A method and device for AI scheduling and allocation based on fresh food orders

Abstract

The present invention discloses a method and device for AI scheduling and allocation based on fresh food orders, belonging to the technical field of semantic processing. The present invention calculates the characteristic overlap between a current food characteristic set and a food characteristic set corresponding to a historical fresh food order, screens and recommends refrigeration equipment, and obtains a first recommended refrigeration equipment set and a first overlap vector; extracts user demand information on the fresh food order, calculates the demand overlap between the user demand and the remaining refrigeration equipment in the warehouse, and obtains a second recommended refrigeration equipment set and a second overlap vector; uses an AI scheduling and allocation model to extract semantics from the current food characteristic set, the first recommended refrigeration equipment set, and the second recommended refrigeration equipment set, and obtains the allocated refrigeration equipment based on the weights applied to the first overlap vector and the second overlap vector; the present invention achieves accurate matching of refrigeration equipment through in-depth analysis of food characteristics and user needs of fresh food orders.

Description

Technical Field

[0001] The present invention relates to the field of semantic processing technology, and in particular to a method and device for AI scheduling and allocation based on fresh food orders. Background Art

[0002] With the rapid development of e-commerce platforms and fresh food e-commerce, fresh food cold chain logistics has become a key link in the modern food supply chain. At present, there are serious pain points in the traditional storage and transportation scheduling of fresh food: the traditional method mainly relies on manual experience and fixed refrigeration equipment allocation mode, which cannot accurately match the special temperature control requirements of different fresh foods, resulting in poor food preservation effect, low energy utilization efficiency, and difficulty in quickly responding to diverse user personalized needs. Traditional scheduling systems often adopt static and extensive allocation methods, ignoring the differences in food types and the specific needs of users, which not only increases the risk of food spoilage, but also causes waste of refrigeration resources. Summary of the invention

[0003] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method and device for AI scheduling and allocation based on fresh food orders, which solves the problem of low allocation accuracy of refrigeration equipment in the prior art.

[0004] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a method for AI scheduling and allocation based on fresh food orders, comprising the following steps:

[0005] S1, extracting food characteristic information on fresh food orders and constructing a current food characteristic set;

[0006] S2, calculating the overlap between the current food feature set and the food feature set corresponding to the historical fresh food order, screening the recommended refrigeration equipment, and obtaining a first recommended refrigeration equipment set and a first overlap vector;

[0007] S3, extracting user demand information on fresh food orders, calculating the overlap between user demand and demand of remaining refrigeration equipment in the warehouse, and obtaining a second recommended refrigeration equipment set and a second overlap vector;

[0008] S4. Use the AI ​​scheduling allocation model to extract semantics from the current food feature set, the first recommended refrigeration equipment set, and the second recommended refrigeration equipment set, and obtain the allocated refrigeration equipment based on the weights applied to the first coincidence vector and the second coincidence vector.

[0009] Furthermore, the food characteristic information in S1 includes: food type, food name, shelf life, storage temperature, food volume and food perishability;

[0010] The process of constructing the current food characteristic set in S1 is specifically as follows: segmenting the text corresponding to the food characteristic information according to food type, food name, shelf life, storage temperature, food volume and food perishability to obtain multiple phrases, vectorizing each phrase to obtain a phrase vector, and arranging the multiple phrase vectors in sequence to obtain the current food characteristic set.

[0011] Further, S2 includes the following sub-steps:

[0012] S21, taking the intersection of the current food characteristic set and the food characteristic set corresponding to each historical fresh food order, and counting the number of phrase vectors in the intersection;

[0013] S22, taking the ratio of the number of phrase vectors in the intersection to the total number of phrase vectors in the current food characteristic set as the characteristic overlap;

[0014] S23, extracting refrigeration equipment that matches historical fresh food orders and has a feature overlap greater than 0.5 as a set of candidate refrigeration equipment;

[0015] S24, extracting the remaining refrigeration equipment in the warehouse and constructing a set of remaining refrigeration equipment in the warehouse;

[0016] S25, taking the intersection of the candidate refrigeration equipment set and the remaining refrigeration equipment set in the warehouse to obtain a first recommended refrigeration equipment set;

[0017] S26: Using the characteristic coincidences corresponding to the refrigeration equipment in the first recommended refrigeration equipment set as elements to construct a first coincidence vector.

