Quality Detection Method, Device, Equipment and Storage Medium for Logistics Network Points

Abstract

The present invention relates to the field of artificial intelligence, and discloses a quality detection method, device, equipment and storage medium for logistics outlets. The method includes: obtaining the outlet information, historical weather data, historical business index data and historical outlet quality grades of a plurality of logistics outlets, and preprocessing them; calibrating the preprocessed data based on the historical outlet quality grades, and constructing a training data set; calling a network model to calculate the multi-classification matching probability distribution between each logistics outlet in the training data set and a plurality of preset outlet quality grade labels; calculating a loss value based on a preset loss function and data calibration, and adjusting the network parameters of the model according to the loss value to obtain a quality inspection model for logistics outlets; calling the quality inspection model to process the outlet information, current month weather information and current month business index data of the logistics outlet to be detected, so as to obtain the outlet quality grade. The present invention combines machine learning, thereby improving the quality inspection efficiency of logistics outlets.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to a method, device, equipment and storage medium for quality detection of logistics outlets. Background Art

[0002] Logistics quality refers to the sum of the ability characteristics of logistics activities to meet the production needs of enterprises and the consumption needs of customers. For logistics transportation enterprises with numerous logistics outlets, it is particularly important to conduct quality detection in a timely manner, issue early warnings, and strictly control the business quality of each logistics outlet. [[ID=I0]]

[0003] The existing quality detection methods for logistics outlets are inefficient because they conduct sampling inspections on the business data of logistics outlets and evaluate their business quality based on the sampling inspection results. Summary of the Invention

[0004] The main purpose of the present invention is to solve the problem of low accuracy of the existing quality detection methods for logistics outlets.

[0005] The first aspect of the present invention provides a method for quality detection of logistics outlets, including:

[0006] Obtain the outlet information, historical weather data, historical business index data and historical outlet quality grades of multiple logistics outlets, where the outlet information is used to represent the geographical location and outlet scale of the logistics outlet;

[0007] Preprocess the outlet information, the historical weather data and the historical business data, calibrate the preprocessed data based on the historical outlet quality grades, and construct a training data set according to the calibrated data;

[0008] Call a preset network model to calculate the multi-classification matching probability distribution between each logistics outlet in the training data set and a preset plurality of outlet quality grade labels;

[0009] Based on a preset loss function and the result of data calibration, calculate the loss value corresponding to the multi-classification matching probability distribution, and adjust the network parameters of the network model according to the loss value to obtain a quality inspection model for logistics outlets;

[0010] Call the quality inspection model of the logistics outlet to process the outlet information, the current month's weather information and the current month's business index data of the logistics outlet to be detected, and obtain the outlet quality grade of the logistics outlet to be detected.

[0011] Optionally, in the first implementation manner of the first aspect of the present invention, after obtaining the outlet information, historical weather data, historical business index data and historical outlet quality grades of multiple logistics outlets, it further includes:

[0012] Perform data cleaning on the said network point information, the said historical weather data, the said historical business indicator data, and the said historical network point quality grade to obtain the source-attached layer data of the logistics network points;

[0013] Based on a preset common layer data model, classify the source-attached layer data to obtain common layer data of different subject domains;

[0014] According to the preset business logic, perform data extraction on the common layer data of each subject domain to obtain application layer data;

[0015] Construct a data warehouse for the logistics network points based on the source-attached layer data, the common layer data, and the application layer data.

[0016] Optionally, in the second implementation manner of the first aspect of the present invention, after calling the quality inspection model of the logistics network point to process the network point information, the current month's weather information, and the current month's business indicator data of the logistics network point to be detected, and obtaining the network point quality grade of the logistics network point to be detected, it further includes:

[0017] Receive a report generation request sent by the terminal, and according to the report generation request, obtain the target identification information corresponding to the report to be generated;

[0018] According to the target identification information, obtain the report template and data filling rules corresponding to the report to be generated;

[0019] Based on the data filling rules, the report template, and the network point quality grade of the logistics network point to be detected, generate a quality inspection report for the logistics network point to be detected.

[0020] Optionally, in the third implementation manner of the first aspect of the present invention, the calling of the preset network model to calculate the multi-classification matching probability distribution between each logistics network point in the training data set and a preset plurality of network point quality grade labels includes:

[0021] Call the input layer network in the network model to encode each piece of data in the training data set to obtain the feature vector of the training sample;

[0022] Call the hidden layer network in the network model, and based on the feature vector of the training sample, perform feature learning to obtain the cross feature vector and the target feature vector;

[0023] Call the output layer network in the network model to calculate the cross feature vector and the target feature vector to obtain the multi-classification matching probability distribution between each logistics network point in the training data set and a preset plurality of network point quality grade labels.

[0024] Optionally, in the fourth implementation manner of the first aspect of the present invention, the hidden layer network includes a factorization network and a deep neural network. Invoking the hidden layer network in the network model and performing feature learning based on the feature vector of the training sample to obtain a cross feature vector and a target feature vector includes:

[0025] Invoking the factorization network to perform feature crossing on the discrete feature vector to obtain a cross feature vector, where the cross feature vector is used to represent the association relationship between discrete feature vectors;

[0026] Invoking the deep neural network to perform non-linear activation on the continuous feature vector to obtain a target feature vector.

