Express courier station information data processing method and equipment

Abstract

The invention relates to an express delivery courier station information data processing method and device, and the method comprises the following steps: S1, collecting historical parcel data, the data comprising traffic records, time distribution and user behavior information; s2, performing deep mining on the data by using a machine learning model, and predicting package traffic, peak time periods and user preferences in a future preset time period; s3, dynamically planning manpower resources, storage space and equipment configuration of the courier station according to a prediction result; and S4, based on a user behavior analysis result, generating a personalized parcel pickup suggestion, and pushing the personalized parcel pickup suggestion to the user terminal.The method can accurately predict the parcel traffic, the peak period and the user preference in the future preset period by collecting historical parcel data and performing deep mining and analysis by using the machine learning model. Based on the prediction results, human resources, storage space and equipment configuration of the courier station can be dynamically planned, so that the package flow pressure in the peak period is effectively dealt with, and the operation efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of express delivery stations, and specifically to a method and device for processing express delivery station information data. Background Art

[0002] In the field of express delivery logistics, with the booming development of e-commerce, express delivery stations, as important nodes in the logistics chain, are undertaking increasing parcel processing and distribution tasks. Most traditional express delivery station management methods rely on manual operations and simple information systems. These data processing methods are inefficient and difficult to cope with the parcel flow pressure during peak periods. At the same time, due to the lack of in-depth analysis and prediction of user behavior, express delivery stations often have difficulty providing personalized services, resulting in a poor user experience.

[0003] According to the publicly disclosed patent 202110656913.9, an intelligent access information management system applied to an express delivery station includes: a goods information acquisition unit for scanning and identifying the goods information entering and leaving the access window; a first image capture unit for acquiring the first image information of the goods to be stored; a second image capture unit for acquiring the second image information of the goods stored on the shelf in real time; a data processing unit for calculating the size and volume of the goods to be stored through a machine vision algorithm based on the first image information; identifying all standard storage space positions on the shelf where no goods are placed and the remaining unoccupied space information in all standard storage space positions where goods have been stored through a machine vision algorithm based on the second image information, and planning a storage space that can match the size and volume of the goods to be stored from the remaining unoccupied space in each standard storage space position as the storage position for the goods to be stored. It realizes the intelligent warehousing management of express delivery stations and improves the utilization rate of shelf space.

[0004] However, in actual operation, express delivery stations need to process a large number of parcels, generating a vast amount of operation data, including the throughput of goods, access time, user pick-up habits, etc. If these data cannot be deeply analyzed and utilized, express delivery stations will have difficulty dynamically adjusting operation strategies based on historical data and real-time situations, which may lead to low operation efficiency and resource waste. Moreover, there is a lack of comprehensive analysis of user behavior and the provision of personalized services. In the modern logistics system, the user experience is crucial. However, traditional solutions have not fully paid attention to the needs and behavior patterns of users. For example, the system does not provide personalized pick-up services according to users' pick-up habits and preferences, nor does it predict future pick-up needs based on users' behavior data, making it difficult to provide a more efficient and convenient pick-up experience for users. Therefore, new technical solutions need to be designed to solve this problem. Summary of the Invention

[0005] The object of the present invention is to overcome the deficiencies of the prior art, meet the actual needs, and provide a method and device for processing express delivery station information data, so as to solve the problem that current express delivery stations need to process a large number of packages, generating a huge amount of operation data, including the throughput of goods, storage and retrieval time, users' pick-up habits, etc. If these data cannot be deeply analyzed and utilized, it is difficult for express delivery stations to dynamically adjust operation strategies according to historical data and real-time situations, which may lead to low operation efficiency and resource waste. Moreover, there is a lack of comprehensive analysis of user behavior and provision of personalized services. In the modern logistics system, the user experience is crucial. However, traditional solutions have not fully paid attention to users' needs and behavior patterns. For example, the system does not provide personalized pick-up services according to users' pick-up habits and preferences, nor does it predict future pick-up needs based on users' behavior data, thus making it difficult to provide users with a more efficient and convenient pick-up experience.

