Artificial intelligence model construction method for intelligent selling
By building a large AI model, the problem of initial debugging errors in smart vending equipment was solved, automatic parameter matching and intelligent heating were achieved, operational efficiency and sales results were improved, and corporate risks were reduced.
Patent Information
- Application Number
- CN202510601541.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-12
- Publication Date
- 2025-09-12
AI Technical Summary
Existing smart vending equipment has human errors during initial factory debugging and is unable to provide marketing strategies and intelligent parameter configurations, resulting in low operational efficiency and poor sales results.
Build a large AI model through data collection, cleaning and organization, use deep neural network architecture, combine high-performance computing hardware for training and hyperparameter tuning, optimize model parameters, apply model quantization and knowledge distillation technology, realize automatic parameter allocation and intelligent heating, and make intelligent recommendations based on sensor data.
It realizes automatic parameter matching of smart vending equipment, improves operational efficiency, ensures optimal food temperature, formulates delivery and replenishment plans in a timely manner, reduces corporate risks, and provides intelligent marketing and recommendation capabilities.
Smart Images
Figure CN120632445A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of artificial intelligence model technology, and in particular to a method for constructing an artificial intelligence model for smart sales. Background Art
[0002] With the continuous development of artificial intelligence (AI) technology, the construction of large AI models (artificial intelligence models) has become a hot topic across various industries. Large AI models are characterized by large scale, high complexity, and extensive structure. These models require extensive data and computing resources during training to achieve learning and optimization. Existing smart vending equipment is often manually debugged during initial factory delivery, which can lead to human errors and an inability to provide customers with marketing strategies and parameter configurations that match real-world scenarios. This results in low project operational efficiency, a lack of intelligent recommendation capabilities, and poor overall sales performance. Summary of the Invention
[0003] In view of this, the present invention provides an artificial intelligence model construction method for intelligent sales to solve complex technical problems such as human errors in the initial debugging of the existing technology and the inability to provide marketing strategies and intelligent parameter configurations.
[0004] The present invention provides a method for constructing an artificial intelligence model for smart vending, which is applied to hot food smart vending equipment. The method comprises: step 1, collecting meal data, temperature data, heating parameters, user preferences, and venue data based on the equipment data platform and various sensors of the smart vending equipment; step 2, cleaning and organizing the collected data, and constructing an artificial intelligence model using a deep neural network architecture; step 3, using high-performance computing hardware to train the artificial intelligence model, focusing on training the differences in machine parameter configuration strategies under different scenarios to achieve parameter optimization of the artificial intelligence model; step 4, using large model hyperparameter tuning technology to optimize the artificial intelligence model algorithm; step 5, after the algorithm optimization is completed, evaluating and verifying the artificial intelligence model through the following function formula, ， , , Among them, Perplexity is the perplexity value, total_loss is the cross entropy loss, BLEU is the BLEU score, PB is the penalty factor, BLEU_N is the accuracy of N consecutive words, Accuracy is the accuracy rate, correct_predictions is the number of predictions, and total_predictions is the total number of samples; Step 6, after the evaluation is completed, remove redundant parameters in the artificial intelligence model, use model quantization and knowledge distillation technology to reduce the computational complexity and storage requirements of the artificial intelligence model, so that it can run on resource-constrained smart vending equipment, thereby realizing the functions of automatic parameter allocation, intelligent heating, and formulation of distribution plans.
[0005] Furthermore, the step 4 includes: step 41, automatically searching for the optimal learning rate, batch size and network depth; step 42, using the cosine annealing learning rate to adjust the learning rate of the model training to avoid the training falling into the local optimum; step 43, dividing the neural network into an input layer, a shallow layer, an intermediate layer, a deep layer and an output layer, performing L2 regularization in the input layer and the shallow layer to suppress the complexity of the model, performing a balanced use of L2 regularization and Dropout technology in the intermediate layer, using Dropout technology in the deep layer to close the neurons and their links in the neural network, and the output layer outputs the processed data, and prevents overfitting through the combination of L2 regularization and Dropout technology; step 44, based on the performance of the validation set, determining the timing of training termination to achieve algorithm optimization and model effect enhancement of the artificial intelligence model.
[0006] Furthermore, step 2 also includes dividing the collected data into a training set, a validation set and a test set, the training set is used for model training, the validation set is used for parameter adjustment during the training process, and the test set is used to evaluate the final effect of the model.
[0007] Furthermore, step 1 also includes: performing data cleaning, data labeling, and data normalization preprocessing on the collected data to improve the quality of the data.
