

[0097] Based on the loss, the model parameters are iteratively adjusted using the backpropagation algorithm until the loss is less than a preset loss threshold, thus obtaining a trained convolutional neural network model.

[0098] It should be noted that the integrity check first checks whether there are missing or empty values ​​in the key fields (such as recipient address, contact information, and package ID) in the preliminary attribute information; and checks whether there are low confidence fields marked by S11 that do not match the records in the cloud platform database.

[0099] For example, if the "recipient address" field in the preliminary attribute information output after all processes in S11 is marked as "inconsistent" because it does not fully match in the database, then the result of this integrity check is "incomplete". When the check result is "incomplete", the system triggers the subsequent physical state analysis process and acquires the high-definition scanned image.

[0100] It should be noted that the system uses a pre-trained lightweight convolutional neural network model to perform multi-scale feature extraction on the high-definition scanned image, identify and locate unique texture patterns related to physical damage such as tearing, denting or leakage on the package surface, and finally output a set of accurate pixel coordinates representing the location of the damage.

[0101] It is worth noting that the pre-trained convolutional neural network model is obtained through supervised learning, and its training process includes the following steps:

[0102] Step one involves acquiring a training dataset containing a large number of labeled samples. This dataset construction process includes collecting thousands of high-resolution scan images of packages under different lighting and angles, and based on the physical detection results, precisely labeling all damaged areas such as tears, dents, and leaks on these images at the pixel level or border level, generating labels that correspond one-to-one with each image and contain the actual location of the damaged pixels.

[0103] Step two, construct the topology of a convolutional neural network. For example, this structure may contain three convolutional layers, each using 32 3×3 convolutional kernels, followed by a max pooling layer and a ReLU activation function.

[0104] Step three involves inputting the aforementioned training sample dataset into the constructed network model, which outputs its predicted coordinates of damaged pixels. The system calculates the error between these predicted coordinates and the true location labels of the samples using a loss function (e.g., cross-entropy loss). This error value drives the optimizer (e.g., the Adam optimizer) to adjust the weight parameters of all convolutional kernels in the network through a backpropagation algorithm. This process iterates repeatedly on the training set and evaluates its performance on an independent validation set until the model's loss value on the validation set is less than a preset loss threshold (e.g., 0.01). At this point, training terminates, and the parameters of the best-performing model are saved.

[0105] In one implementation, after obtaining the set of damaged pixel coordinates, the system calculates the total area of ​​the damaged region by counting the total number of pixels in the set. Then, this total area is divided by the total number of pixels in the image to obtain the area percentage of damaged pixels. The system uses a preset linear mapping function to convert the calculated area percentage into a standardized, quantified damage index. The formula for this linear mapping function is as follows:

[0106] D = (A d / A t )×K

[0107] Where D represents the damage index; A d A represents the total area of ​​the damaged region calculated from the set of damaged pixel coordinates; t The total pixel area of ​​the image is represented by K; K represents a preset constant used to adjust the range of the exponent.

[0108] For example, if an image is 1920x1080 pixels (total area A) t In an image with a resolution of 2,073,600, the total number of damaged pixels identified was 10,000 (i.e., A). d =i0,000), with a preset constant K of 100, then its damage index D is 0.48.

[0109] It is worth noting that the system uses a heap sort algorithm to process the damage index of all abnormal packages in the current batch. This algorithm efficiently sorts all packages in descending order according to their damage index by constructing a max-heap and extracting the maximum value one by one, ultimately obtaining the repair priority queue.

[0110] In step S13, based on the repair priority queue, similar package features are matched from a pre-established historical database of abnormal package features to obtain inferential physical attribute information, including:

[0111] The size and material of the current package are extracted from the preliminary attribute information and the entered information to obtain its physical characteristics;

[0112] Based on the repair priority queue and in conjunction with the physical characteristics, match historical package cases with similar characteristics in the abnormal package feature history database;

[0113] Based on the attribute records of the historical package cases, the missing attributes of the current package are supplemented to obtain the inferred physical attribute information.