[0006] To achieve the object of the present invention, the technical solution adopted by the present invention is: designing a method for processing express delivery station information data, including the following steps:

[0007] S1. Collect historical package data, including traffic records, time distribution, and user behavior information;

[0008] S2. Use a machine learning model to deeply mine the data to predict the package traffic, peak period, and user preferences in a preset future period;

[0009] S3. Dynamically plan the human resources, storage space, and equipment configuration of the station according to the prediction results;

[0010] S4. Generate personalized pick-up suggestions based on the user behavior analysis results and push them to the user terminal.

[0011] Preferably, the prediction of future package traffic specifically includes:

[0012] Adopt a time series analysis algorithm to perform trend fitting on historical traffic data and correct the prediction results in combination with seasonal characteristics;

[0013] Introduce external variable data, including weather and promotion activity information, and optimize the traffic prediction accuracy through a random forest model.

[0014] Preferably, the prediction of the peak period is achieved through the following steps:

[0015] Extract the peak time window of daily package arrivals in historical data and construct a Gaussian mixture model for clustering;

[0016] Associate the user pick-up delay duration with the peak period congestion degree, and dynamically adjust the start time and duration of the prediction window.

[0017] Preferably, the user preference analysis includes:

[0018] Based on the user's pick-up time, collection selection, and complaint records, use the collaborative filtering algorithm to construct a preference profile;

[0019] Cluster and group users, and optimize the generation logic of the recommendation strategy for different groups.

[0020] Preferably, the dynamic programming resource allocation specifically includes:

[0021] Before the predicted peak period, increase the temporary storage shelves and staff scheduling volume proportionally;

[0022] According to the proportion of the "instant pick-up" label in the user preferences, pre-allocate the fast channel resources.

[0023] Preferably, the generation of personalized pick-up suggestions includes:

[0024] Analyze the historical pick-up time distribution of users and calculate the standard deviation of their habitual time periods;

[0025] When the congestion index of the predicted period exceeds the threshold, preferentially recommend the off-peak pick-up plan to users with a standard deviation greater than the preset value.

[0026] Preferably, it further includes a real-time optimization mechanism:

[0027] Real-time collect the package accumulation data in the post station through Internet of Things devices;

[0028] When the deviation between the actual traffic and the predicted value exceeds 20%, trigger the dynamic resource reallocation algorithm and adjust the task priorities of the staff.

[0029] Preferably, it further includes an exception handling module:

[0030] Monitor the packages that have not been picked up for a long time, and establish a risk scoring model in combination with the user's historical behavior;

[0031] When the score exceeds the critical value, automatically switch to phone notification and generate a pending work order.

[0032] Preferably, it further includes an effectiveness evaluation stage:

[0033] Compare the package processing efficiency indicators before and after the resource allocation adjustment;

[0034] Use A / B testing to verify the user adoption rate of different recommendation strategies, and update the optimal solution to the model training data set.

[0035] A device adopting the express delivery station information data processing method, including:

[0036] A data collection module for real-time obtaining the inbound and outbound records of the post station and the user operation logs;

[0037] A prediction engine, equipped with an LSTM neural network to process time-series data and output prediction results of traffic and peak periods.

[0038] A resource scheduling center that generates a shift schedule and a space allocation plan based on the prediction results.

[0039] A recommendation system that matches the optimal pick-up strategy based on the user profile and issues it through a multi-channel push interface.

[0040] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0041] 1. By collecting historical parcel data and using machine learning models for in-depth mining and analysis, the present invention can accurately predict the parcel traffic, peak periods, and user preferences in a preset future period. Based on these prediction results, the present invention can dynamically plan the human resources, storage space, and equipment configuration of the courier station, thus effectively coping with the parcel traffic pressure during peak periods and improving the operation efficiency.

[0042] 2. The present invention can generate personalized pick-up suggestions based on the analysis results of user behavior and issue them to the user terminal through a multi-channel push interface, providing a convenient and efficient pick-up experience.

