Fresh food supply chain data analysis method and system based on big data and deep learning
Through the analysis method of fresh food supply chain data based on big data and deep learning, we can identify abnormal state of fresh food supply chain and analyze the relationship between key indicators and influencing factors, and solve the problem that traditional methods are difficult to cope with the complex fresh food supply chain environment, and improve the analysis Accuracy and efficiency optimize supply chain management.
Patent Information
- Application Number
- CN202410953113.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-16
- Publication Date
- 2024-11-15
AI Technical Summary
Traditional data analysis methods are difficult to effectively deal with the complex and changeable fresh food supply chain environment, resulting in problems such as large demand fluctuations, short shelf life and high loss rate.
Adopt the fresh food supply chain data analysis method based on big data and deep learning. By obtaining the basic indicator data of fresh food categories to be monitored, identifying abnormal situations, obtaining key indicators and alternative influencing factors, and using deep learning algorithms to analyze the impact dependencies. Optimize fresh food supply chain management decisions.
It improves the accuracy and efficiency of fresh food supply chain data analysis, helps optimize fresh food supply chain management decisions, promptly respond to abnormal situations, and improves overall operational efficiency.
Smart Images

Figure CN118966964A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a fresh food supply chain data analysis method and system based on big data and deep learning. Background Art
[0002] Fresh food supply chain management currently faces many challenges, including large fluctuations in demand, short shelf life, high loss rates, etc. Traditional data analysis methods often rely on limited historical data and empirical judgments, and are difficult to effectively cope with the complex and changing fresh food supply chain environment. Summary of the invention
[0003] The purpose of the present invention is to provide a fresh food supply chain data analysis method and system based on big data and deep learning.
[0004] In a first aspect, an embodiment of the present invention provides a fresh food supply chain data analysis method based on big data and deep learning, comprising:
[0005] Acquire multiple fresh food categories to be monitored, and acquire basic indicator data of each of the fresh food categories to be monitored;
[0006] When there is an abnormality in the fresh food supply chain status represented by the basic indicator data, the corresponding fresh food category to be monitored is used as the target fresh food category;
[0007] The key indicators and multiple candidate influencing factors of the target fresh food category are obtained, and the influence dependency relationship between the key indicators and the target influencing factors determined from the multiple candidate influencing factors is used as the fresh food supply chain data analysis result of the target fresh food category.
[0008] In a second aspect, an embodiment of the present invention provides a server system, including a server, wherein the server is used to execute the method described in the first aspect.
[0009] Compared with the prior art, the beneficial effects provided by the present invention include: using the fresh food supply chain data analysis method and system based on big data and deep learning disclosed in the embodiments of the present invention, first obtaining basic indicator data of multiple fresh food categories to be monitored, identifying abnormal conditions in the fresh food supply chain status through data analysis, and taking the corresponding fresh food categories as target analysis objects. Furthermore, the present invention obtains key indicators and multiple alternative influencing factors of the target fresh food category, and uses a deep learning algorithm to analyze the influence dependency relationship between the key indicators and the target influencing factors, thereby obtaining the fresh food supply chain data analysis results of the target fresh food category. Such a design improves the accuracy and efficiency of fresh food supply chain data analysis, and helps to optimize fresh food supply chain management decisions. BRIEF DESCRIPTION OF THE DRAWINGS
[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can also be obtained based on these drawings without creative work.
[0011] Figure 1 A schematic diagram of the steps of a fresh food supply chain data analysis method based on big data and deep learning provided by an embodiment of the present invention;
[0012] Figure 2 A schematic block diagram of the structure of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0013] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.
[0014] The specific implementation modes of the present invention are described in detail below in conjunction with the accompanying drawings.
[0015] In order to solve the technical problems in the aforementioned background technology, Figure 1 A flow chart of a fresh food supply chain data analysis method based on big data and deep learning provided in an embodiment of the present disclosure is provided below. The fresh food supply chain data analysis method based on big data and deep learning is introduced in detail.
[0016] Step S201, obtaining a plurality of fresh food categories to be monitored, and obtaining basic indicator data of each of the fresh food categories to be monitored;
[0017] Step S202, when there is an abnormality in the fresh food supply chain status represented by the basic indicator data, the corresponding fresh food category to be monitored is used as the target fresh food category;
[0018] Step S203, obtaining key indicators and multiple candidate influencing factors of the target fresh food category, and taking the influence dependency relationship between the key indicators and the target influencing factors determined from the multiple candidate influencing factors as the fresh food supply chain data analysis result of the target fresh food category.
[0019] In an embodiment of the present invention, illustratively, the server first automatically captures multiple fresh categories to be monitored from the fresh supply chain management system, such as vegetables (such as spinach, tomatoes), fruits (such as apples, bananas) and meat (such as pork, beef), etc. For each category, the server collects its basic indicator data, including but not limited to inventory, daily sales, loss rate, purchase price, sales price, and seasonal change trends. For example, for the category of spinach, the server will obtain its daily inventory, sales, loss rate due to decay, and sales fluctuations in different months of the past year. The server uses the built-in big data analysis engine to analyze the collected basic indicator data and identify categories with abnormal supply chain status. Assume that the spinach category has a serious inventory backlog in the past week, the loss rate has risen sharply, and the sales volume is far below the average level. These abnormal indicators trigger the server's early warning mechanism, and the server immediately identifies spinach as a target fresh category and prepares for in-depth analysis. For spinach, which was identified as a target fresh product, the server further collected its key indicators, such as daily sales volume and loss rate, and listed possible alternative influencing factors, such as weather changes, promotional activities, holiday effects, supplier stability, logistics transportation time, etc. The server used a deep learning model, combined with historical data and real-time data, to rank the importance of these alternative factors and conduct correlation analysis. After analysis, the server found that the recent continuous rainy weather was the main factor leading to the decline in spinach sales and the increase in loss rate. Specifically, the model revealed that rainy weather prolonged the transportation time of spinach, increased its in-transit loss, and reduced consumers' willingness to buy. The server finally identified "rainy weather" as the target influencing factor, and output the negative dependency between it and the two key indicators of "daily sales volume" and "loss rate" as the fresh supply chain data analysis result. Based on this analysis result, managers of the fresh supply chain can adjust inventory strategies in a timely manner, optimize logistics transportation routes, and even consider increasing promotional activities before rainy weather to stimulate consumption, thereby effectively responding to abnormal situations in the supply chain and improving overall operational efficiency.
[0020] In an embodiment of the present invention, the key indicators and multiple alternative influencing factors of the target fresh food category are obtained, and the influence dependency relationship between the key indicators and the target influencing factors determined from the multiple alternative influencing factors is used as the fresh food supply chain data analysis result of the target fresh food category. It can be implemented through the following examples.