[0027] Optionally, in the fifth implementation manner of the first aspect of the present invention, the output layer network includes a multi-classifier. Invoking the output layer network in the network model to calculate the cross feature vector and the target feature vector to obtain the multi-class matching probability distribution between each logistics network point in the training dataset and a preset plurality of network point quality level labels includes:

[0028] Invoking the multi-classifier to respectively average the cross feature vector and the target feature vector to obtain an average feature vector, and calculating a plurality of initial classification scores corresponding to the average feature vector according to a preset scoring function;

[0029] Based on a preset diffusion function, performing value diffusion on each initial classification score to obtain a plurality of target classification scores;

[0030] Performing normalization processing on each target classification score to obtain the multi-class matching probability distribution between each logistics network point in the training dataset and a preset plurality of network point quality level labels.

[0031] Optionally, in the sixth implementation manner of the first aspect of the present invention, calculating the loss value corresponding to the multi-class matching probability distribution based on a preset loss function and the result of data calibration, and adjusting the network parameters of the network model according to the loss value to obtain a quality inspection model for logistics network points includes:

[0032] Calculating the loss value corresponding to the multi-class matching probability distribution based on a preset loss function and the result of data calibration;

[0033] Backpropagating the loss value corresponding to the multi-class matching probability distribution from the output layer network of the network model to the input layer network of the network model;

[0034] Iteratively update the network parameters of the network model according to the loss value corresponding to the multi-classification matching probability distribution and the stochastic gradient descent algorithm until the network model converges, determine the current network parameters as the target parameters, and obtain the quality inspection model of the logistics network point.

[0035] The second aspect of the present invention provides a quality inspection device for logistics network points, including:

[0036] A data acquisition module, configured to acquire the network point information, historical weather data, historical business index data, and historical network point quality levels of multiple logistics network points, where the network point information is used to represent the geographical location and scale of the logistics network point;

[0037] A data set construction module, configured to preprocess the network point information, the historical weather data, and the historical business data, calibrate the preprocessed data based on the historical network point quality levels, and construct a training data set according to the calibrated data;

[0038] A model calculation module, configured to call a preset network model to calculate the multi-classification matching probability distribution between each logistics network point in the training data set and a preset plurality of network point quality level labels;

[0039] A model optimization module, configured to calculate the loss value corresponding to the multi-classification matching probability distribution based on a preset loss function and the result of data calibration, and adjust the network parameters of the network model according to the loss value to obtain a quality inspection model of the logistics network point;

[0040] A model calling module, configured to call the quality inspection model of the logistics network point to process the network point information, the current month's weather information, and the current month's business index data of the logistics network point to be detected, and obtain the network point quality level of the logistics network point to be detected.

[0041] Optionally, in the first implementation manner of the second aspect of the present invention, the model calculation module specifically includes:

[0042] A vector encoding unit, configured to call the input layer network in the network model to encode each piece of data in the training data set to obtain a feature vector of the training sample;

[0043] A feature learning unit, configured to call the hidden layer network in the network model and perform feature learning based on the feature vector of the training sample to obtain a cross feature vector and a target feature vector;

[0044] A probability calculation unit, configured to call the output layer network in the network model to calculate the cross feature vector and the target feature vector, so as to obtain a multi-classification matching probability distribution between each logistics network point in the training dataset and a plurality of preset network point quality level labels.

[0045] Optionally, in the second implementation manner of the second aspect of the present invention, the feature learning unit is specifically configured to:

[0046] Call the factorization network to perform feature crossing on the discrete feature vector to obtain a cross feature vector, where the cross feature vector is used to represent the association relationship between discrete feature vectors;

[0047] Call the deep neural network to perform non-linear activation on the continuous feature vector to obtain a target feature vector.

[0048] Optionally, in the third implementation manner of the second aspect of the present invention, the probability calculation unit is specifically configured to:

[0049] Call the multi-classifier to respectively average the cross feature vector and the target feature vector to obtain an average feature vector, and calculate a plurality of initial classification scores corresponding to the average feature vector according to a preset scoring function;

[0050] Based on a preset diffusion function, perform value diffusion on each initial classification score to obtain a plurality of target classification scores;

[0051] Perform normalization processing on each target classification score to obtain a multi-classification matching probability distribution between each logistics network point in the training dataset and a plurality of preset network point quality level labels.

[0052] Optionally, in the fourth implementation manner of the second aspect of the present invention, the model optimization module specifically includes:

[0053] A loss calculation unit, configured to calculate a loss value corresponding to the multi-classification matching probability distribution based on a preset loss function and the result of data calibration;

[0054] A backpropagation unit, configured to backpropagate the loss value corresponding to the multi-classification matching probability distribution from the output layer network of the network model to the input layer network of the network model;

[0055] An iterative update unit, configured to iteratively update the network parameters of the network model according to the loss value corresponding to the multi-classification matching probability distribution and the stochastic gradient descent algorithm until the network model converges, determine the current network parameters as target parameters, and obtain a quality inspection model for logistics network points.

[0056] In a third aspect of the present invention, there is provided a quality inspection device for logistics outlets, including: a memory and at least one processor, wherein instructions are stored in the memory; the at least one processor invokes the instructions in the memory so that the quality inspection device for logistics outlets executes the above-mentioned quality inspection method for logistics outlets.

[0057] In a fourth aspect of the present invention, there is provided a computer-readable storage medium, in which instructions are stored, and when it runs on a computer, it enables the computer to execute the above-mentioned quality inspection method for logistics outlets.