[0043] 3. Through the exception handling module and the effect evaluation stage, the present invention can monitor and handle exceptions in real time, and continuously optimize and adjust the resource configuration and recommendation strategy to ensure the efficient and stable operation of the courier station. Brief Description of the Drawings

[0044] Figure 1 is a schematic flow chart of the present invention;

[0045] Figure 2 is a schematic diagram of the equipment of the present invention. Detailed Embodiments

[0046] The present invention will be further described below with reference to the drawings and embodiments:

[0047] A method for processing information data of a courier station, see Figure 1 , including the following steps:

[0048] S1. Data collection

[0049] Export historical parcel data from the system of the courier station. These data should cover traffic records (such as the number of parcels per day, the number of parcels arriving per hour, etc.), time distribution (such as peak periods and off-peak periods of parcel arrival), and user behavior information (such as user pick-up time, whether to choose agency collection, complaint records, etc.) within a past period of time;

[0050] S2. Data preprocessing

[0051] Clean the collected historical data, remove invalid data such as outliers and duplicate values, and ensure the accuracy and integrity of the data;

[0052] S3. Machine learning model training and prediction

[0053] Use machine learning models (such as LSTM neural network, random forest, etc.) to deeply mine the preprocessed data. First, use time series analysis algorithm to fit the trend of traffic data, and correct the prediction results combined with seasonal characteristics to obtain the predicted values of parcel traffic in the preset future period (such as next week, next month);

[0054] Then, introduce external variable data (such as weather forecast, e-commerce platform promotion activity information, etc.), and further optimize the traffic prediction accuracy through the random forest model;

[0055] At the same time, extract the peak time window of daily parcel arrivals in historical data, construct a Gaussian mixture model for clustering analysis, associate the user pick-up delay duration with the peak period congestion degree, and dynamically adjust the start time and duration of the prediction window to predict the future peak period;

[0056] S4. User preference analysis

[0057] Based on information such as user pick-up time, agency collection selection, and complaint records, use collaborative filtering algorithm to construct a user preference profile and cluster and group users;

[0058] S5. Dynamic programming resource allocation

[0059] According to the predicted future parcel traffic and peak period, dynamically plan the human resources (such as increasing the scheduling volume of temporary staff), storage space (such as adding temporary storage shelves), and equipment configuration (such as pre-assigning fast channel resources) of the post station;

[0060] S6. Generate personalized pick-up suggestions

[0061] Analyze the historical pick-up time distribution of users, calculate the standard deviation of their habitual time periods. When the congestion index in the predicted period exceeds the threshold, preferentially recommend off-peak pick-up solutions to users with a standard deviation greater than the preset value, and push personalized pick-up suggestions through user terminals (such as mobile APPs, text messages, etc.);

[0062] S7. Real-time optimization and exception handling

[0063] Real-time collect the parcel accumulation data in the post station through Internet of Things devices. When the deviation between the actual traffic and the predicted value exceeds 20%, trigger the dynamic resource reallocation algorithm to adjust the task priorities of the staff;

[0064] Monitor the situation where packages are not picked up for a long time, establish a risk scoring model based on the user's historical behavior. When the score exceeds the critical value, automatically switch to phone notification for the user and generate a work order to be processed;

[0065] S8. Effect Evaluation and Model Optimization

[0066] Compare the package processing efficiency indicators (such as average pick-up waiting time, package processing volume, etc.) before and after the resource allocation adjustment to evaluate the optimization effect;

[0067] Use A / B testing to verify the user adoption rate of different recommendation strategies, update the optimal solution to the model training dataset, and continuously optimize the machine learning model and recommendation strategies.

[0068] Through the above steps, the express delivery station can achieve a comprehensive optimization of the operation process, improve service quality and user satisfaction.

