Agricultural product sales prediction and inventory management method and system based on big data analysis

Abstract

This invention relates to the field of warehousing management technology for pre-packaged rice, flour, and oil products, and particularly to a method and system for agricultural product sales forecasting and inventory management based on big data analysis. The method includes: a data collection step, where multimodal time-series quality characteristic data of agricultural products are collected in real time using a composite sensing array deployed in the storage unit; an analysis step, where the data is input into a physical information fusion neural network to output the theoretical aging cycle and dynamic quality decay curve of individual batches; a depreciation measurement step, where the book inventory is dynamically converted into standardized effective inventory equivalent based on the decay curve; a procurement decision step, where a procurement plan is generated; and an inventory decision step, where outbound instructions are generated based on the effective inventory equivalent and remaining aging cycle, and inventory parameters are adaptively adjusted. This invention solves the technical problems of discrepancies between book inventory and actual sellable inventory, distorted replenishment decisions, and high loss rates caused by factors such as aging, moisture absorption, and mold growth in agricultural products. It achieves synergistic optimization of procurement and inventory, significantly reducing losses and capital occupation.

Description

Technical Field

[0001] This invention relates to the field of warehousing management technology for pre-packaged agricultural products such as rice, flour, and oil, and more specifically, to a method and system for agricultural product sales forecasting and inventory management based on big data analysis. Background Technology

[0002] Existing agricultural product supply chain management systems (such as ERP and WMS) typically treat procurement, inventory, and sales as independent processes, lacking cross-process information sharing and collaborative decision-making. Procurement departments often rely solely on historical sales data or manual experience to formulate procurement plans, failing to anticipate the quality degradation process of agricultural products after warehousing due to physiological activities such as aging, moisture absorption, mold, and oxidative rancidity. They also fail to incorporate the true effective value of current inventory (rather than book quantity) into procurement demand calculations. Inventory management largely depends on static shelf-life or simple warehousing time sequences, ignoring the actual biochemical state of individual agricultural products, leading to significant discrepancies between book inventory and available-for-sale inventory, and distorted replenishment timing. Sales forecasting is also conducted independently, without linking to dynamic inventory quality data, making accurate inventory depletion planning difficult.

[0003] Pre-packaged grain and oil products (such as bagged rice, flour, and cooking oil) are essential necessities, and their inventory management directly impacts food security, capital occupation, and loss control. Unlike fresh agricultural products, the main quality changes in these products manifest as aging, moisture absorption, mold growth, and oxidative rancidity. Their quality degradation is slow, non-linear, and significantly affected by environmental temperature and humidity. Existing warehouse management systems generally employ a static shelf-life management model, managing batches based on the time of entry and preset shelf-life, without considering the dynamic impact of storage environment, stacking methods, and packaging integrity on product quality. Procurement decisions rely heavily on historical sales and human experience, failing to incorporate the true quality of current inventory and future aging trends. This leads to inaccurate replenishment timing, a coexistence of inventory backlog and shortages, high loss rates, and significant capital occupation. Furthermore, while packaging materials may block some environmental factors, temperature and humidity can still be slowly transmitted through the packaging, affecting the quality of the internal products. Therefore, an intelligent management solution capable of real-time sensing and prediction of quality dynamics is urgently needed. Summary of the Invention

[0004] The technical problem this invention aims to solve is how to break down the independence between procurement, inventory, and sales processes, achieve real-time perception and prediction of dynamic changes in agricultural product quality, and integrate this into procurement decisions and inventory management, thereby reducing losses and optimizing inventory levels. In response to the above-mentioned deficiencies of existing technologies, this invention provides a method and system for agricultural product sales forecasting and inventory management based on big data analysis.