[0114] In one implementation, this embodiment first extracts and structures a series of features that can describe the physical form of the package from the preliminary attribute information and the input information produced in S11, such as the package size specifications, outer packaging material and transportation method.

[0115] It should be noted that in this embodiment, abnormal packages are processed sequentially according to the repair priority queue generated in S12. The physical characteristics of the current package (e.g., {material: cardboard box, size: medium}) and the physical characteristics of all cases in the abnormal package feature historical database are uniformly converted into a computable feature vector. Next, the system uses preset weights (e.g., material weight 0.4, size weight 0.3, transportation method weight 0.3) and a weighted similarity function (e.g., weighted cosine similarity) to calculate the similarity score between the feature vector of the current package and each historical case in the historical database. The system sorts all calculated similarity scores in descending order and retrieves the historical package case with the highest similarity score from the sorted results.

[0116] For example, suppose the current package A123 matches historical case C789 in both "material" and "size", and its weighted similarity score is 0.4 + 0.3 = 0.7; while it matches historical case E123 only in "material", and its score is 0.4. After traversing the entire database, if case C789 has the highest score, it is taken as the most similar historical package case.

[0117] In one implementation, after matching the most similar historical package case, the system refers to the attribute records of that case to supplement the missing attributes of the current package. These missing attributes may include the physical location of the damage to the package, the specific type of damage (such as tear, dent, or leakage), and the quantified degree of damage.

[0118] For example, if the damage location recorded in the most similar historical case is "edge tear," the system will add the "edge tear" attribute value to the "damage location" field of the current package that is missing. All inferred physical attributes are then structured and integrated to form the inferred physical attribute information.

[0119] It's worth noting that for missing continuous numerical attributes like the degree of damage, the system can use linear interpolation to supplement them. This method retrieves multiple highly similar historical cases, calculates the average of the "degree of damage" field in these cases, and uses this average as supplementary data to fill in the corresponding field of the current package.

[0120] In step S14, the entered information is parsed to obtain a set of candidate destinations. The optimal destination probability is calculated by combining the preliminary attribute information and the inferred physical attribute information to determine the final package destination, including:

[0121] The semantic structure of the entered information is analyzed, and keywords related to the address are extracted to obtain the candidate destination set;

[0122] For each candidate destination in the candidate destination set, at least one business feature is extracted from the address, estimated timeliness, or path cost corresponding to the candidate destination, and a multi-dimensional feature vector is constructed by combining the preliminary attribute information and the inferred physical attribute information.

[0123] The multidimensional feature vector is compared with the feature vector of historical successful delivery cases in the pre-established historical successful delivery database to calculate the similarity and find the k nearest neighbor historical cases for each candidate destination.

[0124] The frequency of each destination in the k nearest neighbor historical cases is counted, and the frequency of occurrence is used as the basic score. Combined with the preset sorting rules, the candidate destination with the highest score is determined as the final package destination.

[0125] In one implementation, this embodiment performs natural language processing operations on the input information obtained in S11, including: preprocessing the input text string, including removing irrelevant symbols and formatting; performing word segmentation on the preprocessed text to divide the text into independent word sequences; and performing part-of-speech tagging and named entity recognition on the word sequences to identify nouns, verbs, and proper nouns with geographical location attributes, i.e., location entities.

[0126] For example, if the entered information is a text "Sent to Haidian District, Beijing, please prioritize delivery", the system first segments it into ["sent to", "Beijing", "Haidian District", ",", "please", "priority", "delivery"]. Through named entity recognition, the system identifies "Beijing" and "Haidian District" as location entities. The system extracts all identified location entities and, according to preset administrative division rules (e.g., keywords must contain units such as "city" or "district"), filters and combines the keyword set to obtain the candidate destination set.

[0127] It should be noted that after obtaining the set of candidate destinations, the system will construct an independent multidimensional feature vector for each candidate destination for subsequent quantitative decision-making. The steps for constructing the multidimensional feature vector are as follows: normalizing the package weight, volume, and damage index extracted from the preliminary attribute information and the inferred physical attribute information; extracting the business features of each candidate destination, such as the delivery distance, estimated delivery time, and path cost that combines distance and time; and concatenating all the above numerical features in a preset order to obtain the multidimensional feature vector.