[0021] Obtain key indicators and multiple candidate influencing factors of the target fresh food category;
[0022] Obtaining a supply chain data sample data set according to the relevant parameters corresponding to each of the candidate influencing factors and the key indicators;
[0023] Iterative feature engineering is performed according to the supply chain data sample data set to determine the target influencing factors screened stage by stage from each of the candidate influencing factors; wherein the supply chain data samples of the next stage of feature engineering include relevant parameters of the key indicators and relevant parameters of the pending influencing factors selected by the feature engineering of the previous stage, and the relevant parameters of the pending influencing factors selected by the feature engineering of the previous stage are different from the data configuration status of the relevant parameters of the influencing factors in the supply chain data samples of the feature engineering of the previous stage; the correlation strength between the pending influencing factors and the key indicators reaches a preset correlation strength threshold;
[0024] The influence dependency relationship between the key indicator and the target influencing factor is used as the fresh food supply chain data analysis result of the target fresh food category.
[0025] In an embodiment of the present invention, illustratively, for spinach previously identified as a target fresh product category, the server has determined that its key indicators are daily sales volume and loss rate. Next, the server collects alternative influencing factors related to spinach from multiple data sources, including weather changes (such as temperature, humidity, whether it rains), promotional activities (such as discounts, buy one get one free), holiday effects (such as the Spring Festival, the Mid-Autumn Festival), supplier stability (such as whether the supply is timely, whether the quality is stable), logistics transportation time (such as average transportation time, number of delays), etc. The server integrates the collected daily sales volume and loss rate of spinach with the relevant parameters of each alternative influencing factor to form a supply chain data sample data set. For example, the data set may contain daily records for the past year, each record including the weather conditions of the day (temperature, humidity, whether it rains), whether there are promotional activities, whether it is a holiday, supplier supply conditions, logistics transportation time, and the corresponding daily sales volume and loss rate. The server uses a machine learning algorithm to perform iterative feature engineering to screen the target influencing factors that have the greatest impact on the daily sales volume and loss rate of spinach stage by stage. In the first stage, the server can find that weather changes and promotions have a strong correlation with key indicators, so these two factors are selected to enter the next stage of feature engineering. In the second stage, the server further analyzes the specific parameters in weather changes (such as temperature, humidity, and rainfall), and finds that rainfall has the most significant impact on sales and loss rate, so rainfall is taken as a pending influencing factor. At the same time, the server will also consider the promotion factor selected in the first stage and analyze the impact of different data configurations of the promotion factor and the rainfall factor on the key indicators. After multiple rounds of iterations, the server finally determines rainfall and promotions as target influencing factors because the correlation strength between them and the key indicators reaches the preset correlation strength threshold. The server quantitatively analyzes the influence dependency between the two key indicators of daily sales and loss rate and the two target influencing factors of rainfall and promotion, and generates fresh supply chain data analysis results. The results may show that in rainy weather, the sales volume of spinach will drop by about 20% and the loss rate will increase by about 15%; during the promotion period, the sales volume will increase by about 30% and the loss rate will decrease by about 10%. These specific influence dependencies provide valuable decision-making basis for managers of the fresh supply chain so that they can adjust the supply chain strategy according to actual conditions.
[0026] In the embodiment of the present invention, obtaining the key indicators and multiple candidate influencing factors of the target fresh food category may be implemented through the following examples.
[0027] Obtaining an influencing factor identification indication for the target fresh food category, and extracting a key indicator of the target fresh food category from the influencing factor identification indication;
[0028] Determine multiple candidate influencing factors of the target fresh food category from the data relationship network where the key indicators are located.
[0029] In an embodiment of the present invention, illustratively, the server first receives an identification instruction of influencing factors for a target fresh food category (such as spinach) from a fresh food supply chain management system or a related data analysis platform. This instruction can be a data analysis task, requiring the server to analyze the influencing factors of spinach sales and loss rate. The server extracts the key indicators of spinach, namely daily sales and loss rate, from this instruction. The server then accesses a data warehouse storing a large amount of fresh food supply chain data, and locates the data relationship network where the two key indicators of spinach daily sales and loss rate are located. In this network, the server finds that the data points associated with the daily sales and loss rate of spinach include weather changes (such as temperature, humidity, whether it rains), promotional activities (such as discounts, buy one get one free), holiday effects, supplier stability, and logistics transportation time. The server determines these data points associated with the key indicators as multiple candidate influencing factors of spinach. Through these two steps, the server successfully obtains the key indicators (daily sales and loss rate) of the target fresh food category (spinach) and multiple candidate influencing factors (weather changes, promotional activities, holiday effects, supplier stability, logistics transportation time, etc.), laying the foundation for subsequent in-depth analysis and feature engineering.
[0030] In the embodiment of the present invention, determining a plurality of candidate influencing factors of the target fresh food category from the data relationship network where the key indicators are located can be implemented through the following examples.
[0031] Determine the fresh product category type of the target fresh product category, and obtain a data relationship network of the fresh product category type; each element of the data relationship network includes a first element and a second element where the key indicator is located; the second element is the remaining element except the first element;
[0032] Determine a plurality of undetermined elements from each of the second elements according to the distribution information corresponding to each element in the data relationship network;
[0033] The influencing parameter corresponding to the undetermined element is determined as the candidate influencing factor of the target fresh food category.
[0034] In an embodiment of the present invention, illustratively, the server first determines the type of fresh category to which the target fresh category (such as spinach) belongs, namely, "vegetables". Then, the server accesses the data warehouse to obtain the data relationship network of the "vegetable" fresh category. This network contains various indicators and influencing factors of multiple vegetable categories, including the two key indicators of daily sales volume and loss rate of spinach. In the data relationship network of the "vegetable" fresh category, the server analyzes the distribution information of each element (the first element including the two key indicators of daily sales volume and loss rate of spinach, and the indicators and influencing factors of other vegetable categories as the second element). The server will find that some second elements (such as weather changes, promotional activities) show a strong correlation with the key indicators of spinach in the network, that is, their distribution in the network is closely related to the key indicators of spinach. Based on this correlation, the server determines these second elements as pending elements. The server extracts the specific influencing parameters (such as temperature, humidity, whether it rains, discounts, buy one get one free, etc.) corresponding to the pending elements determined in the previous step (such as weather changes, promotions), and determines these influencing parameters as alternative influencing factors for the target fresh category of spinach. These alternative influencing factors will be further considered in subsequent analysis to determine their specific impact on the daily sales volume and loss rate of spinach. Through these three steps, the server successfully identified multiple alternative influencing factors (such as specific parameters of weather changes, specific forms of promotions, etc.) for the target fresh category (spinach) from the data relationship network where the key indicators are located, providing strong support for subsequent in-depth analysis and feature engineering.
[0035] In an embodiment of the present invention, the key indicators of the target fresh food category are obtained in the following manner.