[0008] Furthermore, the sensors include: a temperature sensor, a humidity sensor, a pickup cart motion sensor, a microwave heating sensor, and an infrared photoelectric signal sensor.
[0009] Furthermore, the method also includes: step 7, deploying the artificial intelligence model to the smart vending equipment to realize the automatic parameter allocation, intelligent heating, and distribution plan formulation functions of the smart vending equipment.
[0010] Furthermore, the step 7 specifically includes: step 71, deploying the artificial intelligence model to the smart vending equipment to automatically match the equipment factory parameters, pickup cart operating parameters, refrigerator freezing parameters, and defrosting parameters; step 72, through the data classification of the artificial intelligence model, the smart vending equipment adjusts the intelligent heating parameters of the food according to the current scenario to ensure that the food sales temperature reaches the preset optimal taste temperature; step 73, based on the real-time sales data of the smart vending equipment, the artificial intelligence model is used to predict and analyze future sales and formulate distribution and replenishment plans.
[0011] Furthermore, the method also includes: step 8, providing the device with a product push plan and price combination that meets the current scenario based on the digital base provided by the artificial intelligence model.
[0012] Furthermore, the current scene includes the current ambient temperature, food sales situation, and food types.
[0013] The present invention provides a method for constructing an artificial intelligence model for smart vending. By constructing a large AI model, the method enables smart vending equipment to automatically match parameters, reduce equipment debugging time, and improve machine operating efficiency. It can also enable smart vending equipment distributed across the country to intelligently adjust the heating parameters of meals according to local ambient temperature, sales conditions, and meal types, to ensure that the meal selling temperature reaches the optimal taste temperature. The technical solution can also timely predict the sales situation of each device through real-time analysis of real-time sales data, scientifically formulate intelligent distribution and replenishment plans, and provide production references for meal supply chain companies to reduce corporate risks. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] Figure 1 This is a flow chart of a method for building an artificial intelligence model for smart selling provided by the present invention; Figure 2 This is a flow chart of the method for deploying an artificial intelligence model on a smart vending device provided by the present invention; Figure 3 This is a flow chart of a method for algorithm optimization of an artificial intelligence model in training provided by the present invention; Figure 4 This is the neuron architecture diagram of the artificial intelligence model provided by the present invention. DETAILED DESCRIPTION
[0015] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.
[0016] Example 1: This application provides a method for constructing an artificial intelligence model for smart vending, which is applied to hot food vending equipment. This method uses an AI big model to combine the software and hardware of the hot food vending machine, that is, the hot food smart vending equipment, to achieve the comprehensive application of AI technology and smart vending equipment. Figure 1 As shown, the method includes the following steps.
[0017] Step 1: Collect meal data, temperature data, heating parameters, user preferences, and venue data based on the device data platform and various sensors of the smart vending equipment; Step 2: Clean and organize the collected data and build an artificial intelligence model using a deep neural network architecture; Step 3: Use high-performance computing hardware to train the AI model, focusing on the differences in machine parameter configuration strategies in different scenarios to achieve parameter optimization of the AI model; The focus of the training is on the differences in machine parameter configuration strategies in different scenarios, such as the differences in parameter configurations for airports and gas stations. Step 4: Use large-scale model hyperparameter tuning technology to optimize the AI model algorithm; Step 5: After the algorithm optimization is completed, the artificial intelligence model is evaluated and verified through the following function: , , , Among them, Perplexity is the perplexity value, total_loss is the cross entropy loss, BLEU is the BLEU score, PB is the penalty factor, BLEU_N is the accuracy of N consecutive words, Accuracy is the accuracy rate, correct_predictions is the number of predictions, and total_predictions is the total number of samples; Step 6: Remove redundant parameters from the AI model and use model quantization and knowledge distillation technology to reduce the computational complexity and storage requirements of the AI model, enabling it to run on resource-constrained smart vending equipment, thereby realizing the functions of automatic parameter allocation, intelligent heating, and formulation of distribution plans.
[0018] The present invention provides a method for constructing an artificial intelligence model for smart vending. By constructing a large AI model, the method enables smart vending equipment to automatically match parameters, reduce equipment debugging time, and improve machine operation efficiency. It can also enable smart vending equipment distributed across the country to intelligently adjust the heating parameters of meals according to the ambient temperature, sales situation, and meal types in different scenarios, to ensure that the meal selling temperature reaches the optimal taste temperature. The technical solution can also timely predict the sales situation of each device through real-time analysis of real-time sales data, scientifically formulate intelligent distribution and replenishment plans, and provide production references for meal supply chain companies to reduce corporate risks.