[0128] For example, the preset order can prioritize the physical characteristics of the package itself and prioritize the business characteristics of the destination, forming a sequence such as [package weight, package volume, damage index, delivery distance, estimated time, route cost].

[0129] It's worth noting that the system uses the K-Nearest Neighbors (KNN) algorithm to determine the final package destination. For each candidate destination's multi-dimensional feature vector, a K-Nearest Neighbor search is performed on historical successful delivery cases. This search calculates the Euclidean distance between the current vector and each case vector in the historical database, finding the k most similar historical cases (i.e., those with the smallest Euclidean distance). After finding the k nearest neighbor historical cases, the system calculates the frequency of each candidate destination in the actual destinations of these k cases. This frequency is used as a base score, reflecting the probability of the candidate destination. The historical successful delivery cases are obtained from a pre-built database containing a massive number of historical successful delivery cases.

[0130] For example, suppose there are two candidate destinations, A and B, and the k value is set to 10. The system finds 10 nearest neighbor historical cases for A and finds that 9 of them actually lead to A, so A's base score is 0.9. However, among the 10 nearest neighbors found for B, only 2 actually lead to B, so B's base score is 0.2.

[0131] In another implementation, the system also incorporates a weighted adjustment to the base score based on preset sorting rules. For example, for urgent orders, the weight of candidate destinations with shorter estimated delivery times is increased. Ultimately, the candidate destination with the highest overall score is determined as the final package destination.

[0132] In step S15, based on the final package destination, real-time device status and path occupancy data are acquired to generate a preliminary allocation path, including:

[0133] Obtain the preset path topology model of the sorting center, as well as the real-time status of the equipment and the path occupancy data;

[0134] Based on the path topology model, and combined with the device status and the path occupancy data, path optimization calculation is performed to obtain an initial path scheme;

[0135] The initial path scheme is subjected to path conflict detection and adjustment to obtain the preliminary allocation path.

[0136] In one implementation, after determining the final destination of the package, this embodiment plans a preliminary physical transport path for it within the sorting center. This process first acquires two types of core data: a static path topology model and dynamic real-time operational data.

[0137] It should be noted that the preset path topology model is a pre-built digital map describing the internal physical layout of the sorting center. This model includes information on all sorting nodes (such as intersections, scanners, and sorting machines) and the paths connecting these nodes (such as conveyor belts), and assigns fixed attributes, such as physical length, to each path. The real-time equipment status and path occupancy data are collected in real-time by various sensors deployed within the sorting center (such as conveyor belt speed monitors and path node occupancy sensors), reflecting whether each piece of equipment is operating normally and the congestion level of each path at any given moment.

[0138] It is worth noting that the system performs path optimization calculations based on the aforementioned path topology model and in conjunction with real-time device status and path occupancy data. According to a preset cost function, the system calculates the passage cost (a dimensionless abstract value used for quantitative evaluation of path quality, not involving any economic cost) for each segment of all feasible paths in the path topology model. In one implementation, this cost function can be defined as a weighted sum function, as shown in the following example:

[0139] C seg =w d ×L+w t ×T dyn

[0140] Among them, C seg T represents the total cost of a path segment; L represents the fixed physical length of the path segment defined in the path topology model; T dyn This represents the real-time dynamic travel time through this segment, determined by the path occupancy data; w d with w t These are preset weighting coefficients. It should be noted that these weighting coefficients are configured as inverse units with normalizing properties, for example, w. d The unit can be m -1 w t The unit can be s -1 These two weighting coefficients are used to adjust the importance of physical length and dynamic time in the total cost calculation, respectively.

[0141] The system calculates the total cost of a complete path by summing the travel costs of all segments. Using computational methods such as traversal or heuristic search, the system compares the total costs of all feasible paths and selects the path with the lowest total cost as the initial path solution.