[0036] Acquire multiple key indicator instances of multiple fresh food business scenarios, each of the key indicator instances including an indicator feature, a key indicator type target value, and a key indicator scenario target value;
[0037] Extracting indicator semantic features from the indicator features using an initial indicator recognition model, performing semantic recognition on the key indicator instance according to the indicator semantic features to obtain a semantic recognition result, and performing business scenario classification on the key indicator instance according to the indicator semantic features to obtain a business scenario classification result;
[0038] Determine a first error according to the business scenario classification result and the key indicator scenario target value, and determine a second error according to the semantic recognition result and the key indicator type target value;
[0039] According to the first error and the second error, a training process is performed on the initial indicator recognition model to obtain a trained indicator recognition model;
[0040] The target fresh food category is input into the trained indicator recognition model to obtain the key indicators of the target fresh food category.
[0041] In an embodiment of the present invention, illustratively, the server obtains multiple key indicator instances of multiple fresh business scenarios (such as vegetable sales, fruit procurement, meat distribution, etc.) from the fresh supply chain management system, the sales data analysis platform, and the market research report. Each key indicator instance includes indicator features (such as daily sales volume, monthly procurement volume, delivery on-time rate, etc.), key indicator type target values (such as numerical type, proportional type, etc.) and key indicator scenario target values (such as in the vegetable sales scenario, the target value of daily sales volume may be 500 kilograms). The server uses an initial indicator recognition model to extract indicator semantic features from the acquired key indicator instances. For example, for the indicator feature of "daily sales volume", the model can extract the two semantic features of "day" and "sales volume". Then, the model performs semantic recognition on the key indicator instances based on these semantic features to determine which type of indicators they belong to (such as sales volume, procurement volume, etc.). At the same time, the model also classifies the key indicator instances into business scenarios based on the semantic features to determine which fresh business scenarios they belong to (such as vegetable sales, fruit procurement, etc.). According to the business scenario classification results, the server assigns the key indicator instance to the corresponding fresh business scenario, compares it with the target value of the scenario, and calculates the first error. For example, in the vegetable sales scenario, if the model mistakenly classifies a key indicator instance into the fruit procurement scenario, then the difference between the scenario target value of the instance and the target value of the vegetable sales scenario is the first error. At the same time, the server also determines whether the model correctly identifies the type of the key indicator based on the semantic recognition results, and compares it with the key indicator type target value to calculate the second error. The server uses the calculated first error and second error as training signals to train the initial indicator recognition model. Through continuous iterative training, the model gradually learns how to more accurately extract semantic features from indicator features, perform semantic recognition and business scenario classification. Finally, the server obtains a trained indicator recognition model. When it is necessary to determine the key indicators of the target fresh category (such as spinach), the server inputs the relevant information of the category into the trained indicator recognition model. Based on the input information, the model outputs the key indicators of spinach, such as daily sales volume and loss rate. These key indicators will be used in subsequent fresh supply chain data analysis.
[0042] In an embodiment of the present invention, the indicator recognition model includes: a semantic feature extraction unit, a semantic recognition unit and a business scenario classification unit;
[0043] The semantic feature extraction unit is used to receive the indicator feature and extract the indicator semantic feature from the indicator feature;
[0044] The semantic recognition unit is used to receive the indicator semantic feature, and perform semantic recognition on the key indicator instance according to the indicator semantic feature to obtain the semantic recognition result;
[0045] The business scenario classification unit is used to receive the indicator semantic feature, and perform business scenario classification on the key indicator instance according to the indicator semantic feature to obtain the business scenario classification result.
[0046] In an embodiment of the present invention, illustratively, the server obtains a key indicator instance from the fresh supply chain management system, and the indicator feature of the instance is "spinach daily sales volume". The semantic feature extraction unit receives this indicator feature and extracts three semantic features of "spinach", "day" and "sales volume" from it. These semantic features will be passed to the semantic recognition unit and the business scenario classification unit for further processing. The semantic recognition unit receives the three semantic features of "spinach", "day" and "sales volume" from the semantic feature extraction unit. Based on these semantic features, the semantic recognition unit determines that this key indicator instance belongs to the "sales volume" type indicator, and passes this semantic recognition result to the subsequent business scenario classification unit. The business scenario classification unit receives the semantic recognition result from the semantic recognition unit and the original semantic feature from the semantic feature extraction unit. Based on this information, the business scenario classification unit determines that this key indicator instance belongs to the fresh business scenario of "vegetable sales". At the same time, the business scenario classification unit will also obtain the sales volume target value corresponding to the "vegetable sales" scenario from the preset scenario target value based on this classification result, such as 500 kilograms / day, for subsequent error calculation and model training. Through the cooperation of these three units, the server can accurately extract semantic features from key indicator instances of fresh food business scenarios, perform semantic recognition and business scenario classification, and thus provide accurate data support for subsequent error calculation and model training. Ultimately, the trained indicator recognition model will be used to identify key indicators of the target fresh food category.
[0047] In an embodiment of the present invention, the business scenario classification unit is used to receive the indicator semantic features and the semantic recognition results, and perform business scenario classification on the key indicator instance according to the indicator semantic features and the semantic recognition results to obtain the business scenario classification results.
[0048] In an embodiment of the present invention, illustratively, the server has extracted indicator semantic features, such as "spinach", "day" and "sales volume", from the key indicator instance of the fresh business scenario through the semantic feature extraction unit. At the same time, the semantic recognition unit has also determined that this key indicator instance belongs to the "sales volume" type indicator based on these semantic features, and generated a semantic recognition result. Next, the business scenario classification unit receives these indicator semantic features and semantic recognition results. It first checks the semantic recognition result and understands that this is an indicator of the "sales volume" type. Then, it further analyzes the indicator semantic features, especially the feature of "spinach", because it is usually associated with a specific fresh business scenario. Based on this information and preset business scenario classification rules (for example, indicators containing "spinach" are usually associated with the "vegetable sales" scenario), the business scenario classification unit determines that this key indicator instance belongs to the "vegetable sales" fresh business scenario. Finally, the business scenario classification unit generates a business scenario classification result and passes this result to the subsequent process for calculating errors, training models or performing other related operations. Through this process, the server can accurately classify the business scenarios of the key indicator instances according to the indicator semantic features and semantic recognition results, and provide accurate information for subsequent data analysis and decision-making.
[0049] In an embodiment of the present invention, a training process is performed on the initial indicator recognition model according to the first error and the second error, including:
[0050] Performing a first optimization operation on the network parameters of the business scenario classification unit; the first optimization operation is used to optimize the first error to a minimum value;
[0051] Performing a second optimization operation on the network parameters of the semantic recognition unit; the second optimization operation is used to optimize the second error to a minimum value;
[0052] A third optimization operation is performed on the network parameters of the semantic feature extraction unit; the third optimization operation is used to optimize the first error to a maximum value and optimize the second error to a minimum value.