[0018] Furthermore, S3 includes the following sub-steps:

[0019] S31, extracting user demand information on fresh food orders;

[0020] S32, performing word segmentation processing on the user demand information to obtain multiple phrases, vectorizing each phrase to obtain a phrase vector, and arranging the multiple phrase vectors in sequence to obtain a user demand set;

[0021] S33, taking the intersection of the user demand set and the performance set of the remaining refrigeration equipment in each warehouse, and counting the number of phrase vectors in the intersection;

[0022] S34, taking the ratio of the number of phrase vectors in the intersection to the total number of phrase vectors in the user demand set as the demand overlap;

[0023] S35. Taking the remaining refrigeration equipment in the warehouse corresponding to the demand overlap greater than 0.5 as the recommended refrigeration equipment, and constructing a second recommended refrigeration equipment set;

[0024] S36: Use the demand overlaps corresponding to the recommended refrigeration equipment in the second recommended refrigeration equipment set as elements to construct a second overlap vector.

[0025] Further, the AI ​​scheduling allocation model in S4 includes: a first equipment feature extraction unit, a second equipment feature extraction unit, a food semantic extraction unit, a first coincidence applying unit, a second coincidence applying unit, a first combined semantic extraction unit, a second combined semantic extraction unit, and a classification unit;

[0026] The input end of the first equipment feature extraction unit is used to input the first recommended refrigeration equipment set, and the output end thereof is connected to the first input end of the first coincidence applying unit;

[0027] The second input end of the first coincidence applying unit is used to input the first coincidence vector;

[0028] The input end of the food semantic extraction unit is used to input the current food characteristic set;

[0029] The input end of the second equipment feature extraction unit is used to input the second recommended refrigeration equipment set, and the output end thereof is connected to the first input end of the second coincidence applying unit;

[0030] The second input end of the second coincidence applying unit is used to input a second coincidence vector;

[0031] The output end of the food semantic extraction unit is connected to the first input end of the first combined semantic extraction unit and the first input end of the second combined semantic extraction unit respectively;

[0032] The second input end of the first combined semantic extraction unit is connected to the output end of the first coincidence applying unit;

[0033] The second input end of the second combined semantic extraction unit is connected to the output end of the second coincidence applying unit;

[0034] The input end of the classification unit is connected to the output end of the first combined semantic extraction unit and the output end of the second combined semantic extraction unit respectively, and its output end serves as the output end of the AI ​​scheduling allocation model.

[0035] Furthermore, the first device feature extraction unit, the second device feature extraction unit and the food semantic extraction unit each include: a plurality of LSTM networks, a plurality of phrase feature output layers and a concatenation layer;

[0036] An LSTM network is used to process a phrase vector. The output of an LSTM network is concatenated to obtain a feature vector, where a Cell unit in the LSTM network inputs an element in the phrase vector.

[0037] The phrase feature output layer is used to process the feature vector to obtain the phrase feature;

[0038] The concatenation layer is used to concatenate the phrase features output by multiple phrase feature output layers to obtain an output vector.

[0039] Furthermore, the expression of the phrase feature output layer is: , where r is the phrase feature, h i is the i-th element in the eigenvector, ω h,i is the phrase feature output layer for h i The weight assigned, b h,i is the phrase feature output layer for h i The bias assigned, i is a positive integer, and N is the number of elements in the feature vector.

[0040] Furthermore, the expression of the first overlap applying unit is: , where X1 is the output vector of the first coincidence applying unit, C1 is the output vector of the first device feature extraction unit, R1 is the first coincidence vector, is element-wise multiplication;

[0041] The expression of the second coincidence application unit is: , where X2 is the output vector of the second coincidence applying unit, C2 is the output vector of the second device feature extraction unit, and R2 is the second coincidence vector.

[0042] Furthermore, the expression of the combined semantic extraction unit is:

[0043] ,

[0044] ,

[0045] Among them, y1 is the first feature value output by the combined semantic extraction unit, y2 is the second feature value output by the combined semantic extraction unit, and x 1,i is the i-th element of the first input terminal of the combined semantic extraction unit, x 2,i is the i-th element input at the second input terminal of the combined semantic extraction unit, f is the fully connected layer, σ is the S-type activation function, tanh is the hyperbolic tangent activation function, M1 is the number of elements input at the first input terminal of the combined semantic extraction unit, M2 is the number of elements input at the second input terminal of the combined semantic extraction unit, W1 is the first weight in the combined semantic extraction unit, W2 is the second weight in the combined semantic extraction unit, W3 is the third weight in the combined semantic extraction unit, and W4 is the fourth weight in the combined semantic extraction unit.