[0142] For example, suppose a package needs to be delivered from point A to point C. The path topology model shows two possible paths: Path 1 (A→B→C) with a total length of 50 meters, and Path 2 (A→D→C) with a total length of 60 meters. During periods of low load, the system will select Path 1 as the initial path option. However, if the current real-time path occupancy data shows that the A→B segment is normal, but the B→C segment is slow due to package backlog, with its real-time travel time being much longer than that of the A→D→C segment, then the optimization calculation will determine that Path 2 has a lower total cost and will select it as the initial path option.

[0143] In another implementation, the system also performs a final path conflict detection and adjustment on the generated initial path plan. The planned path resources (such as a section of conveyor belt) are compared again with the latest real-time occupancy data. If the resource is found to be occupied by another higher-priority task, a path conflict is identified. At this point, the system adjusts the path plan to avoid the conflict, for example, by selecting an alternative path or adjusting the time order in which tasks enter the path node. After conflict detection and adjustment, a preliminary allocation path that is feasible at the current moment is finally obtained.

[0144] In step S16, peak-hour route congestion records are extracted from a pre-established historical database of sorting route congestion, and the initial allocation route is optimized to obtain an optimized allocation route, including:

[0145] Filter and extract peak-hour route congestion records from the historical congestion database of the sorting route;

[0146] Identify the set of congested road segments in the preliminary dispatch route that overlap with the peak-hour route congestion records;

[0147] For the set of congested road segments, based on preset path adjustment rules and allocation weights, path replanning calculations are performed to obtain the optimized allocation path.

[0148] In one implementation, this embodiment utilizes historical operational data to optimize the initial allocation path. This optimization process first involves executing a database query command to filter and extract relevant congestion records from a pre-established historical database of sorting path congestion. It should be noted that the construction process of the historical database of sorting path congestion may include: the system periodically recording and storing operational data such as the real-time occupancy rate, package throughput rate, and average travel time of each sorting path node; identifying time periods and road segments with operational efficiency below a preset efficiency threshold and marking these events as congestion events; and storing the key parameters of each congestion event, such as the road segment ID where congestion occurred, the specific time and duration of the congestion, and the average travel time during the congestion period, in a structured manner into the database, ultimately forming the historical database of sorting path congestion.

[0149] For example, this extraction operation can be accomplished by executing a database query command. For instance, the system can execute an SQL query with the filter conditions set to "Time: Weekday peak hours" and "Road segment capacity: Below a preset threshold," thereby retrieving all path congestion records that meet the conditions from the historical database. Each congestion record can contain information such as road segment ID, congestion period, and historical average travel time.

[0150] It should be noted that after obtaining peak-hour route congestion records, the system compares these records with the preliminary dispatched routes to identify potential congestion risk points within the routes. This comparison process can be performed within a Geographic Information System (GIS) environment, using spatial query functions to determine whether the route nodes traversed by the preliminary dispatched routes overlap with the road segment IDs and time periods in the congestion records. All identified overlapping road segments are aggregated into a congested road segment set.

[0151] It is worth noting that the system dynamically modifies the cost parameters of path optimization calculation and performs path replanning calculation for the set of congested road segments based on a set of preset path adjustment rule bases and corresponding allocation weights.

[0152] For example, the route adjustment rule base may contain a rule like this: {“Rule ID”: R01, “Trigger Condition”: “Road segment B→C congested”, “Adjustment Strategy”: “Activate alternative route D→E”}. The allocation weights corresponding to this rule are a set of numerical coefficients used to modify the cost function. When the system identifies that road segment B→C in the initial allocation path belongs to the set of congested road segments, rule R01 is triggered. The system then retrieves the corresponding allocation weights from the rule base when performing route replanning calculations. For example, it dynamically increases the toll cost weight coefficient of road segment B→C to 1.5, while keeping the weight coefficients of alternative routes and other smooth routes at 1.0. The system applies this adjusted weight to the cost function described in S15 and recalculates the total cost of all feasible paths. Since the calculation cost of congested road segments is artificially and significantly increased, the result of the route optimization calculation will naturally select a new path with the lowest total cost that bypasses the congested road segment. This new path is the final optimized allocation path.

[0153] In step S17, based on the optimized allocation path, load balancing analysis is performed on the equipment to generate a final allocation plan and execute it, including:

[0154] Obtain the rated capacity and real-time operating data of each piece of equipment, and calculate the load rate of each piece of equipment;

[0155] The statistical coefficient of variation is calculated based on the load rate to obtain a quantitative assessment result of load balance.