[0058] In the technical solution provided by the present invention, by obtaining the outlet information, historical weather data, historical business index data, and historical outlet quality grades of multiple logistics outlets, and performing preprocessing and data calibration on them, a training data set for model training is obtained. Secondly, a preset network model is trained based on the training data set, and the corresponding loss value is calculated based on the result of the model training. Finally, the network parameters of the network model are optimized according to the loss value to obtain a quality inspection model for logistics outlets. Finally, the quality inspection model is called to process the relevant data of the logistics outlets to be inspected. The present invention combines machine learning and trains a quality inspection model for logistics outlet quality inspection through the historical data of logistics outlets, thereby improving the quality inspection efficiency of logistics outlets. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a schematic diagram of the first embodiment of the quality inspection method for logistics outlets in an embodiment of the present invention;

[0060] Figure 2 It is a schematic diagram of the second embodiment of the quality inspection method for logistics outlets in an embodiment of the present invention;

[0061] Figure 3 It is a schematic diagram of the third embodiment of the quality inspection method for logistics outlets in an embodiment of the present invention;

[0062] Figure 4 It is a schematic diagram of the fourth embodiment of the quality inspection method for logistics outlets in an embodiment of the present invention;

[0063] Figure 5 It is a schematic diagram of an embodiment of the quality inspection device for logistics outlets in an embodiment of the present invention;

[0064] Figure 6 It is a schematic diagram of another embodiment of the quality inspection device for logistics outlets in an embodiment of the present invention;

[0065] Figure 7 It is a schematic diagram of an embodiment of the quality inspection device for logistics outlets in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0066] An embodiment of the present invention provides a method, device, equipment and storage medium for quality inspection of logistics outlets, with higher monitoring efficiency.

[0067] The terms "first", "second", "third", "fourth", etc. (if any) in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances, so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the term "comprising" or "having" and any variation thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or equipment comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or equipment.

[0068] For ease of understanding, the specific process of the embodiment of the present invention will be described below. Please refer to Figure 1 , an embodiment of the quality inspection method for logistics outlets in the embodiment of the present invention includes:

[0069] 101. Obtain the outlet information, historical weather data, historical business indicator data and historical outlet quality grades of multiple logistics outlets, where the outlet information is used to represent the geographical location and outlet scale of the logistics outlets;

[0070] It can be understood that a logistics company may have multiple logistics outlets distributed across the country. The business quality of a logistics outlet refers to the sum of the ability characteristics of logistics activities to meet the production needs of the enterprise and the consumption needs of customers. In this embodiment, the business quality of logistics outlets is quantitatively graded to distinguish the business quality of each logistics outlet. There are many factors affecting the business quality of logistics outlets. In addition to business indicator data, the geographical location, outlet scale and weather conditions of the logistics outlets all have a certain impact on the business quality. Among them, business indicator data is the key performance indicator in the logistics field. For example, business indicator data includes the total inventory amount (the total amount calculated at the incoming cost price of all inventory products at the end of the month), the total inventory quantity (the quantity of all inventory products at the end of the month), the material incoming error rate (the number of incoming errors / the total number of incoming), the order matching accuracy rate ((the number of lost orders + the number of wrong orders) / the total number of orders), the defective product outflow rate (the number of defective products flowing out per month / the total number of transactions per month), the order delay rate (the number of times of delayed order sending per month / the total number of order sendings per month), the express bill matching error rate (the number of shipping matching errors per month / the total number of order shipments per month), the invoice matching accuracy rate (the number of invoice matching errors for shipping per month / the total number of orders per month), the document storage loss rate (the number of lost documents at the end of the month / (the total number of shipping times + the number of receiving times) per month), the warehouse environment audit (whether the warehouse environment is clean and orderly), etc.

[0071] Optionally, the server can also construct a unified data warehouse for logistics outlets based on the obtained outlet information, historical weather data, historical business indicator data, and historical outlet quality grades (hereinafter referred to as the original data of logistics outlets). Specifically, the server first performs data cleaning on the original data of logistics outlets to obtain source-attached layer data; secondly, based on a preset common layer data model, classifies the source-attached layer data to obtain common layer data for different subject domains; then, according to the preset business logic, extracts the common layer data for each subject domain to obtain application layer data; finally, constructs a data warehouse based on the source-attached layer data, common layer data, and application layer data. Through hierarchical data storage, it abstracts a common framework for the future and can empower other business lines, while providing stable and accurate data support for business development and enabling data-driven and empowerment for the development of new businesses according to the existing models.

[0072] 102. Preprocess the outlet information, historical weather data, and historical business data, calibrate the preprocessed data based on the historical outlet quality grades, and construct a training data set according to the calibrated data.

[0073] It can be understood that data is the raw material for machine learning. Before performing the training task on the network model, the server performs preprocessing means such as data cleaning and data conversion on the outlet information, historical weather data, and historical business data, so as to standardize and normalize the data to further improve the speed of model calculation and processing. The specific method of preprocessing in this embodiment is not limited.

[0074] Furthermore, the server uses the outlet information, historical weather data, and historical business data of the logistics outlet as sample data, and adds corresponding data labels to them through a preset data annotation tool (such as doccano, labelme), and the annotation content is the quality grade of the logistics outlet, which is convenient for the network model to perform supervised training.

[0075] 103. Invoke a preset network model to calculate the multi-classification matching probability distribution between each logistics outlet in the training data set and a preset multiple outlet quality grade labels.

[0076] It can be understood that the preset network model preferably adopts the Deep Factorization Machines (DeepFM) model, which is mainly composed of two parts: the Factorization Machines (FM) and the Deep Neural Networks (DNN). Among them, the DNN network serves as the backbone network of the prediction model, and the FM network serves as an auxiliary network. The two share the same input layer, so as to learn the low-order and high-order feature cross processes simultaneously. Finally, the output vectors and their weights of the DNN network and the FM network are input into the fully connected layer of the prediction model, so as to calculate the probability of the logistics network point and each network point quality level label. Among them, the network point quality level label is the same as the data label added during the data calibration process. The server classifies the quality level of the logistics network point through the given data label.