[0019] Example 2: This application provides a method for building an artificial intelligence model for smart vending, which is applied to hot food vending equipment, such as Figure 1 As shown, the method includes the following steps Step 1: Collect meal data, temperature data, heating parameters, user preferences, and venue data based on the device data platform and various sensors of smart vending equipment.
[0020] The construction of AI large models requires a large amount of real-time data from hardware devices as training samples, so the IoT sensors installed on the smart vending equipment need to provide the required data in real time. The IoT sensors include: temperature sensors, humidity sensors, pickup cart motion sensors, microwave heating sensors, and infrared light signal sensors. Figure 4 As shown in the figure, the hardware used for data collection on the hot food smart vending equipment mainly consists of four parts: refrigerator, pickup cart, microwave heating, and thermal insulation cabinet (the refrigerator module contains modules such as temperature sensor and humidity sensor, the pickup cart module contains photoelectric sensor and vehicle control module, the microwave module contains limit module, microwave control module, etc., and the thermal insulation cabinet module contains photoelectric sensor, etc.). Each part contains various types of digital sensors. The main control board of the equipment performs simple cleaning and preprocessing on the collected signals, and maintains real-time communication with the AI big model platform through the 5G smart gateway.
[0021] Step 2: Clean and organize the collected data and build an artificial intelligence model using a deep neural network architecture; After data collection is complete, it needs to be preprocessed, including data cleaning, data labeling, and data normalization. The purpose of data preprocessing is to improve data quality and lay a solid foundation for building large AI models. After data preprocessing is complete, the large AI model needs to be designed, including aspects such as the model's structure, algorithms, and optimization. The goal of model design is to achieve the model's learning and generalization capabilities to adapt to different application scenarios. Large AI models use a deep neural network architecture, which can capture long-range dependencies when processing sequential data and is particularly suitable for natural language processing and sequence generation tasks.
[0022] Step 3: Use high-performance computing hardware to train the AI model, focusing on the differences in machine parameter configuration strategies in different scenarios to achieve parameter optimization of the AI model; After the model design is completed, the AI large model needs to be trained to achieve model parameter optimization. During the training process, attention should be paid to the convergence of the model, the performance evaluation of the model, etc. The step 2 also includes dividing the collected data into a training set, a validation set, and a test set. The training set is used for model training, the validation set is used for parameter adjustment during the training process, and the test set is used to evaluate the final effect of the model. The training process of the AI large model is a long-term resource-intensive computing process. At present, high-performance computing hardware is used for the training of large models. At the same time, because the existing equipment is distributed over a wide area, the training process can be effectively accelerated by working together with multiple GPU or TPU nodes. The high-performance computing hardware refers to hardware that can perform complex mathematical operations and data processing tasks at an extremely fast speed, which is usually achieved by increasing the number of processor cores, optimizing the instruction set architecture, and adopting advanced semiconductor technology.
[0023] Step 4: Use large-scale model hyperparameter tuning technology to optimize the AI model algorithm; During the training process, the AI large model mainly uses hyperparameter tuning technology (hyperparameter tuning refers to the parameters set before model training, which improves the performance and effect of learning by selecting a set of optimal hyperparameters) to optimize the algorithm. In addition, the AI large model continuously improves the performance of the model through technologies such as learning rate scheduling, regularization, and dropout (dropout is a regularization technology commonly used in deep learning, mainly used to prevent overfitting of neural networks during training). Step 3 includes the following steps Step 41, automatically searching for the optimal learning rate, batch size and network depth; Step 42: Use cosine annealing learning rate to adjust the learning rate of model training to avoid training falling into local optimum; Step 43: The neural network is divided into an input layer, a shallow layer, an intermediate layer, a deep layer, and an output layer. L2 regularization is implemented in the input layer and the shallow layer to suppress the complexity of the model. A balanced use of L2 regularization and Dropout technology is performed in the intermediate layer. Dropout technology is used in the deep layer to shut down neurons and their links in the neural network. The output layer outputs the processed data. The combination of L2 regularization and Dropout technology prevents overfitting. L2 regularization, also known as weight decay, suppresses model complexity by adding a penalty term based on the sum of squared weights to the loss function. L2 regularization is implemented by adding an L2 penalty term to the original loss function and dynamically adjusting the regularization strength based on the training stage. Dropout is a regularization technique used to prevent overfitting in neural networks. Its core idea is to randomly "drop out" (temporarily shut down) a portion of neurons and their connections during training, reducing the network's overreliance on any particular neuron and improving the model's generalization ability. Neural networks can be divided into input, shallow, intermediate, deep, and output layers. The input layer applies to the raw input features, aiming to retain most of the original information while filtering out some noise. The shallow layers, applied at the front end of the network, are responsible for extracting basic features and preventing the model from over-relying on certain prominent features. The intermediate layers, applied in the middle of the network, handle feature combination and abstraction, balancing feature retention and overfitting prevention. The deep layers, applied at the back end of the network, extract highly abstract features. Dropout is typically not directly applied to the output layer to ensure stable prediction results. In some cases, a slight dropout may be applied to the last hidden layer before the output layer. This hierarchical and progressive strategy can effectively prevent overfitting while retaining the model's expressiveness and improve the model's generalization ability in different scenarios.