[0156] Based on the quantitative evaluation results, task optimization and redistribution calculations are performed to obtain an optimized task allocation scheme;

[0157] The optimized task allocation scheme was simulated and verified to obtain the final allocation scheme that passed the verification.

[0158] The final allocation plan is converted into equipment control commands and sent to the controller for execution.

[0159] In one implementation, two types of data for the equipment required to execute the path are first obtained: one is static information obtained from the equipment ledger, such as the rated capacity of each machine; the other is dynamic operating data collected in real time, such as the actual running time and processing volume.

[0160] It should be noted that the system calculates the load rate of each machine based on the acquired data. This calculation is performed by dividing the actual processing volume of the machine over a period of time by its theoretical maximum capacity (i.e., rated capacity multiplied by the running time) during that period, resulting in a percentage value.

[0161] For example, if a sorting machine has a rated capacity of 1000 pieces / hour and has actually processed 7200 packages in the past 8 hours, its load rate is 7200 / (1000*8) = 0.9, or 90%. The system performs this calculation on all relevant machines to obtain a set of load rate data.

[0162] It is worth noting that when the average load rate of all machines is greater than zero, the system calculates the standard deviation of the load rate data based on the obtained load rate data, divides the standard deviation by the arithmetic mean of the load rate data to obtain the coefficient of variation, and uses the coefficient of variation as an evaluation result that can quantify the overall load balancing degree. When the coefficient of variation exceeds a preset balancing threshold (e.g., 0.15), or the load rate of any single machine exceeds a preset reasonable range (e.g., 0.8-1.0), the system initiates task optimization and redistribution calculations based on the evaluation result.

[0163] In one implementation, the task optimization and reallocation calculation is performed under a series of preset constraints. These constraints may include the requirement that task allocation avoid pre-set equipment downtime for maintenance (maintenance constraint), that the amount of tasks allocated to a single piece of equipment not exceed its rated capacity (capacity constraint), and that some time-sensitive packages must be processed within a specific time window (time window constraint). Under these constraints, the calculation process will reallocate some tasks originally allocated to high-load machines to low-load machines.

[0164] In another implementation, the system also performs simulation verification of the optimized task allocation scheme. This verification process involves inputting the new allocation scheme into a digital twin or simulation environment of a sorting center, simulating its performance over a future period. The system checks whether the various performance indicators in the simulation results (such as the predicted load rate of each machine, coefficient of variation, average package waiting time, etc.) meet the preset efficiency and balance targets.

[0165] It is worth noting that after the simulation verification was passed, the task allocation scheme was confirmed as the final allocation scheme. The system converts the task allocation instructions into message formats that conform to specific industrial communication protocols (such as Modbus or Profinet), and sends them to the programmable logic controllers (PLCs) and other controllers of the sorting equipment for execution, adjusting the conveyor belt speed or sorting priority, etc., to complete the final closed-loop control.

[0166] To facilitate understanding of the present invention, some preferred embodiments of the present invention will be described in further detail below.

[0167] In one implementation, the present invention provides a complete intelligent management workflow from anomaly identification to closed-loop control. Exemplarily, the workflow is as follows:

[0168] Step 1: A package with a partially soiled label enters the sorting line. After the system acquires its image and inputs the information, its preliminary attribute information is marked as "inconsistent" after the final consistency check because the matching degree between the text recognized by OCR and the input information is less than a preset threshold of 90%.

[0169] Step two: Due to incomplete preliminary attribute information, the system initiates physical state analysis. Using high-resolution scan images, a pre-trained convolutional neural network identifies dent-like damage at the corners of the package and calculates its damage index as 0.65. The package information is then added to the repair priority queue.

[0170] Step three: The system processes the package according to the queue order. It extracts the physical characteristics {material: cardboard box, size: small} from the entered information and uses a collaborative filtering algorithm to match them in the historical database of abnormal package characteristics. The results show that the historical cases most similar to these characteristics mostly have "corner crush damage" as their damage attribute. The system uses this as inferential physical attribute information to supplement the package's file.