[0069] A device adopting the express delivery station information data processing method, see Figure 2 , including:

[0070] A data acquisition module, used to obtain the inbound and outbound records of the station and the user operation logs in real time;

[0071] A prediction engine, equipped with an LSTM neural network to process time series data and output the prediction results of traffic and peak periods;

[0072] A resource scheduling center, which generates a work schedule and a space allocation plan according to the prediction results;

[0073] A recommendation system, which matches the optimal pick-up strategy based on the user profile and issues it through a multi-channel push interface.

[0074] Example 1: LSTM Time Series Prediction Model

[0075] Application scenario: Package traffic prediction

[0076] Data description:

[0077] Input data: The number of packages arriving per hour in the past 30 days (a total of 720 time series data)

[0078] Feature dimensions: Timestamp (hour), package volume, whether it is a working day (0 / 1)

[0079] Target value: Prediction of the number of packages per hour in the next 7 days

[0080] Model parameters:

[0081] Network structure: Double-layer LSTM (64 hidden layer nodes) + fully connected layer

[0082] Training parameters: Sliding window length = 24 hours, batch size = 32, number of training epochs = 100, learning rate = 0.001

[0083] Loss function: MAE (Mean Absolute Error)

[0084] Implementation effect:

[0085] Prediction error: MAE = 15.2 pieces / hour (validation set)

[0086] Peak period prediction accuracy: 89% (within an error of ±30 minutes)

[0087] Example data: The predicted package volume from 14:00 to 15:00 on the 3rd day is 235 pieces, and the actual value is 228 pieces

[0088] Comparative advantage:

[0089] Compared with the traditional ARIMA model (comparative example 1), the MAE = 28.7 pieces. The LSTM captures the non - linear time - series features, and the error is reduced by 47%.

[0090] Example 2: Random forest multi - factor prediction model

[0091] Application scenario: Traffic correction during the promotion period

[0092] Data description:

[0093] Input features:

[0094] Historical traffic mean (same period in the past 7 days)

[0095] Weather index (0 - 10, encoded by temperature / rainfall)

[0096] Promotion activity intensity (e - commerce platform activity level 1 - 5)

[0097] Target value: Percentage increase in traffic on the promotion day

[0098] Model parameters:

[0099] Number of trees = 200, maximum depth = 10, minimum number of samples per leaf = 5

[0100] Feature importance ranking: Promotion intensity (weight 0.52) > Weather index (0.31) > Historical mean (0.17)

[0101] Implementation effect:

[0102] Prediction for the Double Eleven promotion day: Traffic growth of 78% (actual growth of 82%)

[0103] Average error for regular promotion days: ±6.5%

[0104] Example data: Input [historical average = 200 pieces, weather index = 8 (sunny day), promotion level = 4], output predicted increase of 65%

[0105] Comparative advantage:

[0106] Compared with the linear regression model (R 2 = 0.71), the R of the random forest 2 = 0.89, improving the modeling ability of the multi-variable coupling relationship.

[0107] Example 3: Gaussian Mixture Model (GMM) peak period clustering

[0108] Application scenario: Dynamically adjust peak hours

[0109] Data description:

[0110] Input data: Daily parcel arrival time distribution in the past 90 days (arrival density of 1440 minutes / day)

[0111] Clustering dimensions: Peak start time, peak duration, peak intensity

[0112] Model parameters:

[0113] Number of components = 3 (corresponding to morning / noon / evening peaks)

[0114] Covariance type = full covariance, maximum number of iterations = 500

[0115] Implementation effect:

[0116] Clustering results:

[0117] Morning peak (8:30 - 10:00, average intensity of 145 pieces / hour)

[0118] Noon peak (12:00 - 13:30, average intensity of 112 pieces / hour)

[0119] Evening peak (18:00 - 19:30, average intensity of 178 pieces / hour)

[0120] Example of dynamic adjustment: The evening peak on rainy days is extended to 20:00 (verified accuracy of 92%)

[0121] Comparative advantage:

[0122] Compared with K-means clustering (Comparative Example 2), the probability membership model of GMM reduces the partitioning error of overlapping periods by 21%.