[0053] In an embodiment of the present invention, exemplarily, the server has calculated the first error and the second error, which represent the difference between the predicted results and the actual results of the business scenario classification unit and the semantic recognition unit, respectively. In order to improve the performance of the model, the server will train different units in the initial indicator recognition model. First, the server performs a first optimization operation on the network parameters of the business scenario classification unit. The goal of this operation is to adjust the network parameters so that the first error (i.e., the business scenario classification error) is optimized to a minimum value. Through continuous iteration and adjustment, the server gradually finds a combination of network parameters that can minimize the first error. Next, the server performs a second optimization operation on the network parameters of the semantic recognition unit. The goal of this operation is to adjust the network parameters so that the second error (i.e., the semantic recognition error) is optimized to a minimum value. Similarly, through continuous iteration and adjustment, the server finds a combination of network parameters that can minimize the second error. Finally, the server performs a third optimization operation on the network parameters of the semantic feature extraction unit. The goal of this operation is relatively special. It is to maximize the first error (this can be a reversal in expression, which should actually reduce the contribution of the first error to the overall error, or seek the optimal solution under certain constraints), and to optimize the second error to a minimum value. This may require a complex optimization strategy, such as a multi-objective optimization algorithm. In this way, the server finds a combination of network parameters that can meet these two conditions. After completing these optimization operations, the server obtains a trained indicator recognition model, which should show better performance in business scenario classification, semantic recognition, and semantic feature extraction.
[0054] In an embodiment of the present invention, the business scenario classification unit includes an information conversion component;
[0055] The information conversion component is used to receive initial data during the forward transmission process and transmit the initial data to the business scenario classification unit; wherein the initial data is calculated based on the indicator semantic features generated by the semantic feature extraction unit and the semantic recognition result generated by the semantic recognition unit;
[0056] The information conversion component is also used to reversely process the gradient direction of the business scenario classification unit and transmit it to the semantic feature extraction unit during the reverse transmission process.
[0057] In an embodiment of the present invention, illustratively, the server has extracted indicator semantic features, such as "spinach", "day" and "sales volume", from the key indicator instance of the fresh business scenario through the semantic feature extraction unit. The semantic recognition unit has also determined that this key indicator instance belongs to the "sales volume" type indicator based on these semantic features, and generated a semantic recognition result. The information conversion component receives the indicator semantic features generated by the semantic feature extraction unit and the semantic recognition result generated by the semantic recognition unit. The information conversion component performs necessary calculations or processing (if any) on these initial data, and then transmits the processed data to the business scenario classification unit. The business scenario classification unit classifies the key indicator instance into business scenarios based on the received data to obtain a business scenario classification result. In the model training phase, the server calculates the gradient direction of the business scenario classification unit, which represents the sensitivity of the model to the business scenario classification error under the current parameters. The information conversion component receives this gradient direction and performs inverse processing. The inverse processing may include operations such as negation and scaling, depending on the training algorithm and model structure. The gradient direction after inverse processing is transmitted to the semantic feature extraction unit. The semantic feature extraction unit updates its network parameters according to the received gradient direction to reduce the classification error of the business scenario. Through the role of the information conversion component in the forward and reverse transmission processes, the server can ensure that the data flow and gradient transmission during the model training process are correct and effective, thereby training a better performance indicator recognition model.
[0058] In an embodiment of the present invention, the business scenario classification unit further includes: at least one integrated learning component;
[0059] The at least one integrated learning component is used to receive the initial data from the information conversion component, perform business scenario classification on the key indicator instance according to the initial data, and obtain and output the business scenario classification result.
[0060] In an embodiment of the present invention, illustratively, the server extracts a key indicator instance, such as "daily sales volume of spinach", from the fresh supply chain management system. The semantic feature extraction unit extracts indicator semantic features, such as "spinach", "day" and "sales volume", from the indicator instance. The semantic recognition unit determines that the indicator instance belongs to the "sales volume" type indicator based on these semantic features, and generates a semantic recognition result. The information conversion component receives the indicator semantic features and semantic recognition results generated by the semantic feature extraction unit and the semantic recognition unit. The information conversion component performs necessary calculations or processing on these initial data and prepares to transmit them to the integrated learning component. At least one integrated learning component receives initial data from the information conversion component. The integrated learning component contains multiple base learners, such as decision trees, support vector machines, etc. Each base learner performs independent business scenario classification judgment on the key indicator instance based on the received initial data. The integrated learning component integrates the classification results of all base learners through strategies such as voting or weighted average. The integrated learning component obtains the final business scenario classification result, such as "vegetable sales". The result is output and can be used for subsequent error calculation, model training or other related operations. In this way, the server can use the powerful ability of ensemble learning to accurately classify business scenarios for key indicators in the fresh food supply chain. This not only helps to better understand business performance in different scenarios, but also provides targeted suggestions and support for supply chain optimization. For example, in the "vegetable sales" scenario, the server can further analyze which vegetable varieties have higher sales volumes and which periods of time have more concentrated sales, thereby providing data support for inventory management, promotion strategies, etc.
[0061] In an embodiment of the present invention, the semantic feature extraction unit includes: a preprocessing component, a word embedding component, a perceptron component and an integrated learning component;
[0062] The preprocessing component is used to receive the indicator feature;
[0063] The word embedding component is used to map the indicator feature into a semantic vector;
[0064] The perceptron component is used to determine the weight coefficient corresponding to the semantic vector, and perform a linear superposition operation on the semantic vector according to the weight coefficient to obtain a processed semantic vector;
[0065] The integrated learning component is used to generate the indicator semantic feature according to the processed semantic vector.
[0066] In an embodiment of the present invention, illustratively, the fresh supply chain management system needs to extract semantic features for key indicators (such as "daily sales of spinach") for subsequent business scenario classification and data analysis. The server receives the original key indicator features, such as "daily sales of spinach", from the fresh supply chain management system. The preprocessing component cleans and normalizes these raw data, such as removing irrelevant characters, unifying the time format, etc., and prepares to pass them to the word embedding component. The preprocessed key indicator features are passed to the word embedding component. The word embedding component uses a pre-trained word embedding model (such as Word2Vec, BERT, etc.) to map words such as "spinach", "day" and "sales volume" to corresponding semantic vectors. The perceptron component receives the semantic vector generated by the word embedding component. It determines the weight coefficients corresponding to each semantic vector through training, and these weight coefficients reflect the importance of different words in the indicator. The perceptron component performs a linear superposition operation on the semantic vector according to these weight coefficients to obtain a processed semantic vector, which integrates the semantic information of the original words and their importance. The integrated learning component receives the processed semantic vector. It can contain multiple base learners (such as decision trees, neural networks, etc.), which jointly analyze and learn semantic vectors. The ensemble learning component generates the final indicator semantic features, such as "vegetable sales daily index", by synthesizing the outputs of all base learners. In this way, the server can accurately extract rich semantic features from the original key indicators, which not only contain the semantic information of the original vocabulary, but also incorporate their importance in the indicator. These semantic features provide strong support for subsequent business scenario classification and data analysis. For example, in the optimization of the fresh food supply chain, the server can use these semantic features to identify sales patterns, inventory requirements, etc. in different scenarios, thereby providing decision makers with accurate insights and suggestions.