[0171] Step four: The system initiates the destination decision-making process. Through natural language processing, the candidate destination "Chaoyang District, Beijing" is extracted from the remarks field "sent to Chaoyang District, Beijing" in the entered information. Subsequently, the system constructs a multi-dimensional feature vector from all known information (preliminary attribute information, inferred physical attribute information, business characteristics of candidate destinations, etc.) and performs K-nearest neighbor matching with the successful delivery history database. Finally, "Chaoyang District, Beijing" is determined as the final destination of the package with high confidence.

[0172] Step 5: Based on the destination, the system combines real-time path occupancy data to perform path optimization calculations and generate a preliminary allocation path of A→B→C.

[0173] Step six: When optimizing the route, the system finds from the historical congestion database that route B→C is a highly congested section during the current peak hours. It then triggers route replanning calculations, adjusting the route to the less congested A→D→C and generating an optimized allocation route.

[0174] Step seven, finally, the system analyzes the load rate of each device along the optimized path A→D→C. Calculations show that the load rate of node D has reached 95%, and the coefficient of variation exceeds the balancing threshold. The system then initiates task optimization and redistribution calculations, transferring some tasks from node D to the less loaded node E. After simulation verification, the final allocation scheme is converted into control commands and sent to the PLC for execution.

[0175] In another implementation, the system architecture of this invention can be configured differently. For example, in a large, centralized automated sorting center, all image acquisition units, sensors, execution controllers, and computing servers can be connected via wired networks such as industrial Ethernet. All calculations are performed on local edge computing nodes, ensuring minimal latency and maximum response speed. Alternatively, in a distributed network consisting of multiple small regional sorting stations, sensor data from each station can be uploaded in real-time to a central cloud server via wireless networks such as 5G. The cloud server performs complex model calculations (such as CNN recognition and KNN decision-making) and accesses historical databases, then distributes the final allocation plan generated to the local controllers of each station for execution. This approach provides greater deployment flexibility and scalability.

[0176] In summary, this invention discloses an intelligent management method for sorting centers based on a cloud platform, comprising: acquiring and verifying image data and input information of abnormal packages to obtain preliminary attribute information; if the preliminary attribute information is incomplete, acquiring high-resolution scanned images of the abnormal packages, calculating and sorting damage indices, and generating a repair priority queue; matching similar package features from a pre-established historical database of abnormal package features to obtain inferred physical attribute information; parsing the input information to obtain a set of candidate destinations, and combining the preliminary attribute information and the inferred physical attribute information to calculate the optimal destination probability and determine the final package destination; acquiring equipment status and path occupancy data in real time based on the final package destination to generate a preliminary allocation path; extracting peak-period path congestion records from a pre-established historical database of sorting path congestion, optimizing the preliminary allocation path to obtain an optimized allocation path; and performing load balancing analysis on the equipment based on the optimized allocation path to generate a final allocation plan and issue it for execution. This invention solves the technical problem of low efficiency in handling abnormal packages in existing technologies by constructing an intelligent closed-loop control process that integrates multi-dimensional information intelligent completion, dynamic path optimization, and load balancing scheduling, thereby improving the automation level and operational efficiency of sorting centers.

[0177] Reference Figure 2 The second embodiment of the present invention provides a cloud-based intelligent management system for sorting centers, comprising:

[0178] The data acquisition module is used to acquire and verify the image data and input information of the abnormal package to obtain preliminary attribute information;

[0179] The damage assessment module is used to acquire high-resolution scan images of the abnormal package if the preliminary attribute information is incomplete, calculate and sort the damage index, and generate a repair priority queue.

[0180] The attribute completion module is used to match similar package features from a pre-established historical database of abnormal package features according to the repair priority queue, and obtain inferential physical attribute information.

[0181] The destination decision module is used to parse the input information, obtain a set of candidate destinations, and combine the preliminary attribute information with the inferred physical attribute information to calculate the optimal destination probability and determine the final package destination.

[0182] The preliminary route generation module is used to generate a preliminary allocation route based on the final package destination, by acquiring device status and route occupancy data in real time.