[0046] A device for AI scheduling and allocation based on fresh food orders, comprising: a food characteristic extraction system, a first recommendation system, a second recommendation system and an allocation system;

[0047] The food property extraction system is used to extract food property information on fresh food orders and build the current food property set;

[0048] The first recommendation system is used to calculate the characteristic overlap between the current food characteristic set and the food characteristic set corresponding to the historical fresh food order, screen and recommend refrigeration equipment, and obtain a first recommended refrigeration equipment set and a first overlap vector;

[0049] The second recommendation system is used to extract user demand information on fresh food orders, calculate the overlap between user demand and demand of remaining refrigeration equipment in the warehouse, and obtain a second recommended refrigeration equipment set and a second overlap vector;

[0050] The allocation system is used to extract semantics from the current food characteristic set, the first recommended refrigeration equipment set and the second recommended refrigeration equipment set using an AI scheduling allocation model, and obtain the allocated refrigeration equipment based on the weights applied to the first overlap vector and the second overlap vector.

[0051] The beneficial effects of the present invention are:

[0052] 1. The present invention calculates the degree of characteristic overlap between the current food characteristic set and the food characteristic set corresponding to the historical fresh food order, screens the refrigeration equipment used for fresh food with similar food characteristics in history, and constructs a first recommended refrigeration equipment set and a first overlap vector. The first recommended refrigeration equipment set reflects the type of refrigeration equipment, and the first overlap vector reflects the fit between the refrigeration equipment in the first recommended refrigeration equipment set and the current food characteristic set.

[0053] 2. The present invention calculates the overlap between user demand and the demand of remaining refrigeration equipment in the warehouse, screens out refrigeration equipment that meets user demand, and obtains a second recommended refrigeration equipment set and a second overlap vector. The second recommended refrigeration equipment set reflects refrigeration equipment that meets user demand, and the second overlap vector reflects the compatibility of the second recommended refrigeration equipment set with user demand.

[0054] 3. The AI ​​scheduling and allocation model of the present invention refers to two sets of recommended refrigeration equipment and the current set of food characteristics to analyze the refrigeration equipment that should be allocated to the current fresh food order, and adjusts the importance of different recommended refrigeration equipment in the model based on the weights applied by the first coincidence vector and the second coincidence vector, thereby further improving the accuracy of the allocated refrigeration equipment.

[0055] 4. The present invention can achieve accurate matching of refrigeration equipment through in-depth analysis of food characteristics and user needs of fresh food orders. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1A flowchart of a method for AI scheduling and allocation based on fresh food orders;

[0057] Figure 2 This is a schematic diagram of the structure of the AI ​​scheduling allocation model;

[0058] Figure 3 It is a structural schematic diagram of a first device feature extraction unit, a second device feature extraction unit and a food semantic extraction unit. DETAILED DESCRIPTION

[0059] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0060] Embodiment 1, as Figure 1 As shown, a method for AI scheduling and allocation based on fresh food orders includes the following steps:

[0061] S1, extracting food characteristic information on fresh food orders and constructing a current food characteristic set;

[0062] S2, calculating the overlap between the current food feature set and the food feature set corresponding to the historical fresh food order, screening the recommended refrigeration equipment, and obtaining a first recommended refrigeration equipment set and a first overlap vector;

[0063] S3, extracting user demand information on fresh food orders, calculating the overlap between user demand and demand of remaining refrigeration equipment in the warehouse, and obtaining a second recommended refrigeration equipment set and a second overlap vector;

[0064] S4. Use the AI ​​scheduling allocation model to extract semantics from the current food feature set, the first recommended refrigeration equipment set, and the second recommended refrigeration equipment set, and obtain the allocated refrigeration equipment based on the weights applied to the first coincidence vector and the second coincidence vector.