[0005] The technical solution adopted by this invention to solve its technical problem is: on the one hand A method for agricultural product sales forecasting and inventory management based on big data analysis includes the following steps: Data Acquisition Steps: Multimodal time-series quality characteristic data of agricultural products are acquired in real time through a composite sensing array deployed in the storage unit. The composite sensing array includes at least a near-infrared spectroscopy module for monitoring changes in product water activity and fatty acid value, a gas chromatography module for monitoring characteristic gases of mold growth in the storage environment, and a temperature and humidity sensing network for monitoring the temperature and humidity distribution of agricultural products. Analysis steps: Input the multimodal time-series quality feature data into a pre-trained physical information fusion neural network; the physical information fusion neural network is a deep learning model that embeds a dynamic model of agricultural product aging and mold growth as a physical constraint, and outputs the theoretical aging cycle and dynamic quality decay curve of an individual batch. Effective inventory measurement step: Based on the dynamic quality decay curve, the book inventory quantity of agricultural products is dynamically converted into standardized effective inventory equivalent, realizing the integrated calculation of inventory quantity and inventory quality; Procurement decision-making steps: Based on the standardized effective inventory equivalent, future sales demand forecasts, and historical supply quality data, a procurement plan is generated; Inventory decision-making steps: Based on the standardized effective inventory equivalent and individual theoretical aging cycle, generate differentiated inventory disposal instructions and adaptively adjust inventory management parameters.

[0006] Preferably, the near-infrared spectroscopy module is deployed at the sampling and testing station or automatic sampling device of the storage unit, using a near-infrared spectrometer in the 900-1700nm band, and using average spectral data from no less than 32 scans per sample to provide feedback on the product's moisture content, fatty acid value, protein content and starch aging degree. The gas chromatography module uses a microelectromechanical system integrated gas sensor, which is embedded in the warehouse ventilation duct or the top space of the shelf to collect the concentration of musty odor markers in the head air in real time, including aflatoxin metabolites, ethanol and acetic acid. The temperature and humidity sensing network is distributed in the form of wireless sensor nodes between shelf layers and inside the warehouse to monitor ambient temperature and relative humidity in real time and predict the risk of condensation and mold growth.

[0007] Preferably, the physical information fusion neural network includes: The multimodal feature encoding subnetwork adopts a parallel temporal convolutional network structure to extract features from the multimodal temporal quality feature data and fuse them to form a high-dimensional hidden layer feature vector. The physical information constrained subnetwork contains a differential equation model of the aging and mold growth dynamics of agricultural products. Using the high-dimensional hidden layer feature vector as the initial state and parameters, time integration is performed to deduce the theoretical trajectory of quality changes. The quality prediction subnetwork outputs the theoretical aging period and dynamic quality decay curve. The loss function of the physical information fusion neural network is composed of a weighted average of the data loss term between the predicted value and the measured value, and the physical constraint loss term between the theoretical trajectory and the predicted trajectory.

[0008] Preferably, in the commutation measurement step: The formula for calculating the standardized effective inventory equivalent is as follows: ,in For the first The quality degradation curve values ​​of each batch at time t. This is a conversion function based on a preset sellable quality threshold; when When the quality is greater than or equal to the marketable quality threshold, ;when When it is below the marketable quality threshold Furthermore, the batch was marked as an object to be downgraded or processed.

[0009] Preferably, the procurement decision-making steps include: Procurement demand forecasting steps: Based on the current standardized effective inventory level Preset safety stock equivalent threshold and future sales demand forecast Calculate the recommended purchase quantity ,in Equivalent to in-transit effective inventory; Standardized effective inventory equivalent This represents the true inventory after removing unsellable portions; future sales demand forecast. Forecasted sales volume within a predetermined future timeframe; Supplier quality assessment steps: Construct a supplier quality index based on the dynamic quality degradation curves of historical batches after warehousing. ,in, This refers to the number of historical supply batches from the supplier. For the first Theoretical marketable lifespan of the batch; For the first The batch dynamic quality degradation curve, wherein the supplier quality index is used to evaluate the storability and aging rate of agricultural products from different suppliers; Purchase order generation steps: Combine the suggested purchase quantity, the supplier quality index, and the time and price information to generate a purchase order that includes the quantity of supplied goods and the expected delivery time.