[0067] In an embodiment of the present invention, the semantic recognition unit includes: at least one integrated learning component;
[0068] The at least one integrated learning component is used to receive the indicator semantic features from the semantic feature extraction unit, and perform semantic recognition on the key indicator instance according to the indicator semantic features to obtain and output the semantic recognition result.
[0069] In an embodiment of the present invention, illustratively, a fresh supply chain management system needs to perform semantic recognition on a key indicator instance (e.g., "daily sales volume of spinach") to determine the semantic category to which it belongs (e.g., "sales volume" type indicator) so as to perform subsequent business scenario classification and data analysis. The semantic recognition unit receives the extracted indicator semantic features from the semantic feature extraction unit, which integrate the semantic information of the original vocabulary and their importance in the indicator. At least one integrated learning component receives these indicator semantic features as input. The integrated learning component contains multiple base learners, such as decision trees, support vector machines, etc., which jointly analyze and learn the input semantic features. Each base learner independently judges the semantic category of the key indicator instance based on the received semantic features. The integrated learning component synthesizes the output results of all base learners through strategies such as voting or weighted average. The integrated learning component obtains the final semantic recognition result, such as the "sales volume" type indicator. The result is output and can be used for subsequent business scenario classification, data analysis or other related operations. In this way, the server can accurately identify the semantic category to which the key indicator instance belongs. In the optimization of the fresh food supply chain, this helps to conduct targeted analysis and processing of different types of indicators. For example, for indicators of the "sales volume" type, the server can further analyze sales trends, the impact of promotional activities on sales, etc.; while for indicators of the "inventory volume" type, it can focus on inventory turnover, replenishment strategies, etc. This refined analysis provides strong data support for the optimization of the fresh food supply chain.
[0070] In an embodiment of the present invention, a supply chain data sample data set is obtained according to the relevant parameters corresponding to each of the candidate influencing factors and the key indicators, which can be implemented through the following examples.
[0071] Performing priori processing on relevant parameters corresponding to each of the candidate influencing factors and the key indicators to obtain priori processed data;
[0072] Performing feature recognition operations on the priori processed data to obtain influencing parameter features required for feature engineering;
[0073] A supply chain data sample data set containing the influencing parameter features is obtained.
[0074] In an embodiment of the present invention, illustratively, the fresh supply chain management system needs to analyze the impact of multiple alternative influencing factors (such as weather, promotional activities, holidays, etc.) on key indicators (such as "spinach daily sales"). In order to perform this analysis, the server needs to first obtain a supply chain data sample data set containing the parameters corresponding to these influencing factors and key indicators. The server extracts the relevant parameters (such as weather conditions, promotional efforts, holiday types, etc.) corresponding to each alternative influencing factor (such as weather data, promotional activity records, holiday arrangements, etc.) and the key indicator ("spinach daily sales") from the fresh supply chain management system. These raw data are processed a priori, including data cleaning (removing invalid or outliers), data normalization (unifying data formats and dimensions), etc., to obtain a priori processed data. Feature recognition operations are performed on the data after a priori processing, which include feature selection (selecting parameters that have an impact on the analysis), feature construction (such as calculating derived indicators), etc. Through the feature recognition operation, the server obtains the influencing parameter features required to achieve feature engineering, and these features can reflect the impact of alternative influencing factors on key indicators. The server integrates the data containing the influencing parameter features into a supply chain data sample data set, which contains all the necessary information for subsequent analysis. The supply chain data sample data set is stored in the server's database, ready for subsequent data analysis, model training and other operations. In this way, the server can prepare a supply chain data sample data set containing rich features, providing strong data support for subsequent in-depth analysis. In the optimization of the fresh food supply chain, this helps to more accurately understand the degree of influence of different influencing factors on key indicators, so as to develop more effective supply chain optimization strategies. For example, by analyzing the impact of weather on sales, the server can recommend adjusting inventory or promotion strategies under different weather conditions to improve sales performance.
[0075] In an embodiment of the present invention, iterative feature engineering is performed based on the supply chain data sample data set to determine the target influencing factors obtained by screening each of the candidate influencing factors stage by stage, which can be implemented through the following examples.
[0076] Performing feature engineering on the supply chain data sample dataset to obtain a prediction model;
[0077] Predicting the influence of the key indicator according to the prediction model to obtain the correlation strength between each candidate influencing factor and the key indicator;
[0078] From each of the candidate influencing factors, select the pending influencing factor whose correlation strength reaches a preset correlation strength threshold;
[0079] The data configuration status of the supply chain data sample is optimized according to the pending influencing factors, an optimized supply chain data sample data set is determined, and the steps of performing feature engineering on the supply chain data sample data set are iterated in a loop to obtain target influencing factors; the optimized supply chain data sample data set includes relevant parameters corresponding to the pending influencing factors and the key indicators.
[0080] In an embodiment of the present invention, exemplarily, the server performs feature engineering on the supply chain data sample data set, including steps such as feature selection, feature construction and feature conversion, to obtain a prediction model for predicting key indicators. This prediction model can be an algorithm based on machine learning, such as linear regression, decision tree or neural network. The server uses the prediction model to predict the influence of the key indicators, that is, to predict the changes of the key indicators under different candidate influencing factors. Through the prediction results, the server calculates the correlation strength between each candidate influencing factor and the key indicator, and this correlation strength reflects the influence of the influencing factor on the key indicator. The server selects the pending influencing factors whose correlation strength reaches the preset correlation strength threshold from each candidate influencing factor. For example, if the preset correlation strength threshold is 0.5, then the influencing factor with a correlation strength greater than or equal to 0.5 will be selected as the pending influencing factor. The server optimizes the data configuration status of the supply chain data sample according to the pending influencing factors, that is, adjusts the parameters and features contained in the data set to determine the optimized supply chain data sample data set. The server uses the optimized supply chain data sample data set to iterate the steps of feature engineering, that is, repeat the operations such as feature selection, feature construction and feature conversion to obtain a more accurate prediction model and a higher correlation strength. Through multiple iterations, the server gradually screens out the target influencing factors that have the greatest impact on key indicators. In this way, the server can accurately determine the target influencing factors that have the greatest impact on key indicators. In the optimization of the fresh food supply chain, this helps supply chain managers focus on the most important influencing factors and formulate corresponding strategies to improve the efficiency and effectiveness of the supply chain. For example, if the analysis results show that weather is the factor that has the greatest impact on sales, then supply chain managers can formulate sales strategies and inventory plans for different weather conditions to improve sales performance and customer satisfaction.
[0081] In an embodiment of the present invention, feature engineering is performed on the supply chain data sample dataset to obtain a prediction model, which can be implemented through the following examples.
[0082] Determine the characteristic domain corresponding to each of the relevant parameters in the supply chain data sample data set;
[0083] The deep learning components associated with each of the feature domains are used to perform feature engineering on the supply chain data sample dataset to obtain a prediction model.