[0065] In this embodiment, the food characteristic information in S1 includes: food type, food name, shelf life, storage temperature, food volume and food perishability;

[0066] The process of constructing the current food characteristic set in S1 is specifically as follows: segmenting the text corresponding to the food characteristic information according to food type, food name, shelf life, storage temperature, food volume and food perishability to obtain multiple phrases, vectorizing each phrase to obtain a phrase vector, and arranging the multiple phrase vectors in sequence to obtain the current food characteristic set.

[0067] For example, there is a fresh food order that contains the following information: food type: fruit, food name: imported Chilean blueberries, shelf life: 7 days, storage temperature: 0-4℃, food volume: 500ml, food perishability: high; the word segmentation results are: fruit, imported Chilean blueberries, 7 days, 0-4℃, 500ml, high, that is, 6 phrases are obtained.

[0068] Refrigeration equipment includes: refrigerators, refrigerated boxes, freezers, deep freezers, fresh-keeping cabinets, constant temperature and humidity cabinets, refrigerated containers, intelligent cold storage and biosafety refrigerators, etc.

[0069] In this embodiment, S2 includes the following sub-steps:

[0070] S21, taking the intersection of the current food characteristic set and the food characteristic set corresponding to each historical fresh food order, and counting the number of phrase vectors in the intersection;

[0071] S22, taking the ratio of the number of phrase vectors in the intersection to the total number of phrase vectors in the current food characteristic set as the characteristic overlap;

[0072] S23, extracting refrigeration equipment that matches historical fresh food orders and has a feature overlap greater than 0.5 as a set of candidate refrigeration equipment;

[0073] S24, extracting the remaining refrigeration equipment in the warehouse and constructing a set of remaining refrigeration equipment in the warehouse;

[0074] S25, taking the intersection of the candidate refrigeration equipment set and the remaining refrigeration equipment set in the warehouse to obtain a first recommended refrigeration equipment set;

[0075] S26: Using the characteristic coincidences corresponding to the refrigeration equipment in the first recommended refrigeration equipment set as elements to construct a first coincidence vector.

[0076] In step S26, if there are multiple characteristic overlaps corresponding to the same type of refrigeration equipment, the maximum characteristic overlap is used.

[0077] The present invention takes the intersection of the current food characteristic set and the historical food characteristic set, and can count the number of identical phrase vectors. The more phrase vectors there are in the intersection, the higher the characteristic overlap, and the historical fresh food orders with characteristic overlap greater than 0.5 are screened out. The refrigeration equipment configured for these historical fresh food orders is the refrigeration equipment used for foods with similar characteristics to the current food, but some of these refrigeration equipment have been used. Therefore, the intersection of the candidate refrigeration equipment set and the remaining refrigeration equipment set in the warehouse is taken to obtain the first recommended refrigeration equipment set, and the first overlap vector is constructed according to the characteristic overlap, so that different refrigeration equipment has different recommendation weights.

[0078] For example, the first recommended refrigeration equipment set is: {freezer, refrigerator, refrigerated box, fresh-keeping cabinet}. The second recommended refrigeration equipment set is: {fresh-keeping cabinet, refrigerator, refrigerated box}. When entering the AI ​​scheduling allocation model for processing, each refrigeration device in the recommended refrigeration equipment set is vectorized to obtain a phrase vector.

[0079] In this embodiment, S3 includes the following sub-steps:

[0080] S31, extracting user demand information on fresh food orders;

[0081] S32, performing word segmentation processing on the user demand information to obtain multiple phrases, vectorizing each phrase to obtain a phrase vector, and arranging the multiple phrase vectors in sequence to obtain a user demand set;

[0082] In the present invention, vectorization processing is to encode text into computer-recognizable code values;

[0083] S33, taking the intersection of the user demand set and the performance set of the remaining refrigeration equipment in each warehouse, and counting the number of phrase vectors in the intersection;

[0084] S34, taking the ratio of the number of phrase vectors in the intersection to the total number of phrase vectors in the user demand set as the demand overlap;

[0085] S35. Taking the remaining refrigeration equipment in the warehouse corresponding to the demand overlap greater than 0.5 as the recommended refrigeration equipment, and constructing a second recommended refrigeration equipment set;

[0086] S36: Use the demand overlaps corresponding to the recommended refrigeration equipment in the second recommended refrigeration equipment set as elements to construct a second overlap vector.