[0024] Step 44, based on the performance of the validation set, decide when to terminate the training to achieve algorithm optimization and model effect enhancement of the artificial intelligence model.
[0025] Step 5: After the algorithm optimization is completed, the artificial intelligence model is evaluated and verified through the following function: , , , Among them, Perplexity is the perplexity value, total_loss is the cross entropy loss, BLEU is the BLEU score, PB is the penalty factor, BLEU_N is the accuracy of N consecutive words, Accuracy is the accuracy rate, correct_predictions is the number of predictions, and total_predictions is the total number of samples; Step 6: After the evaluation is completed, redundant parameters in the AI model are removed, and model quantization and knowledge distillation techniques are used to reduce the computational complexity and storage requirements of the AI model, enabling it to run on resource-constrained smart vending equipment, thereby realizing the functions of automatic parameter allocation, intelligent heating, and formulation of distribution plans.
[0026] Step 7: Deploy the artificial intelligence model to the smart vending equipment to realize the functions of automatic parameter allocation, intelligent heating, and formulation of distribution plans for the smart vending equipment.
[0027] After the model training is completed, the AI big model is deployed to the smart vending equipment. That is, the AI big model is used to combine the hot food vending machine hardware and software to optimize the performance of the hot food vending machine hardware, intelligently adapt to the scene, and provide intelligent digital human customer service, thus realizing the comprehensive application of AI technology and smart vending equipment. Figure 2 As shown, the step 7 specifically includes: Step 71: Deploy the artificial intelligence model to the smart vending device to automatically match the device's factory parameters, pickup cart operating parameters, refrigerator freezing parameters, and defrosting parameters. Step 72: Through data classification by the artificial intelligence model, the intelligent vending device adjusts the food heating parameters according to the current scenario to ensure that the food vending temperature reaches the preset optimal taste temperature; Step 73: Based on the real-time sales data from the smart vending equipment, the artificial intelligence model is used to predict and analyze future sales and formulate distribution and replenishment plans.
[0028] Step 8: Based on the digital base provided by the artificial intelligence model, the device is provided with a product push plan and price combination that suits the current scenario.
[0029] The current scenario includes the current ambient temperature, food availability, and food types. The digital foundation of the AI model typically refers to the infrastructure and technical platform that supports model operation and data processing. This foundation encompasses multiple aspects, including data storage, processing, analysis, and application development.
[0030] The present invention provides a method for constructing an artificial intelligence model for smart vending. By constructing a large AI model, the method enables smart vending equipment to automatically match parameters, reduce equipment debugging time, and improve machine operating efficiency. It can also enable smart vending equipment distributed across the country to intelligently adjust the heating parameters of meals according to local ambient temperature, sales conditions, and meal types, to ensure that the meal selling temperature reaches the optimal taste temperature. The technical solution can also timely predict the sales situation of each device through real-time analysis of real-time sales data, scientifically formulate intelligent distribution and replenishment plans, and provide production references for meal supply chain companies to reduce corporate risks.
[0031] In summary, the embodiment of the present invention provides a method for constructing an artificial intelligence model for smart vending. This technical solution significantly increases the operating efficiency of smart vending equipment. Through the data classification of the AI big model, it can quickly adjust the equipment batches and site environment, significantly reduce the initial debugging time of the equipment, avoid human errors, and automatically optimize the system parameters according to the scene, etc., to ensure that the equipment can achieve maximum efficiency. In addition, the digital base provided by the big model can provide the equipment with a product push solution and price combination that is closer to the scene, and can realize intelligent recommendations and intelligent marketing for customers, so that operators can make scientific decisions. The AI big model constructed by this technical solution can also coordinate the supply chain and distribution as a whole, realize an intelligent one-stop production, sales, and delivery, significantly reduce labor costs, and improve overall social benefits. The establishment of the AI big model can realize the personalized demand display of the overall data, provide visual integrated services for participating units and personnel at all levels, and scientifically deduce the medium- and short-term development trends. It can realize intelligent planning reports for all participating companies, greatly improving the operational efficiency of the overall project.