[0084] In an embodiment of the present invention, illustratively, a fresh food supply chain management system has accumulated a large amount of historical data, including weather conditions, promotion activity records, holiday arrangements, and key indicators such as "spinach daily sales volume". In order to predict the future trend of "spinach daily sales volume", the server will perform feature engineering on these data and build a prediction model based on deep learning. The server first analyzes the supply chain data sample data set to determine the feature domain corresponding to each relevant parameter. For example, for weather condition parameters, its feature domain may include temperature, humidity, wind speed, etc.; for promotion activity record parameters, its feature domain may include promotion intensity, promotion duration, etc. The server selects the deep learning component associated with it according to the characteristics of the feature domain. For example, for a sequential feature domain (such as time series weather data), the server may select an LSTM (long short-term memory network) component; for an image feature domain (such as a poster image of a promotion activity), the server may select a CNN (convolutional neural network) component. The server uses the selected deep learning component to perform feature engineering on the supply chain data sample data set. This includes steps such as data preprocessing (such as normalization, standardization), feature extraction (such as automatic extraction of advanced features through deep learning components), and feature selection (such as feature selection based on importance weights). After feature engineering, the server uses a deep learning algorithm to train the processed data to obtain a prediction model. This prediction model can receive new input data and output a prediction result for the future "daily spinach sales volume". In this way, the server can fully utilize the powerful capabilities of deep learning technology to perform in-depth feature engineering on fresh food supply chain data and build an accurate prediction model. This helps fresh food supply chain managers better understand future sales trends and develop more reasonable inventory plans and sales strategies, thereby improving the responsiveness and overall benefits of the supply chain. For example, through the prediction model, managers can learn in advance about sales fluctuations caused by weather changes in a certain period of time in the future, and adjust inventory and promotion strategies accordingly to cope with potential sales peaks or troughs.
[0085] In an embodiment of the present invention, feature engineering is performed on the supply chain data sample dataset to obtain a prediction model, which can be implemented through the following examples.
[0086] Dividing the supply chain data sample data set into a plurality of sub-supply chain data sample data sets; each of the sub-supply chain data sample data sets includes relevant parameters of the key indicator and relevant parameters of a preset number of the candidate influencing factors;
[0087] Feature engineering is performed on each of the sub-supply chain data sample data sets to obtain a prediction model corresponding to each of the sub-supply chain data sample data sets.
[0088] In an embodiment of the present invention, illustratively, a fresh supply chain management system has accumulated a large amount of historical data, including a variety of alternative influencing factors (such as weather, promotional activities, holidays, etc.) and a key indicator "daily sales of spinach". In order to more finely analyze the impact of different influencing factors on key indicators and build a more accurate prediction model, the server decides to divide the supply chain data sample data set into multiple subsets, and perform feature engineering and model training on each subset respectively. The server divides the supply chain data sample data set into multiple sub-supply chain data sample data sets. Each subset includes relevant parameters of the key indicator "daily sales of spinach" and relevant parameters of a preset number of alternative influencing factors. For example, subset 1 may include relevant parameters of weather and promotional activities, and subset 2 may include relevant parameters of holidays and price changes. The server performs feature engineering on each sub-supply chain data sample data set respectively. This includes data preprocessing (such as filling missing values, removing outliers), feature extraction (such as extracting advanced features through deep learning components), feature selection (such as feature selection based on correlation analysis), and other steps. After feature engineering processing, the server uses a deep learning algorithm to train the data of each subset to obtain a prediction model corresponding to each subset. For example, the prediction model corresponding to subset 1 can be better at predicting the impact of weather and promotions on sales, while the prediction model corresponding to subset 2 can be better at predicting the impact of holidays and price changes on sales. In this way, the server can more finely analyze the impact of different influencing factors on key indicators and build a more accurate prediction model. This helps fresh supply chain managers to have a deeper understanding of how different factors work together on sales trends and formulate more refined management strategies accordingly. For example, through the prediction model of subset 1, managers can accurately predict sales changes under specific weather and promotions in a certain period of time in the future; and through the prediction model of subset 2, managers can predict the specific impact of holidays and price changes on sales. These prediction results will provide strong data support for inventory management, sales strategy formulation, etc. of the supply chain.
[0089] In an embodiment of the present invention, the influence of the key indicator is predicted according to the prediction model to obtain the correlation strength between each candidate influencing factor and the key indicator, which can be implemented through the following examples.
[0090] Predicting the influence of the key indicator according to the prediction model to obtain the initial correlation strength between each candidate influencing factor and the key indicator;
[0091] For each of the candidate influencing factors, a data perturbation operation is performed on a relevant parameter of the candidate influencing factor to obtain perturbation data;
[0092] Determining the prediction stability of the prediction model by the candidate influencing factors according to the relevant parameters of the key indicator and the error between the prediction result of the influence of the prediction model on the disturbance data;
[0093] The association strength between the candidate influencing factors and the key indicator is determined in combination with the initial association strength and the predicted stability.
[0094] In an embodiment of the present invention, illustratively, a fresh supply chain management system has constructed multiple prediction models, corresponding to different sub-supply chain data sample data sets. Now, the server needs to use these prediction models to evaluate the correlation strength of each alternative influencing factor (such as weather, promotional activities, holidays, etc.) on the key indicator "spinach daily sales volume". The server uses the trained prediction model to predict the influence of the key indicator "spinach daily sales volume". Through the prediction results, the server can obtain the initial correlation strength of each alternative influencing factor with the key indicator. For example, the server finds that the initial correlation strength of weather and promotional activities with sales is high. For each alternative influencing factor, the server performs a data perturbation operation on its related parameters to generate perturbation data. For example, for the weather influencing factor, the server can slightly modify some weather data (such as changing sunny days to cloudy days) to generate a new perturbation data set. The server uses the prediction model to predict the influence of the perturbation data and obtain the prediction result. By comparing the error between the prediction result of the perturbation data and the prediction result of the original data, the server can evaluate the prediction stability of the alternative influencing factor on the prediction model. For example, if a small disturbance in weather data causes a large change in sales forecast results, it means that the weather factor has a greater impact on the stability of the forecast model. The server combines the initial correlation strength and forecast stability to determine the final correlation strength of each candidate influencing factor with the key indicator. For example, although the initial correlation strength between weather and sales is high, if its forecast stability is poor (that is, data disturbances cause large fluctuations in forecast results), the server will reduce the final correlation strength between weather and sales. In this way, the server can more accurately evaluate the actual impact of each candidate influencing factor on key indicators. This helps fresh supply chain managers formulate management strategies more scientifically. For example, if the server finds that the correlation strength of promotional activities on sales is high and the forecast stability is good, managers can formulate promotional activities more frequently to increase sales.
[0095] In an embodiment of the present invention, the influence of the key indicator is predicted according to the prediction model to obtain the initial correlation strength between each candidate influencing factor and the key indicator, which can be implemented through the following examples.