[0087] The present invention extracts user demand information on fresh food orders, divides it into multiple phrases, takes the intersection of the user demand set and the performance set of the remaining refrigeration equipment in each warehouse, and can count the number of phrase vectors in the intersection. The more phrase vectors in the intersection, the greater the demand overlap. The remaining refrigeration equipment in the warehouse with a demand overlap greater than 0.5 is screened out, and a second recommended refrigeration equipment set and a second overlap vector are constructed. The second overlap vector reflects that different refrigeration equipment has different recommendation weights in meeting user needs.

[0088] For example, the user demand information is: Refrigeration temperature: 2-6℃, preservation time: long-term, space requirement: medium, humidity control: low humidity, shockproof requirement: high, and after word segmentation processing, it is: 2-6℃, long-term, medium, low humidity, high. The corresponding text information of the performance set of refrigeration equipment is: Refrigeration temperature: 0-5℃, preservation time: short-term, space requirement: medium, humidity control: low humidity, shockproof requirement: medium. Then the total number of phrase vectors of the user demand set is: 5, and the intersection phrase vectors with refrigeration equipment are: "low humidity" and "medium", the number of intersection phrase vectors is: 2, and the demand overlap degree = 2 / 5 = 0.4 (lower than 0.5).

[0089] Performance sets are professional and systematic performance labels for refrigeration equipment in advance, including key indicators such as temperature control, storage capacity, humidity regulation, and shockproof performance. Users in fresh food orders can directly refer to these pre-labeled performance sets and quickly match the most suitable refrigeration equipment according to their specific needs to achieve accurate cold chain storage needs.

[0090] like Figure 2 As shown, the AI ​​scheduling allocation model in S4 includes: a first equipment feature extraction unit, a second equipment feature extraction unit, a food semantic extraction unit, a first coincidence applying unit, a second coincidence applying unit, a first combined semantic extraction unit, a second combined semantic extraction unit and a classification unit;

[0091] The input end of the first equipment feature extraction unit is used to input the first recommended refrigeration equipment set, and the output end thereof is connected to the first input end of the first coincidence applying unit;

[0092] The second input end of the first coincidence applying unit is used to input the first coincidence vector;

[0093] The input end of the food semantic extraction unit is used to input the current food characteristic set;

[0094] The input end of the second equipment feature extraction unit is used to input the second recommended refrigeration equipment set, and the output end thereof is connected to the first input end of the second coincidence applying unit;

[0095] The second input end of the second coincidence applying unit is used to input a second coincidence vector;

[0096] The output end of the food semantic extraction unit is connected to the first input end of the first combined semantic extraction unit and the first input end of the second combined semantic extraction unit respectively;

[0097] The second input end of the first combined semantic extraction unit is connected to the output end of the first coincidence applying unit;

[0098] The second input end of the second combined semantic extraction unit is connected to the output end of the second coincidence applying unit;

[0099] The input end of the classification unit is connected to the output end of the first combined semantic extraction unit and the output end of the second combined semantic extraction unit respectively, and its output end serves as the output end of the AI ​​scheduling allocation model.

[0100] The present invention uses a first device feature extraction unit to extract semantics from a first recommended refrigeration device set, uses a second device feature extraction unit to extract semantics from a second recommended refrigeration device set, and uses a food semantic extraction unit to extract semantics from a current food characteristic set. In a first coincidence applying unit, a weight is applied to the output of the first device feature extraction unit according to a first coincidence vector to adjust the importance of different features. In a second coincidence applying unit, the present invention applies a weight to the output of the second device feature extraction unit according to a second coincidence vector to adjust the importance of different features.

[0101] The present invention obtains a first type of semantic feature by synthesizing the semantics of a first recommended refrigeration equipment set and a current food characteristic set through a first combined semantic extraction unit; the present invention obtains a second type of semantic feature by synthesizing the semantics of a second recommended refrigeration equipment set and a current food characteristic set through a second combined semantic extraction unit; and then obtains the assigned refrigeration equipment by synthesizing the first type of semantic feature and the second type of semantic feature through a classification unit.

[0102] In this embodiment, the classification unit may specifically adopt a fully connected layer and a BP neural network, etc.

[0103] In this embodiment, the weights and biases in the AI ​​scheduling allocation model are trained using the existing gradient descent method.