[0032] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.
Claims
1. A method for constructing an artificial intelligence model for intelligent vending, applied to intelligent hot food vending equipment, characterized in that: The method comprises: Step 1: Collect meal data, temperature data, heating parameters, user preferences, and venue data based on the device data platform and various sensors of the smart vending equipment; Step 2: Clean and organize the collected data and build an artificial intelligence model using a deep neural network architecture; Step 3: Use high-performance computing hardware to train the AI model, focusing on the differences in machine parameter configuration strategies in different scenarios to achieve parameter optimization of the AI model; Step 4: Use large-scale model hyperparameter tuning technology to optimize the AI model algorithm; Step 5: After the algorithm optimization is completed, the artificial intelligence model is evaluated and verified through the following function: ， ， ， Among them, Perplexity is the perplexity value, total_loss is the cross entropy loss, BLEU is the BLEU score, PB is the penalty factor, BLEU_N is the accuracy of N consecutive words, Accuracy is the accuracy rate, correct_predictions is the number of predictions, and total_predictions is the total number of samples; Step 6: After the evaluation is completed, redundant parameters in the AI model are removed, and model quantization and knowledge distillation techniques are used to reduce the computational complexity and storage requirements of the AI model, enabling it to run on resource-constrained smart vending equipment, thereby realizing the functions of automatic parameter allocation, intelligent heating, and formulation of distribution plans.
2. The method for constructing an artificial intelligence model for smart selling according to claim 1, characterized in that: The step 4 comprises: Step 41, automatically searching for the optimal learning rate, batch size and network depth; Step 42: Use cosine annealing learning rate to adjust the learning rate of model training to avoid training falling into local optimum; Step 43: The neural network is divided into an input layer, a shallow layer, an intermediate layer, a deep layer, and an output layer. L2 regularization is implemented in the input layer and the shallow layer to suppress the complexity of the model. A balanced use of L2 regularization and Dropout technology is performed in the intermediate layer. Dropout technology is used in the deep layer to shut down neurons and their links in the neural network. The output layer outputs the processed data. The combination of L2 regularization and Dropout technology prevents overfitting. Step 44, based on the performance of the validation set, decide when to terminate the training to achieve algorithm optimization and model effect enhancement of the artificial intelligence model.
3. The method for constructing an artificial intelligence model for smart selling according to claim 1, characterized in that: The step 2 also includes dividing the collected data into a training set, a validation set and a test set, wherein the training set is used for model training, the validation set is used for parameter adjustment during the training process, and the test set is used to evaluate the final effect of the model.
4. The method for constructing an artificial intelligence model for smart selling according to claim 1, characterized in that: The step 1 also includes: performing data cleaning, data labeling, and data normalization preprocessing on the collected data to improve the quality of the data.
5. The method for constructing an artificial intelligence model for smart selling according to claim 1, characterized in that: The sensors include: a temperature sensor, a humidity sensor, a pickup trolley motion sensor, a microwave heating sensor, and an infrared photoelectric signal sensor.
6. The method for constructing an artificial intelligence model for smart selling according to claim 1, characterized in that: The method also includes: step 7, deploying the artificial intelligence model to the smart vending equipment to realize the automatic parameter allocation, intelligent heating, and distribution plan formulation functions of the smart vending equipment.
7. The method for constructing an artificial intelligence model for smart selling according to claim 6, characterized in that: The step 7 specifically includes: Step 71: Deploy the artificial intelligence model to the smart vending device to automatically match the device's factory parameters, pickup cart operating parameters, refrigerator freezing parameters, and defrosting parameters. Step 72: Through data classification by the artificial intelligence model, the intelligent vending device adjusts the food heating parameters according to the current scenario to ensure that the food vending temperature reaches the preset optimal taste temperature; Step 73: Based on the real-time sales data from the smart vending equipment, the artificial intelligence model is used to predict and analyze future sales and formulate distribution and replenishment plans.
8. The method for constructing an artificial intelligence model for smart selling according to claim 1, characterized in that: The method also includes: step 8, providing the device with a product push solution and price combination that meets the current scenario based on the digital base provided by the artificial intelligence model.
9. The method for constructing an artificial intelligence model for smart selling according to claim 7 or 8, characterized in that: The current scene includes the current ambient temperature, food sales situation, and food types.