[0077] 104. Based on the preset loss function and the result of data calibration, calculate the loss value corresponding to the multi-classification matching probability distribution, and adjust the network parameters of the network model according to the loss value to obtain the quality inspection model of the logistics network point;

[0078] It can be understood that the server calculates the loss value (numerically representing the error between the two) between the multi-classification matching probability distribution and the data calibration result of the training data set based on the preset loss function. When the loss value is smaller, it means that the prediction result is closer to the true sampling inspection result, and the prediction result (multi-classification matching probability distribution) is more accurate.

[0079] Furthermore, the server iteratively adjusts the network parameters of the model according to the size of the loss value, then trains again and calculates the corresponding loss value until the model converges, determines the network parameters with the best effect, and thus obtains the quality inspection model of the logistics network point.

[0080] Specifically, the server inputs the loss value corresponding to the multi-classification matching probability distribution into the prediction model, and propagates the loss value backward from the output layer of the network model to the hidden layer until it reaches the input layer of the prediction model. During the backward propagation process, according to the loss value, a point direction is randomly selected for gradient descent, and according to the result of the gradient descent, the weights of the feature vectors in the input layer and the hidden layer of the model are iteratively updated. After each weight update, the corresponding multi-classification matching probability distribution and its loss value are recalculated until the model converges. For example, when the loss value is less than the preset threshold, the current network parameters are determined as the target parameters, and the quality inspection model of the logistics network point is obtained.

[0081] 105. Call the quality inspection model of the logistics network point, process the network point information, the weather information of the current month, and the business index data of the current month of the logistics network point to be inspected, and obtain the network point quality level of the logistics network point to be inspected.

[0082] It can be understood that the quality inspection model of the logistics network point is essentially a classification model. After the server calculates and processes the network point information, the weather information of the current month, and the business index data of the current month of the logistics network point to be inspected, it obtains the matching probability distribution between the logistics network point and multiple preset data tags. If a certain matching probability in the matching probability distribution is greater than the preset threshold, it is output as the quality level of the logistics network point to be inspected.

[0083] Optionally, the server also receives a report generation request sent by the terminal, and thus generates a quality inspection report of the logistics network point. Specifically, the server obtains the target identification information corresponding to the report to be generated according to the report generation request; then, according to the target identification information, obtains the report template and data filling rules corresponding to the report to be generated; finally, based on the data filling rules, the report template, and the network point quality level of the logistics network point to be inspected, generates a quality inspection report of the logistics network point to be inspected. It should be understood that the target identification information is used to indicate the report caliber size of the report to be generated, and the report template corresponds to a unique data filling rule. The server fills the network point quality level detected by the quality inspection model into the corresponding position in the report template according to the data filling rule, so as to obtain the quality inspection report of the logistics network point. Optionally, the server can also publish the quality inspection report of the logistics network point as a corresponding interface service for unified call and mobile display by each business department.

[0084] In this embodiment, combined with machine learning, a quality inspection model for logistics network point quality detection is trained through the historical data of the logistics network point, thereby improving the quality inspection efficiency of the logistics network point.

[0085] Refer to Figure 2 , the second embodiment of the quality inspection method for logistics network points in the embodiment of the present invention includes:

[0086] 201. Obtain the network point information, historical weather data, historical business index data, and historical network point quality levels of multiple logistics network points, where the network point information is used to represent the geographical location and network scale of the logistics network point;

[0087] 202. Preprocess the network point information, historical weather data, and historical business data, and perform data calibration on the preprocessed data based on the historical network point quality levels, and construct a training data set according to the data after data calibration;

[0088] Among them, the execution steps of steps 201-202 are similar to those of the above steps 101-102, and will not be elaborated here specifically.

[0089] 203. Call the input layer network in the network model to encode each piece of data in the training dataset, and obtain the feature vectors of the training samples;

[0090] It can be understood that the input layer network mainly includes an encoding network. The server inputs the data into the input layer of the prediction model and calls the encoding network in the input layer to encode the data in the training dataset, thereby translating the text data into feature vectors that the model can recognize. Preferably, the encoding network is based on the self-attention mechanism and learns the weights of adjacent nodes before and after during word vector encoding.

[0091] Optionally, in a preferred embodiment, to improve the computing speed of the model, an embedding layer is further introduced between the input layer and the hidden layer of the network model. The server calls this embedding layer to perform dimensionality reduction processing on high-dimensional sparse discrete feature vectors and continuous feature vectors (i.e., call a 1*1 size convolutional kernel to perform a multiplication operation with them), thereby compressing them into dense low-dimensional vectors.

[0092] 204. Call the hidden layer network in the network model, and based on the feature vectors of the training samples, perform feature learning to obtain cross feature vectors and target feature vectors;

[0093] It can be understood that the hidden layer network of the model includes a backbone network (DNN network) and an auxiliary network (FM network). The server calls the backbone network to learn the data features of the non-linear distribution in the continuous feature vectors, such as the geographical location of logistics outlets, some business indicator data, etc., and introduce them and their relationships into the new vector representation. At the same time, use the auxiliary network to extract the feature correlation relationships between discrete feature vectors, such as weather information and some business indicator data. Among them, some business indicator data is text type, while the weather information is numerical type. In normal cases, the model cannot recognize the relationship between these two features. In this case, this auxiliary network can be used to extract the correlation relationship between the two.