[0123] Example 4: Collaborative filtering user preference analysis

[0124] Application scenario: Personalized pick-up recommendation

[0125] Data description:

[0126] User-behavior matrix:

[0127] Rows: 5,000 users

[0128] Columns: Pick-up time period (6 periods), collection on behalf selection (yes / no), complaint type (3 types)

[0129] Sparsity: 82% (most users have no complaint records)

[0130] Model parameters:

[0131] Number of latent factors = 20, learning rate = 0.01, regularization coefficient = 0.05

[0132] Similarity calculation: Cosine similarity (number of neighbors = 50)

[0133] Implementation effect:

[0134] Recommendation adoption rate: 38% (random recommendation in the control group is 12%)

[0135] Typical case: User A's historical preference is "evening collection on behalf" → recommended "pick up on behalf from 19:00 to 20:00"

[0136] Complaint rate reduction: The average weekly complaints after optimization decreased from 4.3 times to 1.7 times

[0137] Comparative advantages:

[0138] The rule-based recommendation system (comparative example 3) only relies on the most recent behavior, and the adoption rate is less than 20%.

[0139] Comparative example 1: ARIMA time series prediction

[0140] Model parameters: p = 2, d = 1, q = 1 (selected by AIC criterion)

[0141] Prediction results:

[0142] MAE on normal days = 28.7 pieces / hour

[0143] The error on promotional days reaches ±35% (due to the inability to introduce external variables)

[0144] Defects: Unable to handle external factors such as weather / promotions, and the prediction lags significantly during peak periods. Comparative example 2: Rule-based peak period division rule definition: Fixed division of early (9:00 - 11:00) and late (17:00 - 19:00) peaks. Implementation problems:

[0145] The actual evening rush hour on rainy days is delayed until 20:00, and the rule system fails to adjust, resulting in resource waste: 15% more manpower is allocated during fixed periods, but the utilization rate is only 73%.

[0146] Error comparison: The accuracy rate of peak period determination is only 68% (vs 89% of GMM).

[0147] Comparative example 3: Rule recommendation based on recent behavior

[0148] Application scenario: Pickup strategy push

[0149] Rule definition:

[0150] 1. Time recommendation rule:

[0151] If the user's last 3 pickup times are all between [18:00 - 20:00], then recommend the same time period; otherwise, default to recommend the idle period of the post station (calculated by historical average traffic).

[0152] 2. Consignment recommendation rule:

[0153] If the user's consignment ratio in the past week > 50%, then forcefully recommend the consignment service; otherwise, do not actively push consignment suggestions

[0154] Implementation data:

[0155] Test samples: The same 5000 - user dataset as in Example 4

[0156] Feature dependence: Only use the binary statistics of pickup time and consignment frequency

[0157] Implementation effect:

[0158] Recommendation adoption rate: 19.6% (vs 38% in Example 4)

[0159] Typical case:

[0160] User B has picked up the package at 21:00 for the past 3 days due to overtime → The system recommends picking up the package at 21:00

[0161] Actual demand: The next day, User B returns to normal working hours, but the rule is not updated

[0162] Negative effect:

[0163] High - active users are repeatedly recommended the same strategy, resulting in the push shielding rate rising to 41%

[0164] Complaint rate: 3.1 times per week on average (vs 1.7 times in Example 4)

[0165] Defect analysis:

[0166] 1. Cold - start problem: New users (without historical behavior) cannot obtain effective recommendations

[0167] 2. Behavioral misjudgment: Temporary behaviors (such as overtime work) are misinterpreted as long-term preferences

[0168] 3. Lack of personalization: Failure to consider multi-dimensional correlations (such as the impact of weather, the correlation between collection on behalf and complaints)

[0169] Through Comparative Example 3, the fundamental defects of the traditional rule system in the recommendation scenario are fully revealed: Static rules cannot adapt to dynamic behavior changes, while machine learning models achieve true personalized decision-making through data-driven methods.