[0096] The relevant parameters corresponding to each of the candidate influencing factors are used as the input of the prediction model, and the relevant parameters of the key indicators are used as the output of the prediction model, and the input parameter effects of the prediction model are statistically analyzed through gradient backtracking in a reverse transfer process; the input parameter effects are represented by parameter effect values corresponding to each of the candidate influencing factors;
[0097] According to each of the parameter effect values, the initial correlation strength between each of the candidate influencing factors and the key indicator is obtained; the initial correlation strength is associated with the parameter effect value of the candidate influencing factor.
[0098] In an embodiment of the present invention, illustratively, a fresh supply chain management system has constructed multiple prediction models, corresponding to different sub-supply chain data sample data sets. Now, the server needs to use these prediction models to evaluate the initial correlation strength of each alternative influencing factor (such as weather, promotional activities, holidays, etc.) to the key indicator "spinach daily sales volume". The server uses the relevant parameters corresponding to each alternative influencing factor as the input of the prediction model and the relevant parameters of the key indicator as the output of the prediction model. Through the reverse transfer process (i.e., the back propagation algorithm in deep learning), the server gradient backtracks the input parameter effect of the prediction model. This input parameter effect is characterized by the parameter effect value corresponding to each alternative influencing factor. For example, for the influencing factor of weather, the server can count the effect values of weather parameters such as temperature and humidity. The server obtains the initial correlation strength of each alternative influencing factor and the key indicator based on each parameter effect value. This initial correlation strength is associated with the parameter effect value of the alternative influencing factor, that is, the larger the parameter effect value, the greater the influence of the influencing factor on the key indicator, and the higher the initial correlation strength. For example, if the server finds that the parameter effect value of temperature is large, it means that the initial correlation strength of temperature to the daily sales volume of spinach is high. In this way, the server can quantitatively evaluate the initial impact of each candidate influencing factor on key indicators. This helps fresh food supply chain managers formulate management strategies more scientifically. For example, if the server finds that the initial correlation strength of temperature on sales is high, managers can pay more attention to weather changes when formulating sales strategies in order to better seize sales opportunities.
[0099] In the embodiment of the present invention, the to-be-determined influencing factor whose association strength reaches a preset association strength threshold is selected from each of the candidate influencing factors, which can be implemented through the following example.
[0100] Determining a preset association strength threshold associated with the current iteration cycle;
[0101] The correlation strength corresponding to each of the candidate influencing factors is compared with the preset correlation strength threshold, and the pending influencing factors whose correlation strength reaches the preset correlation strength threshold are selected from the candidate influencing factors.
[0102] In an embodiment of the present invention, illustratively, the fresh supply chain management system has completed the analysis of the correlation strength between each candidate influencing factor and the key indicator "daily sales volume of spinach". Now, the server needs to select the pending influencing factors from the candidate influencing factors according to the preset correlation strength threshold, so as to further analyze or formulate management strategies. The server first determines the preset correlation strength threshold associated with this round of iteration cycle. This threshold can be set according to historical experience, business needs or expert advice. For example, the server can set 0.6 as the preset correlation strength threshold for this round of iteration cycle, which means that only the influencing factors with a correlation strength of or greater than 0.6 will be selected as pending influencing factors. The server compares the correlation strength corresponding to each candidate influencing factor with the preset correlation strength threshold. For example, the server can find that weather (correlation strength 0.7), promotional activities (correlation strength 0.8) and price changes (correlation strength 0.5) are three candidate influencing factors related to daily sales volume of spinach. From each candidate influencing factor, the server selects the pending influencing factor whose correlation strength reaches the preset correlation strength threshold. In this example, the correlation strengths of weather (correlation strength 0.7) and promotions (correlation strength 0.8) both reached the threshold of 0.6, so they were selected as pending influencing factors. Price changes (correlation strength 0.5) were excluded because their correlation strength did not reach the threshold. In this way, the server can filter out pending influencing factors that have a significant impact on key indicators from multiple candidate influencing factors. This helps fresh food supply chain managers focus more on those truly important influencing factors and formulate more accurate management strategies based on them. For example, in this case, managers will pay more attention to the impact of weather and promotions on spinach sales, and formulate corresponding sales strategies to increase sales.
[0103] The embodiment of the present invention provides a computer device 100, which includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device 100 executes the aforementioned fresh food supply chain data analysis method based on big data and deep learning. Figure 2 As shown, Figure 2 The block diagram of the computer device 100 provided in the embodiment of the present invention. The computer device 100 includes a memory 111, a processor 112 and a communication unit 113. To achieve data transmission or interaction, the memory 111, the processor 112 and the communication unit 113 are directly or indirectly electrically connected to each other. For example, these components can be electrically connected to each other through one or more communication buses or signal lines.
[0104] For illustrative purposes, the foregoing description is made with reference to specific embodiments. However, the above illustrative discussion is not intended to be exhaustive or to limit the present disclosure to the precise form disclosed. Numerous modifications and variations are possible in accordance with the above teachings. These embodiments are selected and described in order to best illustrate the principles of the present disclosure and its practical application, so that those skilled in the art can best utilize the present disclosure and utilize various embodiments with different modifications to suit the intended specific application.
Claims
1. A fresh food supply chain data analysis method based on big data and deep learning, characterized in that: include: Acquire multiple fresh food categories to be monitored, and acquire basic indicator data of each of the fresh food categories to be monitored; When there is an abnormality in the fresh food supply chain status represented by the basic indicator data, the corresponding fresh food category to be monitored is used as the target fresh food category; The key indicators and multiple candidate influencing factors of the target fresh food category are obtained, and the influence dependency relationship between the key indicators and the target influencing factors determined from the multiple candidate influencing factors is used as the fresh food supply chain data analysis result of the target fresh food category.
2. The method according to claim 1, characterized in that The step of obtaining the key indicator and multiple candidate influencing factors of the target fresh food category, and using the influence dependency relationship between the key indicator and the target influencing factor determined from the multiple candidate influencing factors as the fresh food supply chain data analysis result of the target fresh food category, includes: Obtaining an influencing factor identification indication for the target fresh food category, and extracting key indicators of the target fresh food category from the influencing factor identification indication; Determine the fresh product category type of the target fresh product category, and obtain a data relationship network of the fresh product category type; each element of the data relationship network includes a first element and a second element where the key indicator is located; the second element is the remaining element except the first element; Determining a plurality of undetermined elements from each of the second elements according to the distribution information corresponding to each element in the data relationship network; Determine the influencing parameter corresponding to the undetermined element as a candidate influencing factor of the target fresh food category; Performing priori processing on relevant parameters corresponding to each of the candidate influencing factors and the key indicators to obtain priori processed data; Performing feature recognition operations on the priori processed data to obtain influencing parameter features required for feature engineering; Obtaining a supply chain data sample data set containing the influencing parameter characteristics; Performing feature engineering on the supply chain data sample dataset to obtain a prediction model; Predicting the influence of the key indicator according to the prediction model to obtain the initial correlation strength between each candidate influencing factor and the key indicator; For each of the candidate influencing factors, a data perturbation operation is performed on a relevant parameter of the candidate influencing factor to obtain perturbation data; Determining the prediction stability of the prediction model by the candidate influencing factors according to the relevant parameters of the key indicator and the error between the prediction result of the influence of the prediction model on the disturbance data; Determine the correlation strength between the candidate influencing factor and the key indicator in combination with the initial correlation strength and the prediction stability; From each of the candidate influencing factors, select the pending influencing factor whose correlation strength reaches a preset correlation strength threshold; According to the pending influencing factors, the data configuration status of the supply chain data samples is optimized, the optimized supply chain data sample data set is determined, and the step of performing feature engineering on the supply chain data sample data set is iterated cyclically to obtain the target influencing factors; the optimized supply chain data sample data set includes the pending influencing factors and the relevant parameters corresponding to the key indicators; wherein the supply chain data samples of the next stage of feature engineering include the relevant parameters of the key indicators and the relevant parameters of the pending influencing factors selected by the feature engineering of the previous stage, and the relevant parameters of the pending influencing factors selected by the feature engineering of the previous stage are different from the data configuration status of the relevant parameters of the influencing factors in the supply chain data samples of the feature engineering of the previous stage; the correlation strength between the pending influencing factors and the key indicators reaches a preset correlation strength threshold; The influence dependency relationship between the key indicator and the target influencing factor is used as the fresh food supply chain data analysis result of the target fresh food category.