[0094] 205. Call the output layer network in the network model to calculate the cross feature vectors and target feature vectors, and obtain the multi-classification matching probability distribution between each logistics outlet in the training dataset and a preset multiple outlet quality level labels;

[0095] It can be understood that the server inputs the continuous feature vectors output by the DNN network and the cross feature vectors output by the FM network into the output layer network of the model, and based on a multi-classifier, maps them into the (0,1) numerical interval, that is, a multi-classification probability distribution.

[0096] 206. Based on the preset loss function and the result of data calibration, calculate the loss value corresponding to the multi-classification matching probability distribution, and adjust the network parameters of the network model according to the loss value to obtain a quality inspection model for logistics outlets.

[0097] 207. Invoke the quality inspection model of the logistics outlet to process the outlet information, the weather information of the current month, and the business index data of the current month of the logistics outlet to be detected, and obtain the outlet quality grade of the logistics outlet to be detected.

[0098] Among them, the execution steps of steps 206-207 are similar to those of the above steps 104-105, and will not be elaborated here specifically.

[0099] In this embodiment, the process of the network model calculating the multi-classification matching probability distribution is described in detail. After quantifying and expressing the data features, mathematical calculations are performed to accurately calculate the matching probability between the logistics outlet and the preset quality grade label.

[0100] Refer to Figure 3 , the third embodiment of the quality inspection method for logistics outlets in the embodiment of the present invention includes:

[0101] 301. Obtain the outlet information, historical weather data, historical business index data, and historical outlet quality grades of multiple logistics outlets, where the outlet information is used to represent the geographical location and outlet scale of the logistics outlet.

[0102] 302. Preprocess the outlet information, historical weather data, and historical business data, and based on the historical outlet quality grades, calibrate the preprocessed data, and construct a training data set according to the calibrated data.

[0103] Among them, the execution steps of steps 301-302 are similar to those of the above steps 101-102, and will not be elaborated here specifically.

[0104] 303. Invoke the input layer network in the network model to encode each piece of data in the training data set to obtain the feature vector of the training sample, where the feature vector of the training sample includes a discrete feature vector and a continuous feature vector.

[0105] 304. Invoke the factorization network in the network model to perform feature crossing on the discrete feature vector to obtain a cross feature vector, where the cross feature vector is used to represent the correlation relationship between discrete feature vectors.

[0106] It can be understood that the server performs feature crossing on each discrete feature vector through the factorization network in the prediction model, so as to extract the correlation relationship between discrete training samples. For discrete features, the feature crossing is the corresponding Cartesian product. Specifically, for example, there are discrete training sample A and discrete training sample B. The discrete training sample A has 3 feature vectors, namely a1, a2, and a3, and the discrete training sample B has 2 feature vectors, namely b1 and b2. Performing feature crossing on them can obtain 6 new feature vectors, namely a1_b1, a1_b2, a2_b1, a2_b2, a3_b1, and a3_b2.

[0107] 305. Invoke the deep neural network in the network model to perform non-linear activation on the continuous feature vector to obtain the target feature vector;

[0108] It can be understood that the data in each continuous feature vector has both linear and non-linear distributions. The server performs non-linear activation on each continuous feature vector through the deep neural network. Specifically, a non-linear transformation is performed on the continuous feature vector through a preset activation function (such as the Relu function), so as to learn more feature expressions.

[0109] 306. Invoke the output layer network in the network model to calculate the cross feature vector and the target feature vector to obtain the multi-classification matching probability distribution between each logistics network point in the training dataset and a preset plurality of network point quality level labels;

[0110] 307. Based on the preset loss function and the result of data calibration, calculate the loss value corresponding to the multi-classification matching probability distribution, and adjust the network parameters of the network model according to the loss value to obtain the quality inspection model of the logistics network point;

[0111] 308. Invoke the quality inspection model of the logistics network point to process the network point information, the weather information of the current month, and the business index data of the current month of the logistics network point to be detected to obtain the network point quality level of the logistics network point to be detected.

[0112] Among them, the execution steps of steps 306-308 are similar to those of the above steps 205-207, and will not be elaborated here specifically.

[0113] In this embodiment, the process of feature learning is described in detail. By performing feature crossing and non-linear transformation on the sample data, structured data features are extracted from unstructured data samples, and corresponding probabilities are calculated based on the extracted structured data features, thereby improving the accuracy of classification.

[0114] Refer to Figure 4 , the fourth embodiment of the quality inspection method for logistics network points in the embodiment of the present invention includes:

[0115] 401. Obtain the outlet information, historical weather data, historical business indicator data, and historical outlet quality grades of multiple logistics outlets, where the outlet information is used to represent the geographical location and outlet scale of the logistics outlets;

[0116] 402. Preprocess the outlet information, historical weather data, and historical business data, calibrate the preprocessed data based on the historical outlet quality grades, and construct a training dataset according to the calibrated data;

[0117] 403. Invoke the input layer network in the network model to encode each piece of data in the training dataset to obtain the feature vectors of the training samples;

[0118] 404. Invoke the hidden layer network in the network model and perform feature learning based on the feature vectors of the training samples to obtain cross feature vectors and target feature vectors;

[0119] Among them, the execution steps of steps 401-404 are similar to those of the above steps 201-204, and will not be elaborated here specifically.