[0010] Preferably, the inventory decision-making step includes: Based on the individual theoretical aging cycle, all batches are sorted from shortest to longest, and a priority outbound instruction for the highest aging risk is automatically generated, prioritizing the matching of the batch with the shortest aging cycle to the short-chain sales or processing channels. Based on the real-time fluctuations of the standardized effective inventory equivalent, the safety stock threshold is dynamically adjusted. When the rate of decline of the standardized effective inventory equivalent exceeds the preset threshold, the safety stock coefficient is automatically increased. Based on the individual theoretical aging cycle, and according to the preset diversion rule engine, the same batch of agricultural products will be diverted to different disposal channels, which include at least long-term storage channels, regular sales channels and processing and disposal channels.

[0011] Preferably, it also includes a feedback optimization step: The actual sales data and actual loss data of agricultural products after they leave the warehouse, as well as the actual quality data after the purchase order arrives, are collected. The actual sales data, actual loss data, and actual quality data are used as feedback signals and input into the physical information fusion neural network and the procurement decision model to iteratively optimize the network parameters and model parameters.

[0012] on the other hand A sales forecasting and inventory management system for agricultural products based on big data analytics, used to implement the steps described in any one of the above methods, including: Multimodal sensing terminals, deployed in storage units, include near-infrared spectrometers, gas chromatographs, and temperature and humidity sensor networks, used to collect real-time data on the quality characteristics of agricultural products and storage environment. An edge computing gateway is communicatively connected to the multimodal sensing terminal and is used to collect data, perform data cleaning and feature extraction, and form multimodal time-series feature data. The physical information fusion prediction server is communicatively connected to the edge computing gateway and is deployed with a time-series convolutional and physical information fusion neural network. The physical information fusion neural network is embedded with a dynamic model of agricultural product aging and mold growth as a physical constraint, which is used to receive the multimodal time-series feature data and generate dynamic quality decay curves. The effective inventory calculation engine is connected to the physical information fusion prediction server and is used to convert the book inventory into standardized effective inventory equivalent based on the dynamic quality decay curve. The procurement decision engine communicates with the depreciated inventory calculation engine and is used to generate outbound instructions, replenishment instructions, and diversion instructions based on the standardized effective inventory equivalent and individual theoretical aging cycle.

[0013] Preferably, the inventory decision engine includes an outbound optimization module, a replenishment parameter adaptive module, and a multi-channel diversion module; the outbound optimization module is used to generate outbound instructions with the highest aging risk based on the individual theoretical aging cycle; the replenishment parameter adaptive module is used to dynamically adjust the safety stock threshold and capture time point based on the real-time fluctuation of the standardized effective inventory equivalent; the multi-channel diversion module is used to divert agricultural products to differentiated disposal channels according to the individual theoretical aging cycle and preset rules.

[0014] Preferably, the procurement decision engine includes a procurement demand forecasting module, a supplier quality assessment module, and a purchase order generation module; the procurement demand module is used to calculate the recommended procurement quantity based on the current standardized effective inventory equivalent, a preset safety stock equivalent threshold, and future sales demand forecasts; the supplier quality assessment module is used to construct a supplier quality index based on the dynamic quality decay curve after historical batches are put into storage; the purchase order generation module is used to generate a purchase order by combining the recommended procurement quantity, the supplier quality index, and market price information.

[0015] The beneficial effects of this invention are as follows: 1. This solution utilizes near-infrared spectroscopy, gas chromatography, and temperature and humidity sensor networks deployed in the storage unit to monitor the moisture content, fatty acid value, mold gases, and ambient temperature and humidity of pre-packaged agricultural products in real time, overcoming the limitations of traditional inventory management that relies on static shelf life. A temporal convolutional and physical information fusion neural network maps the sensed data into the theoretical aging cycle and dynamic quality decay curve of individual batches, enabling the system to accurately identify high-risk products approaching their marketable threshold. Based on this, the system generates a priority outbound instruction for products with the highest aging risk, ensuring that critically endangered products are preferentially matched to short-chain sales or processing channels. This fundamentally avoids passive scrapping due to failure to detect quality deterioration in a timely manner, significantly reducing the loss rate.