[0170] The specific detailed comparison data is shown in Table 1 below:

[0171]

[0172] Through the comparison between the examples and the comparative examples, the advantages of the machine learning model in scenarios such as parcel flow prediction and resource allocation optimization are verified.

[0173] In addition, the components designed in the present invention are all common standard components or components known to those skilled in the art. Their structures and principles can all be learned by those skilled in the art through technical manuals or obtained through conventional experimental methods. Those skilled in the art can fully implement them without further elaboration. The content protected by the present invention also does not involve improvements to internal structures and methods.

[0174] The embodiments disclosed in the present invention are preferred embodiments, but not limited thereto. Those of ordinary skill in the art can easily understand the spirit of the present invention based on the above embodiments and make different extensions and changes. As long as they do not depart from the spirit of the present invention, they are within the protection scope of the present invention.

Claims

1. A method for processing express post station information data, characterized in that: The following steps are involved: S1. Collect historical package data, including traffic records, time distribution and user behavior information; S2. Use machine learning models to perform in-depth mining on the data to predict package traffic, peak hours, and user preferences during future preset periods; S3. Dynamically plan the station’s human resources, storage space, and equipment configuration based on the prediction results; S4. Generate personalized pickup suggestions based on the user behavior analysis results and push them to the user terminal.

2. The express delivery station information data processing method according to claim 1, characterized in that: The prediction of future package traffic specifically includes: Use time series analysis algorithms to fit the trend of historical traffic data, and modify the forecast results based on seasonal characteristics; Introduce external variable data, including weather and promotion information, and optimize traffic prediction accuracy through the random forest model.

3. The express delivery station information data processing method according to claim 1, characterized in that: The peak period prediction is achieved by the following steps: Extract the peak time window of daily package arrival in historical data and construct a Gaussian mixture model for clustering; Correlate the user pickup delay time with the congestion level during peak hours, and dynamically adjust the start time and duration of the prediction window.

4. The express delivery station information data processing method according to claim 1, characterized in that: The user preference analysis includes: Based on the user's pickup time, collection selection and complaint record, a collaborative filtering algorithm is used to build a preference profile; Cluster users and optimize the generation logic of recommendation strategies for different groups.

5. The express delivery station information data processing method according to claim 1, characterized in that: The dynamic planning resource configuration specifically includes: Increase temporary storage shelves and staff shifts in proportion before the predicted peak period; Pre-allocate fast lane resources based on the proportion of "Instant Pickup" tags in user preferences.

6. The express delivery station information data processing method according to claim 1, characterized in that: The generation of the personalized pickup suggestion includes: Analyze the historical distribution of users’ pickup times and calculate the standard deviation of their usual time periods; When the congestion index during the forecast period exceeds the threshold, off-peak pickup plans are recommended to users whose standard deviation is greater than the preset value.

7. The express delivery station information data processing method according to claim 1, characterized in that: Also includes real-time optimization mechanisms: Collect package accumulation data in the post station in real time through IoT devices; When the actual traffic deviates from the predicted value by more than 20%, the dynamic resource reallocation algorithm is triggered to adjust the staff task priority.

8. The express delivery station information data processing method according to claim 1, characterized in that: Also includes exception handling module: Monitor the long period of time when packages have not been picked up, and build a risk scoring model based on the user's historical behavior; When the score exceeds the critical value, it will automatically switch to phone notification and generate a pending work order.

9. The express delivery station information data processing method according to claim 1, characterized in that: It also includes the effect evaluation stage: Compare the parcel handling efficiency indicators before and after the resource allocation adjustment; Use A / B testing to verify the user adoption rate of different recommendation strategies and update the optimal solution to the model training data set.

10. A device using the express post information data processing method according to claims 1-9, characterized in that: include: Data collection module, used to obtain station in and out records and user operation logs in real time; The prediction engine is equipped with an LSTM neural network to process time series data and output traffic and peak period prediction results; The resource scheduling center generates shift schedules and space allocation plans based on the forecast results; The recommendation system matches the optimal pickup strategy based on user profiles and sends it through a multi-channel push interface.

Images