[0120] 405. Invoke the multi-classifier in the network model, average the cross feature vectors and target feature vectors respectively to obtain average feature vectors, and calculate multiple initial classification scores corresponding to the average feature vectors according to a preset scoring function;

[0121] It should be understood that the multi-classifier adopted in this embodiment is a Softmax classifier. First, the server sums the cross feature vectors containing the cross relationships between discrete data and the target feature vectors containing the linear and non-linear distributions in continuous data, and then calculates the corresponding average vector; secondly, according to the preset scoring function in the classifier, calculate multiple initial classification scores (i.e., the distribution T between the outlet quality grade labels - initial classification scores) of the average vector in different situations (multiple classification labels, here are the outlet quality grade labels), for example, T = [Grade A: 2.8, Grade B: 5.3, Grade C: -1.2].

[0122] 406. Perform value diffusion on each initial classification score based on a preset diffusion function to obtain multiple target classification scores;

[0123] It can be understood that the data dispersion among multiple initial classification scores corresponding to the average vector is small. The server uses a preset diffusion function, such as the exp function (calculating the e to the power of x of the initial classification score, where e is Euler's number, an infinite non-repeating decimal, and x is the initial classification score) to perform value diffusion (numerical expansion) on it. When the score is larger, the data dispersion is larger and the result is more obvious. For example, performing value diffusion on distribution T to obtain distribution X = [A:e 2.8 , B:e 5.3 , C:e -1.2 , that is, X = [A:16.4, B:200.3, C:0.30].

[0124] 407. Perform normalization processing on each target classification score to obtain the multi-classification matching probability distribution between each logistics network point in the training dataset and a preset multiple network point quality level labels.

[0125] It can be understood that the purpose of normalization processing (Normalize) is to map the data to a decimal between 0 and 1 (i.e., a probability). For example, performing normalization on distribution X. By summing each target classification score in distribution X to obtain the total score, and then calculating the ratio between each classification score and the total score, the multi-classification matching probability distribution Z = [A:0.08, B:0.92, C:0.00] is obtained.

[0126] 408. Based on the preset loss function and the result of data calibration, calculate the loss value corresponding to the multi-classification matching probability distribution, and adjust the network parameters of the network model according to the loss value to obtain the quality inspection model of the logistics network point;

[0127] 409. Invoke the quality inspection model of the logistics network point to process the network point information, the weather information of the current month, and the business index data of the current month of the logistics network point to be detected, and obtain the network point quality level of the logistics network point to be detected.

[0128] Among them, the execution steps of steps 408 - 409 are similar to those of the above steps 104 - 105, and will not be elaborated here specifically.

[0129] In this embodiment, the calculation process of the multi-classification matching probability distribution is described in detail. The initial score is calculated through the scoring function in the multi-classifier, value diffusion is performed on it, and finally it is normalized to obtain the matching probability between the logistics network point and each network point quality level label.

[0130] The quality inspection method of the logistics network point in the embodiment of the present invention has been described above. Next, the quality inspection device of the logistics network point in the embodiment of the present invention will be described. Please refer to Figure 5 , an embodiment of the quality inspection device of the logistics network point in the embodiment of the present invention includes:

[0131] A data acquisition module 501 for acquiring the outlet information, historical weather data, historical business metric data, and historical outlet quality grades of multiple logistics outlets, where the outlet information is used to represent the geographical location and outlet scale of the logistics outlets;

[0132] A dataset construction module 502 for preprocessing the outlet information, the historical weather data, and the historical business data, calibrating the preprocessed data based on the historical outlet quality grades, and constructing a training dataset according to the calibrated data;

[0133] A model calculation module 503 for calling a preset network model to calculate the multi-classification matching probability distribution between each logistics outlet in the training dataset and a preset plurality of outlet quality grade labels;

[0134] A model optimization module 504 for calculating the loss value corresponding to the multi-classification matching probability distribution based on a preset loss function and the result of data calibration, and adjusting the network parameters of the network model according to the loss value to obtain a quality inspection model for logistics outlets;

[0135] A model calling module 505 for calling the quality inspection model of the logistics outlets to process the outlet information, the current month's weather information, and the current month's business metric data of the logistics outlet to be inspected, and obtaining the outlet quality grade of the logistics outlet to be inspected.

[0136] In this embodiment, in combination with machine learning, a quality inspection model for logistics outlet quality inspection is trained through the historical data of logistics outlets, thereby improving the quality inspection efficiency of logistics outlets.

[0137] Refer to Figure 6 , another embodiment of the quality inspection device for logistics outlets in the embodiment of the present invention includes:

[0138] A data acquisition module 501 for acquiring the outlet information, historical weather data, historical business metric data, and historical outlet quality grades of multiple logistics outlets, where the outlet information is used to represent the geographical location and outlet scale of the logistics outlets;

[0139] A dataset construction module 502 for preprocessing the outlet information, the historical weather data, and the historical business data, calibrating the preprocessed data based on the historical outlet quality grades, and constructing a training dataset according to the calibrated data;

[0140] A model calculation module 503 for calling a preset network model to calculate the multi-classification matching probability distribution between each logistics outlet in the training dataset and a preset plurality of outlet quality grade labels;

[0141] A model optimization module 504, configured to calculate a loss value corresponding to the multi-classification matching probability distribution based on a preset loss function and the result of data calibration, and adjust network parameters of the network model according to the loss value to obtain a quality inspection model for logistics outlets;

[0142] A model calling module 505, configured to call the quality inspection model for logistics outlets to process the outlet information, the weather information of the current month, and the business index data of the current month of the logistics outlet to be detected, and obtain the outlet quality level of the logistics outlet to be detected.