[0104] like Figure 3 As shown, the first device feature extraction unit, the second device feature extraction unit and the food semantic extraction unit all include: multiple LSTM networks, multiple phrase feature output layers and a concatenation layer;

[0105] An LSTM network is used to process a phrase vector. The output of an LSTM network is concatenated to obtain a feature vector, where a Cell unit in the LSTM network inputs an element in the phrase vector.

[0106] The phrase feature output layer is used to process the feature vector to obtain the phrase feature;

[0107] The concatenation layer is used to concatenate the phrase features output by multiple phrase feature output layers to obtain an output vector.

[0108] In an LSTM network, there are multiple connected Cell units. Each Cell unit processes one element, and each Cell unit outputs a feature value, a feature value corresponding to a phrase vector.

[0109] In this embodiment, the expression of the phrase feature output layer is: , where r is the phrase feature, h i is the i-th element in the eigenvector, ω h,i is the phrase feature output layer for h i The weight assigned, b h,i is the phrase feature output layer for h i The bias assigned, i is a positive integer, and N is the number of elements in the feature vector.

[0110] In this embodiment, the expression of the first coincidence applying unit is: , where X1 is the output vector of the first coincidence applying unit, C1 is the output vector of the first device feature extraction unit, R1 is the first coincidence vector, is element-wise multiplication;

[0111] The expression of the second coincidence application unit is: , where X2 is the output vector of the second coincidence applying unit, C2 is the output vector of the second device feature extraction unit, and R2 is the second coincidence vector.

[0112] The present invention realizes that one phrase vector corresponds to one eigenvalue through processing by the first device feature extraction unit, the second device feature extraction unit and the food semantic extraction unit. Therefore, by multiplying the elements in the first coincidence applying unit and the second coincidence applying unit, weights can be applied to the corresponding eigenvalues ​​to adjust the sizes of different eigenvalues.

[0113] In this embodiment, the expressions of the combined semantic extraction units are all:

[0114] ,

[0115] ,

[0116] Among them, y1 is the first feature value output by the combined semantic extraction unit, y2 is the second feature value output by the combined semantic extraction unit, and x 1,i is the i-th element of the first input terminal of the combined semantic extraction unit, x 2,i is the i-th element input at the second input terminal of the combined semantic extraction unit, f is the fully connected layer, σ is the S-type activation function, tanh is the hyperbolic tangent activation function, M1 is the number of elements input at the first input terminal of the combined semantic extraction unit, M2 is the number of elements input at the second input terminal of the combined semantic extraction unit, W1 is the first weight in the combined semantic extraction unit, W2 is the second weight in the combined semantic extraction unit, W3 is the third weight in the combined semantic extraction unit, and W4 is the fourth weight in the combined semantic extraction unit.

[0117] The present invention adopts two types of activation functions to process two inputs of the combined semantic extraction unit respectively, obtains the combined semantic features, and improves the accuracy of semantic extraction.

[0118] The number of eigenvalues ​​for the classification unit input is 4.

[0119] Embodiment 2, a device for AI scheduling and allocation based on fresh food orders, comprising: a food characteristic extraction system, a first recommendation system, a second recommendation system and an allocation system;

[0120] The food property extraction system is used to extract food property information on fresh food orders and build the current food property set;

[0121] The first recommendation system is used to calculate the characteristic overlap between the current food characteristic set and the food characteristic set corresponding to the historical fresh food order, screen and recommend refrigeration equipment, and obtain a first recommended refrigeration equipment set and a first overlap vector;

[0122] The second recommendation system is used to extract user demand information on fresh food orders, calculate the overlap between user demand and demand of remaining refrigeration equipment in the warehouse, and obtain a second recommended refrigeration equipment set and a second overlap vector;

[0123] The allocation system is used to extract semantics from the current food characteristic set, the first recommended refrigeration equipment set and the second recommended refrigeration equipment set using an AI scheduling allocation model, and obtain the allocated refrigeration equipment based on the weights applied to the first overlap vector and the second overlap vector.

[0124] The specific implementation process of Example 2 is the same as that of Example 1.

[0125] The present invention calculates the degree of characteristic overlap between a current food characteristic set and a food characteristic set corresponding to historical fresh food orders, screens refrigeration equipment used for fresh food with similar food characteristics in history, and constructs a first recommended refrigeration equipment set and a first overlap vector. The first recommended refrigeration equipment set reflects the type of refrigeration equipment, and the first overlap vector reflects the fit between the refrigeration equipment in the first recommended refrigeration equipment set and the current food characteristic set.