[0183] The route optimization module is used to extract peak-hour route congestion records from a pre-established historical database of sorting route congestion, and optimize the initial allocation route to obtain an optimized allocation route.

[0184] The load balancing and scheduling module is used to perform load balancing analysis on the devices based on the optimized allocation path, generate the final allocation plan, and send it out for execution.

[0185] It should be noted that the cloud-based intelligent management system for sorting centers provided in this embodiment of the invention is used to execute all process steps of the cloud-based intelligent management method for sorting centers described in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0186] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a cloud-based intelligent management program for sorting centers. When the processor executes the computer program, it implements the steps described in the various cloud-based intelligent management method embodiments for sorting centers, for example... Figure 1 The step S11 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above-described device embodiments, such as the data acquisition module.

[0187] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0188] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0189] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0190] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0191] If the modules / units integrated in the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0192] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0193] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A cloud-based intelligent management method for sorting centers, characterized in that, include: Acquire and verify the image data and entered information of the abnormal package to obtain preliminary attribute information; If the preliminary attribute information is incomplete, a high-resolution scan image of the abnormal package is obtained, the damage index is calculated and sorted, and a repair priority queue is generated. Based on the repair priority queue, similar package features are matched from a pre-established historical database of abnormal package features to obtain inferential physical attribute information; The entered information is parsed to obtain a set of candidate destinations. The optimal destination probability is calculated by combining the preliminary attribute information and the inferred physical attribute information, and the final destination of the package is determined. Based on the final package destination, real-time device status and path occupancy data are obtained to generate a preliminary allocation path; Peak-hour route congestion records are extracted from a pre-established historical database of sorting route congestion, and the preliminary allocation route is optimized to obtain an optimized allocation route. Based on the optimized allocation path, load balancing analysis is performed on the equipment, a final allocation plan is generated, and the plan is sent out for execution.

2. The intelligent management method for sorting centers based on a cloud platform according to claim 1, characterized in that, The process of acquiring image data of the abnormal package and verifying it with the entered information to obtain preliminary attribute information includes: Real-time acquisition of image data of the abnormal package and the corresponding recorded information; The image data is denoised and enhanced to obtain optimized image data; Extract label content from the optimized image data to generate a label text set; Calculate the matching degree between the tag text set and the entered information, and integrate the tag text set and the entered information based on the matching degree to generate a preliminary attribute set; The initial attribute set is compared with the package records in the pre-established cloud platform database to verify consistency, and the initial attribute information is extracted.

3. The intelligent management method for sorting centers based on a cloud platform according to claim 1, characterized in that, If the preliminary attribute information is incomplete, then a high-resolution scan image of the abnormal package is obtained, the damage index is calculated and sorted, and a repair priority queue is generated, including: The preliminary attribute information is checked for completeness. If the check result is incomplete, the high-resolution scanned image is obtained. The high-resolution scanned image is processed using a pre-trained convolutional neural network model to obtain the coordinates of the damaged pixels on the packaged surface; The area ratio of the damaged pixels is calculated, and the area ratio is mapped using a preset mapping rule to obtain the damage index; The abnormal packages are sorted from high to low according to the damage index to obtain the repair priority queue; The training process of the convolutional neural network model is as follows: Obtain a training sample dataset containing high-resolution scanned images and corresponding labels of the actual damaged pixel locations; Construct a convolutional neural network model and input the training sample dataset into the model to calculate the loss between the predicted damaged pixel coordinates and the actual damaged pixel location labels; Based on the loss, the model parameters are iteratively adjusted using the backpropagation algorithm until the loss is less than a preset loss threshold, thus obtaining a trained convolutional neural network model.

4. The intelligent management method for sorting centers based on a cloud platform according to claim 1, characterized in that, The step of matching similar package features from a pre-established historical database of abnormal package features according to the repair priority queue to obtain inferential physical attribute information includes: The size and material of the current package are extracted from the preliminary attribute information and the entered information to obtain its physical characteristics; Based on the repair priority queue and in conjunction with the physical characteristics, match historical package cases with similar characteristics in the abnormal package feature history database; Based on the attribute records of the historical package cases, the missing attributes of the current package are supplemented to obtain the inferred physical attribute information.