[0143] Wherein, the model calculation module 503 specifically includes:

[0144] A vector encoding unit 5031, configured to call an input layer network in the network model to encode each piece of data in the training data set to obtain a feature vector of the training sample;

[0145] A feature learning unit 5032, configured to call a hidden layer network in the network model and perform feature learning based on the feature vector of the training sample to obtain a cross feature vector and a target feature vector;

[0146] A probability calculation unit 5033, configured to call an output layer network in the network model to calculate the cross feature vector and the target feature vector to obtain a multi-classification matching probability distribution between each logistics outlet in the training data set and a preset plurality of outlet quality level labels.

[0147] Wherein, the feature learning unit 5032 is specifically configured to:

[0148] Call the factorization network to perform feature crossing on the discrete feature vector to obtain a cross feature vector, where the cross feature vector is used to represent the correlation relationship between discrete feature vectors;

[0149] Call the deep neural network to perform non-linear activation on the continuous feature vector to obtain a target feature vector.

[0150] Wherein, the probability calculation unit 5033 is specifically configured to:

[0151] Call the multi-classifier to average the cross feature vector and the target feature vector respectively to obtain an average feature vector, and calculate a plurality of initial classification scores corresponding to the average feature vector according to a preset scoring function;

[0152] Perform value diffusion on each initial classification score based on a preset diffusion function to obtain a plurality of target classification scores;

[0153] Normalize each target classification score to obtain the multi-classification matching probability distribution between each logistics network point in the training dataset and a plurality of preset network quality level labels.

[0154] Among them, the model optimization module 504 specifically includes:

[0155] A loss calculation unit 5041, configured to calculate a loss value corresponding to the multi-classification matching probability distribution based on a preset loss function and the result of data calibration;

[0156] A backpropagation unit 5042, configured to backpropagate the loss value corresponding to the multi-classification matching probability distribution from the output layer network of the network model to the input layer network of the network model;

[0157] An iterative update unit 5043, configured to iteratively update the network parameters of the network model according to the loss value corresponding to the multi-classification matching probability distribution and the stochastic gradient descent algorithm until the network model converges, determine the current network parameters as target parameters, and obtain a quality inspection model for logistics network points.

[0158] In the embodiment of the present invention, the modular design enables the hardware of each part of the quality inspection device for logistics network points to focus on the implementation of a certain function, maximizing the performance of the hardware. At the same time, the modular design also reduces the coupling between the modules of the device, making it more convenient to maintain.

[0159] Above Figure 5 And Figure 6 The quality inspection device for logistics network points in the embodiment of the present invention has been described in detail from the perspective of modular functional entities. Next, the quality inspection equipment for logistics network points in the embodiment of the present invention will be described in detail from the perspective of hardware processing.

[0160] Figure 7FIG. 0 is a schematic structural diagram of a quality inspection device for a logistics network point provided by an embodiment of the present invention. The quality inspection device 700 for the logistics network point may vary greatly due to different configurations or performances, and may include one or more processors (central processing units, CPUs) 710 (for example, one or more processors) and a memory 720, and one or more storage media 730 for storing application programs 733 or data 732 (for example, one or more mass storage devices). Among them, the memory 720 and the storage media 730 may be transient storage or persistent storage. The program stored in the storage media 730 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations on the quality inspection device 700 for the logistics network point. Further, the processor 710 may be configured to communicate with the storage media 730 and execute a series of instruction operations in the storage media 730 on the quality inspection device 700 for the logistics network point.

[0161] The quality inspection device 700 for the logistics network point may further include one or more power supplies 740, one or more wired or wireless network interfaces 750, one or more input / output interfaces 760, and / or one or more operating systems 731, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, and so on. Those skilled in the art can understand that Figure 7 the shown structural diagram of the quality inspection device for the logistics network point does not constitute a limitation on the quality inspection device for the logistics network point, and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0162] The present invention also provides a quality inspection device for a logistics network point. The quality inspection device for the logistics network point includes a memory and a processor. When the computer-readable instructions stored in the memory are executed by the processor, the processor executes the steps of the quality inspection method for the logistics network point in the above-mentioned embodiments.

[0163] The present invention also provides a computer-readable storage medium. The computer-readable storage medium may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions run on a computer, the computer executes the steps of the quality inspection method for the logistics network point.

[0164] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described systems, devices, and units can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0165] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0166] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of various embodiments of the present invention.

Claims

1. A quality inspection method for logistics outlets, characterized in that, The quality inspection method for the logistics network points includes: Obtain the network point information, historical weather data, historical business indicator data, and historical network point quality grades of multiple logistics network points. Among them, the network point information is used to represent the geographical location and scale of the logistics network points; Preprocess the network point information, the historical weather data, and the historical business data, and based on the historical network point quality grades, perform data calibration on the preprocessed data, and construct a training data set according to the calibrated data; Call a preset network model to calculate the multi-classification matching probability distribution between each logistics network point in the training data set and a preset plurality of network point quality grade labels; the network model is a deep factorization machine model, including a factorization machine and a deep neural network, and the factorization machine and the deep neural network share an input layer; Based on a preset loss function and the result of data calibration, calculate the loss value corresponding to the multi-classification matching probability distribution, and adjust the network parameters of the network model according to the loss value to obtain a quality inspection model for the logistics network points; Call the quality inspection model of the logistics network points to process the network point information, the current month's weather information, and the current month's business indicator data of the logistics network point to be detected, and obtain the network point quality grade of the logistics network point to be detected.