3. The method according to claim 2, characterized in that The key indicators of the target fresh food category are obtained through the following methods, including: Acquire multiple key indicator instances of multiple fresh food business scenarios, each of the key indicator instances including an indicator feature, a key indicator type target value, and a key indicator scenario target value; Extracting indicator semantic features from the indicator features using an initial indicator recognition model, performing semantic recognition on the key indicator instance according to the indicator semantic features to obtain a semantic recognition result, and performing business scenario classification on the key indicator instance according to the indicator semantic features to obtain a business scenario classification result; Determine a first error according to the business scenario classification result and the key indicator scenario target value, and determine a second error according to the semantic recognition result and the key indicator type target value; According to the first error and the second error, a training process is performed on the initial indicator recognition model to obtain a trained indicator recognition model; The target fresh food category is input into the trained indicator recognition model to obtain the key indicators of the target fresh food category.
4. The method according to claim 3, characterized in that The indicator recognition model includes: a semantic feature extraction unit, a semantic recognition unit and a business scenario classification unit; The semantic feature extraction unit is used to receive the indicator feature and extract the indicator semantic feature from the indicator feature; The semantic recognition unit is used to receive the indicator semantic feature, and perform semantic recognition on the key indicator instance according to the indicator semantic feature to obtain the semantic recognition result; The business scenario classification unit is used to receive the indicator semantic feature and the semantic recognition result, and perform business scenario classification on the key indicator instance according to the indicator semantic feature and the semantic recognition result to obtain the business scenario classification result; The business scenario classification unit includes an information conversion component; The information conversion component is used to receive initial data during the forward transmission process and transmit the initial data to the business scenario classification unit; wherein the initial data is calculated based on the indicator semantic features generated by the semantic feature extraction unit and the semantic recognition result generated by the semantic recognition unit; The information conversion component is also used to reversely process the gradient direction of the business scenario classification unit and transmit it to the semantic feature extraction unit during the reverse transmission process; The business scenario classification unit further includes: at least one integrated learning component; The at least one integrated learning component is used to receive the initial data from the information conversion component, perform business scenario classification on the key indicator instance according to the initial data, and obtain and output the business scenario classification result; The semantic feature extraction unit includes: a preprocessing component, a word embedding component, a perceptron component and an integrated learning component; The preprocessing component is used to receive the indicator feature; The word embedding component is used to map the indicator feature into a semantic vector; The perceptron component is used to determine the weight coefficient corresponding to the semantic vector, and perform a linear superposition operation on the semantic vector according to the weight coefficient to obtain a processed semantic vector; The integrated learning component is used to generate the indicator semantic feature according to the processed semantic vector; The semantic recognition unit includes: at least one integrated learning component; The at least one integrated learning component is used to receive the indicator semantic features from the semantic feature extraction unit, and perform semantic recognition on the key indicator instance according to the indicator semantic features to obtain and output the semantic recognition result.
5. The method according to claim 4, characterized in that According to the first error and the second error, a training process is performed on the initial indicator recognition model, including: Performing a first optimization operation on the network parameters of the business scenario classification unit; the first optimization operation is used to optimize the first error to a minimum value; Performing a second optimization operation on the network parameters of the semantic recognition unit; the second optimization operation is used to optimize the second error to a minimum value; A third optimization operation is performed on the network parameters of the semantic feature extraction unit; the third optimization operation is used to optimize the first error to a maximum value and optimize the second error to a minimum value.
6. The method according to claim 2, characterized in that Perform feature engineering on the supply chain data sample dataset to obtain a prediction model, including: Determine the characteristic domain corresponding to each of the relevant parameters in the supply chain data sample data set; The deep learning components associated with each of the feature domains are used to perform feature engineering on the supply chain data sample dataset to obtain a prediction model.
7. The method according to claim 2, characterized in that Perform feature engineering on the supply chain data sample dataset to obtain a prediction model, including: Dividing the supply chain data sample data set into a plurality of sub-supply chain data sample data sets; each of the sub-supply chain data sample data sets includes relevant parameters of the key indicator and relevant parameters of a preset number of the candidate influencing factors; Feature engineering is performed on each of the sub-supply chain data sample data sets to obtain a prediction model corresponding to each of the sub-supply chain data sample data sets.
8. The method according to claim 2, characterized in that: The influence of the key indicator is predicted according to the prediction model to obtain the initial correlation strength between each candidate influencing factor and the key indicator, including: The relevant parameters corresponding to each of the candidate influencing factors are used as the input of the prediction model, and the relevant parameters of the key indicators are used as the output of the prediction model, and the input parameter effects of the prediction model are statistically analyzed through gradient backtracking in a reverse transfer process; the input parameter effects are represented by parameter effect values corresponding to each of the candidate influencing factors; According to each of the parameter effect values, the initial correlation strength between each of the candidate influencing factors and the key indicator is obtained; the initial correlation strength is associated with the parameter effect value of the candidate influencing factor.
9. The method according to claim 2, characterized in that: From the candidate influencing factors, the influencing factors whose association strength reaches a preset association strength threshold are selected, including: Determining a preset association strength threshold associated with the current iteration cycle; The correlation strength corresponding to each of the candidate influencing factors is compared with the preset correlation strength threshold, and the pending influencing factors whose correlation strength reaches the preset correlation strength threshold are selected from the candidate influencing factors.
10. A server system, characterized in that: The method comprises a server, wherein the server is used to execute the method described in any one of claims 1 to 9.
Citation Information
Cited By
Intelligent freshness locking system and method of food refrigeration equipment
CN119625721A
Fresh food supply chain full-process digital management system
CN122288554A