[0126] The present invention calculates the overlap between user demand and the demand of remaining refrigeration equipment in the warehouse, screens out refrigeration equipment that meets the user demand, and obtains a second recommended refrigeration equipment set and a second overlap vector. The second recommended refrigeration equipment set reflects the refrigeration equipment that meets the user demand, and the second overlap vector reflects the fit between the second recommended refrigeration equipment set and the user demand.

[0127] The AI ​​scheduling and allocation model of the present invention refers to two recommended refrigeration equipment sets and the current food characteristic set, analyzes the refrigeration equipment that should be allocated to the current fresh food order, and adjusts the importance of different recommended refrigeration equipment in the model based on the weights applied by the first coincidence vector and the second coincidence vector, thereby further improving the accuracy of the allocated refrigeration equipment.

[0128] The present invention can achieve accurate matching of refrigeration equipment through in-depth analysis of food characteristics of fresh food orders and user needs.

[0129] The above are only recommended embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for AI scheduling and allocation based on fresh food orders, characterized in that: The following steps are involved: S1, extracting food characteristic information on fresh food orders and constructing a current food characteristic set; S2, calculating the overlap between the current food feature set and the food feature set corresponding to the historical fresh food order, screening the recommended refrigeration equipment, and obtaining a first recommended refrigeration equipment set and a first overlap vector; S3, extracting user demand information on fresh food orders, calculating the overlap between user demand and demand of remaining refrigeration equipment in the warehouse, and obtaining a second recommended refrigeration equipment set and a second overlap vector; S4. Using the AI ​​scheduling allocation model to extract semantics from the current food feature set, the first recommended refrigeration equipment set, and the second recommended refrigeration equipment set, and obtaining the allocated refrigeration equipment based on the weights applied by the first coincidence vector and the second coincidence vector; The food characteristic information in S1 includes: food type, food name, shelf life, storage temperature, food volume and food perishability; The process of constructing the current food characteristic set in S1 is specifically as follows: performing word segmentation processing on the text corresponding to the food characteristic information according to food type, food name, shelf life, storage temperature, food volume and food perishability to obtain multiple phrases, vectorizing each phrase to obtain a phrase vector, and arranging the multiple phrase vectors in sequence to obtain the current food characteristic set; The AI ​​scheduling allocation model in S4 includes: a first equipment feature extraction unit, a second equipment feature extraction unit, a food semantic extraction unit, a first overlap application unit, a second overlap application unit, a first combination semantic extraction unit, a second combination semantic extraction unit and a classification unit; The input end of the first equipment feature extraction unit is used to input the first recommended refrigeration equipment set, and the output end thereof is connected to the first input end of the first coincidence applying unit; The second input end of the first coincidence applying unit is used to input a first coincidence vector; The input end of the food semantic extraction unit is used to input the current food characteristic set; The input end of the second equipment feature extraction unit is used to input the second recommended refrigeration equipment set, and the output end thereof is connected to the first input end of the second coincidence applying unit; The second input end of the second coincidence applying unit is used to input a second coincidence vector; The output end of the food semantic extraction unit is connected to the first input end of the first combined semantic extraction unit and the first input end of the second combined semantic extraction unit respectively; The second input end of the first combined semantic extraction unit is connected to the output end of the first coincidence applying unit; The second input end of the second combined semantic extraction unit is connected to the output end of the second coincidence applying unit; The input end of the classification unit is connected to the output end of the first combined semantic extraction unit and the output end of the second combined semantic extraction unit respectively, and the output end thereof serves as the output end of the AI ​​scheduling allocation model; The first device feature extraction unit, the second device feature extraction unit and the food semantic extraction unit all include: multiple LSTM networks, multiple phrase feature output layers and a splicing layer; One of the LSTM networks is used to process a phrase vector, and the output of one LSTM network is concatenated to obtain a feature vector, wherein one Cell unit in the LSTM network inputs one element in the phrase vector; A phrase feature output layer is used to process a feature vector to obtain a phrase feature; The concatenation layer is used to concatenate the phrase features output by the multiple phrase feature output layers to obtain an output vector; The expression of the phrase feature output layer is: , where r is the phrase feature, h i is the i-th element in the eigenvector, ω h,i is the phrase feature output layer for h i The weight assigned, b h,i is the phrase feature output layer for h i The bias assigned, i is a positive integer, N is the number of elements in the feature vector; The expression of the first coincidence applying unit is: , where X1 is the output vector of the first coincidence applying unit, C1 is the output vector of the first device feature extraction unit, R1 is the first coincidence vector, is element-wise multiplication; The expression of the second coincidence applying unit is: , wherein X2 is the output vector of the second coincidence applying unit, C2 is the output vector of the second device feature extracting unit, and R2 is the second coincidence vector; The expressions of the combined semantic extraction units are: ， ， Among them, y1 is the first feature value output by the combined semantic extraction unit, y2 is the second feature value output by the combined semantic extraction unit, and x 1,i is the i-th element of the first input terminal of the combined semantic extraction unit, x 2,i is the i-th element input at the second input terminal of the combined semantic extraction unit, f is the fully connected layer, σ is the S-type activation function, tanh is the hyperbolic tangent activation function, M1 is the number of elements input at the first input terminal of the combined semantic extraction unit, M2 is the number of elements input at the second input terminal of the combined semantic extraction unit, W1 is the first weight in the combined semantic extraction unit, W2 is the second weight in the combined semantic extraction unit, W3 is the third weight in the combined semantic extraction unit, and W4 is the fourth weight in the combined semantic extraction unit.