5. The intelligent management method for sorting centers based on a cloud platform according to claim 1, characterized in that, The process of parsing the entered information to obtain a set of candidate destinations, and combining the preliminary attribute information with the inferred physical attribute information to calculate the optimal destination probability and determine the final package destination includes: The semantic structure of the entered information is analyzed, and keywords related to the address are extracted to obtain the candidate destination set; For each candidate destination in the candidate destination set, at least one business feature is extracted from the address, estimated timeliness, or path cost corresponding to the candidate destination, and a multi-dimensional feature vector is constructed by combining the preliminary attribute information and the inferred physical attribute information. The multidimensional feature vector is compared with the feature vector of historical successful delivery cases in the pre-established historical successful delivery database to calculate the similarity and find the k nearest neighbor historical cases for each candidate destination. The frequency of each destination in the k nearest neighbor historical cases is counted, and the frequency of occurrence is used as the basic score. Combined with the preset sorting rules, the candidate destination with the highest score is determined as the final package destination.

6. The intelligent management method for sorting centers based on a cloud platform according to claim 1, characterized in that, The step of obtaining device status and path occupancy data in real time based on the final package destination and generating a preliminary allocation route includes: Obtain the preset path topology model of the sorting center, as well as the real-time status of the equipment and the path occupancy data; Based on the path topology model, and combined with the device status and the path occupancy data, path optimization calculation is performed to obtain an initial path scheme; The initial path scheme is subjected to path conflict detection and adjustment to obtain the preliminary allocation path.

7. The intelligent management method for sorting centers based on a cloud platform according to claim 1, characterized in that, The step of extracting peak-hour route congestion records from a pre-established historical database of sorting route congestion and optimizing the initial allocation route to obtain an optimized allocation route includes: Filter and extract peak-hour route congestion records from the historical congestion database of the sorting route; Identify the set of congested road segments in the preliminary dispatch route that overlap with the peak-hour route congestion records; For the set of congested road segments, based on preset path adjustment rules and allocation weights, path replanning calculations are performed to obtain the optimized allocation path.

8. The intelligent management method for sorting centers based on a cloud platform according to claim 1, characterized in that, The step of performing load balancing analysis on the equipment based on the optimized allocation path, generating a final allocation plan, and issuing it for execution includes: Obtain the rated capacity and real-time operating data of each piece of equipment, and calculate the load rate of each piece of equipment; The statistical coefficient of variation is calculated based on the load rate to obtain a quantitative assessment result of load balance. Based on the quantitative evaluation results, task optimization and redistribution calculations are performed to obtain an optimized task allocation scheme; The optimized task allocation scheme was simulated and verified to obtain the final allocation scheme that passed the verification. The final allocation plan is converted into equipment control commands and sent to the controller for execution.

9. A cloud-based intelligent management system for sorting centers, characterized in that, include: The data acquisition module is used to acquire and verify the image data and input information of the abnormal package to obtain preliminary attribute information; The damage assessment module is used to acquire high-resolution scan images of the abnormal package if the preliminary attribute information is incomplete, calculate and sort the damage index, and generate a repair priority queue. The attribute completion module is used to match similar package features from a pre-established historical database of abnormal package features according to the repair priority queue, and obtain inferential physical attribute information. The destination decision module is used to parse the input information, obtain a set of candidate destinations, and combine the preliminary attribute information with the inferred physical attribute information to calculate the optimal destination probability and determine the final package destination. The preliminary route generation module is used to generate a preliminary allocation route based on the final package destination, by acquiring device status and route occupancy data in real time. The route optimization module is used to extract peak-hour route congestion records from a pre-established historical database of sorting route congestion, and optimize the initial allocation route to obtain an optimized allocation route. The load balancing and scheduling module is used to perform load balancing analysis on the devices based on the optimized allocation path, generate the final allocation plan, and send it out for execution.

10. An electronic device, characterized in that, The system includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the cloud-based intelligent management method for sorting centers as described in any one of claims 1 to 8.

Images