2. The quality inspection method of the logistics network according to claim 1, wherein After obtaining the network point information, historical weather data, historical business indicator data, and historical network point quality grades of multiple logistics network points, it further includes: Perform data cleaning on the network point information, the historical weather data, the historical business indicator data, and the historical network point quality grades to obtain the source-attached layer data of the logistics network points; Based on a preset common layer data model, perform data classification on the source-attached layer data to obtain common layer data in different subject domains; According to preset business logics, perform data extraction on the common layer data in each subject domain to obtain application layer data; Construct a data warehouse for the logistics network points based on the source-attached layer data, the common layer data, and the application layer data.

3. The quality inspection method of the logistics network point according to claim 1, characterized in that After calling the quality inspection model of the logistics network points to process the network point information, the current month's weather information, and the current month's business indicator data of the logistics network point to be detected, and obtaining the network point quality grade of the logistics network point to be detected, it further includes: Receive a report generation request sent by a terminal, and according to the report generation request, obtain the target identification information corresponding to the report to be generated; According to the target identification information, obtain the report template and data filling rules corresponding to the report to be generated; Generate a quality inspection report for the logistics network point to be detected based on the data filling rules, the report template, and the network point quality grade of the logistics network point to be detected.

4. The quality inspection method of the logistics network according to claim 1, characterized in that, The calling the preset network model to calculate the multi-classification matching probability distribution between each logistics network point in the training data set and a preset plurality of network point quality grade labels includes: Call the input layer network in the network model to encode each piece of data in the training data set to obtain the feature vector of the training sample; Invoke the hidden layer network in the network model, and perform feature learning based on the feature vectors of the training samples to obtain cross feature vectors and target feature vectors; Invoke the output layer network in the network model, and calculate the cross feature vectors and the target feature vectors to obtain the multi-classification matching probability distribution between each logistics network point in the training dataset and a plurality of preset network point quality level labels; 5. The quality inspection method of the logistics network point according to claim 4, characterized in that, The hidden layer network includes a factorization network and a deep neural network. The step of invoking the hidden layer network in the network model and performing feature learning based on the feature vectors of the training samples to obtain cross feature vectors and target feature vectors includes: Invoke the factorization network to perform feature crossing on the discrete feature vectors to obtain cross feature vectors, where the cross feature vectors are used to represent the association relationships between the discrete feature vectors; Invoke the deep neural network to perform non-linear activation on the continuous feature vectors to obtain target feature vectors.

6. The quality inspection method of the logistics network according to claim 4, characterized in that The output layer network includes a multi-classifier. The step of invoking the output layer network in the network model, and calculating the cross feature vectors and the target feature vectors to obtain the multi-classification matching probability distribution between each logistics network point in the training dataset and a plurality of preset network point quality level labels includes: Invoke the multi-classifier to respectively average the cross feature vectors and the target feature vectors to obtain average feature vectors, and calculate a plurality of initial classification scores corresponding to the average feature vectors according to a preset scoring function; Based on a preset diffusion function, perform value diffusion on each initial classification score to obtain a plurality of target classification scores; Perform normalization processing on each target classification score to obtain the multi-classification matching probability distribution between each logistics network point in the training dataset and a plurality of preset network point quality level labels.

7. The quality inspection method of the logistics network point according to any one of claims 1-6, characterized in that The step of calculating the loss value corresponding to the multi-classification matching probability distribution based on a preset loss function and the result of data calibration, and adjusting the network parameters of the network model according to the loss value to obtain a quality inspection model for logistics network points includes: Calculate the loss value corresponding to the multi-classification matching probability distribution based on a preset loss function and the result of data calibration; Backpropagate the loss value corresponding to the multi-classification matching probability distribution from the output layer network of the network model to the input layer network of the network model; According to the loss value corresponding to the multi-classification matching probability distribution and the stochastic gradient descent algorithm, iteratively update the network parameters of the network model until the network model converges, determine the current network parameters as target parameters, and obtain a quality inspection model for logistics network points.

8. A quality inspection device for a logistics network point, characterized in that, The quality inspection device for logistics network points includes: A data acquisition module, configured to acquire network point information, historical weather data, historical business index data, and historical network point quality levels of a plurality of logistics network points, where the network point information is used to represent the geographical location and network point scale of the logistics network points; A dataset construction module, configured to preprocess the outlet information, the historical weather data, and the historical business data, calibrate the preprocessed data based on the historical outlet quality grade, and construct a training dataset according to the calibrated data; A model calculation module, configured to call a preset network model to calculate a multi-classification matching probability distribution between each logistics outlet in the training dataset and a preset plurality of outlet quality grade labels; the network model is a deep factorization machine model, including a factorization machine and a deep neural network, and the factorization machine and the deep neural network share an input layer; A model optimization module, configured to calculate a loss value corresponding to the multi-classification matching probability distribution based on a preset loss function and the result of data calibration, and adjust network parameters of the network model according to the loss value to obtain a quality inspection model for logistics outlets; A model calling module, configured to call the quality inspection model for logistics outlets to process the outlet information, the current month's weather information, and the current month's business indicator data of the logistics outlet to be detected, and obtain the outlet quality grade of the logistics outlet to be detected.

9. A quality inspection device for a logistics network point, characterized in that, The quality inspection device for logistics outlets includes: a memory and at least one processor, and instructions are stored in the memory; The at least one processor calls the instructions in the memory so that the quality inspection device for logistics outlets executes the quality inspection method for logistics outlets according to any one of claims 1-7.

10. A computer-readable storage medium, on which instructions are stored, characterized in that, When the instructions are executed by the processor, the quality inspection method for logistics outlets according to any one of claims 1-7 is implemented.

Images