2. The method for AI scheduling and allocation based on fresh food orders according to claim 1 is characterized in that: The S2 comprises the following sub-steps: S21, taking the intersection of the current food characteristic set and the food characteristic set corresponding to each historical fresh food order, and counting the number of phrase vectors in the intersection; S22, taking the ratio of the number of phrase vectors in the intersection to the total number of phrase vectors in the current food characteristic set as the characteristic overlap; S23, extracting refrigeration equipment that matches historical fresh food orders and has a feature overlap greater than 0.5 as a set of candidate refrigeration equipment; S24, extracting the remaining refrigeration equipment in the warehouse and constructing a set of remaining refrigeration equipment in the warehouse; S25, taking the intersection of the candidate refrigeration equipment set and the remaining refrigeration equipment set in the warehouse to obtain a first recommended refrigeration equipment set; S26: Using the characteristic coincidences corresponding to the refrigeration equipment in the first recommended refrigeration equipment set as elements to construct a first coincidence vector.

3. The method for AI scheduling and allocation based on fresh food orders according to claim 1 is characterized in that: The S3 comprises the following sub-steps: S31, extracting user demand information on fresh food orders; S32, performing word segmentation processing on the user demand information to obtain multiple phrases, vectorizing each phrase to obtain a phrase vector, and arranging the multiple phrase vectors in sequence to obtain a user demand set; S33, taking the intersection of the user demand set and the performance set of the remaining refrigeration equipment in each warehouse, and counting the number of phrase vectors in the intersection; S34, taking the ratio of the number of phrase vectors in the intersection to the total number of phrase vectors in the user demand set as the demand overlap; S35. Taking the remaining refrigeration equipment in the warehouse corresponding to the demand overlap greater than 0.5 as the recommended refrigeration equipment, and constructing a second recommended refrigeration equipment set; S36: Use the demand overlaps corresponding to the recommended refrigeration equipment in the second recommended refrigeration equipment set as elements to construct a second overlap vector.

4. A device for AI scheduling and allocation based on fresh food orders, implemented based on the method for AI scheduling and allocation based on fresh food orders according to any one of claims 1 to 3, characterized in that: include: Food property extraction system, first recommendation system, second recommendation system and allocation system; The food property extraction system is used to extract food property information on fresh food orders and construct a current food property set; The first recommendation system is used to calculate the characteristic overlap between the current food characteristic set and the food characteristic set corresponding to the historical fresh food order, screen the recommended refrigeration equipment, and obtain a first recommended refrigeration equipment set and a first overlap vector; The second recommendation system is used to extract user demand information on fresh food orders, calculate the overlap between user demand and demand of remaining refrigeration equipment in the warehouse, and obtain a second recommended refrigeration equipment set and a second overlap vector; The allocation system is used to extract semantics from the current food characteristic set, the first recommended refrigeration equipment set and the second recommended refrigeration equipment set using an AI scheduling allocation model, and obtain the allocated refrigeration equipment based on the weights applied to the first coincidence vector and the second coincidence vector.

Images