[0016] 2. The procurement decision engine generates procurement plans based on standardized effective inventory equivalents, future sales demand forecasts, and historical supply quality data. The procurement demand forecasting step incorporates current effective inventory equivalents, safety stock thresholds, and sales forecasts into a unified calculation model, avoiding over-purchasing or untimely replenishment issues caused by the disconnect between traditional procurement and inventory. The supplier quality assessment step utilizes dynamic quality decay curves from historical batches to construct a supplier quality index, enabling procurement decisions to quantitatively assess the differences in storability of agricultural products from different suppliers. This allows for the selection of suppliers with superior quality and slower decay from the source, achieving deep collaboration between procurement and inventory.

[0017] 3. The system transforms the historical quality of supplier deliveries into quantifiable and comparable continuous indicators. This index comprehensively reflects the supplier's ability to maintain the quality of agricultural products throughout their entire marketable lifespan, encompassing both the time dimension of the aging period and the quality dimension of the rate of quality degradation, thus overcoming the limitations of traditional supplier evaluations that rely on static sampling. As the system operates for an extended period, the supplier quality index is continuously updated, forming an accurate profile of the supplier's storability, providing objective data support for purchase order allocation, dynamic adjustment of safety stock, and supplier tiered management.

[0018] 4. When the rate of decline of effective inventory equivalent exceeds a preset threshold (i.e., a sharp decrease in effective inventory due to rapid quality deterioration), the system automatically increases the safety stock coefficient, triggering replenishment demand in advance and avoiding sales gaps caused by quality issues. This dynamic adjustment mechanism enables inventory management parameters to respond in real time to fluctuations in agricultural product quality, shifting the benchmark of inventory management from static book quantities to dynamic effective value inventory, significantly improving the system's adaptability to quality uncertainties.

[0019] 5. This application, based on individual theoretical aging cycles, automatically diverts batches of the same agricultural product to differentiated disposal channels according to a preset diversion rule engine. Batches with longer aging cycles are matched to long-term storage or high-end sales channels, batches with medium aging cycles are matched to regular sales channels, and batches with shorter remaining sales time are automatically assigned processing orders to food processing enterprises. This refined diversion strategy allows the same batch of agricultural products to realize its maximum value at different quality stages, ensuring the quality requirements of high-end channels while avoiding the waste of value for near-critical products, thus maximizing the value of agricultural products throughout their entire life cycle.

[0020] 6. Timely loss risk warnings and efficient inventory disposal instructions enable critically damaged goods to be quickly sold before they completely lose value, reducing direct economic losses caused by passive scrapping. The introduction of a supplier quality index allows companies to prioritize the procurement of agricultural products with strong storage resistance, extending the average saleable period of inventory and further reducing cash flow pressure. In summary, this solution significantly reduces inventory capital tied up and optimizes corporate cash flow and overall operational efficiency. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be further described below in conjunction with the accompanying drawings and embodiments. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Figure 1 This is a schematic diagram illustrating the steps of the agricultural product sales forecasting and inventory management method based on big data analysis in an embodiment of this application.

[0022] Figure 2 This is a schematic diagram of the composition of the agricultural product sales forecasting and inventory management system based on big data analysis, as described in this application embodiment. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, a clear and complete description will be provided below in conjunction with the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the protection scope of the present invention.

[0024] Example 1 Preferred embodiments of the present invention, for example Figure 1 As shown, a method for agricultural product sales forecasting and inventory management based on big data analysis includes the following steps: S1. Data Acquisition Steps: Real-time acquisition of multimodal time-series quality characteristic data of agricultural products through a composite sensing array deployed in the storage unit; the composite sensing array includes at least a near-infrared spectroscopy module for monitoring changes in product water activity and fatty acid value, a gas chromatography module for monitoring moldy characteristic gases in the storage environment, and a temperature and humidity sensing network for monitoring the temperature and humidity distribution of agricultural products.

[0025] S2. Analysis Steps: Multimodal time-series quality feature data is input into a pre-trained physical information fusion neural network. This network is a deep learning model that embeds a dynamic model of agricultural product aging and mold growth as physical constraints. The output of the physical information fusion neural network is the remaining high-quality sales period and dynamic quality decay curve for individual or batch agricultural products. The physical information fusion neural network includes a multimodal feature encoding subnetwork, a physical information constraint subnetwork, and a quality prediction subnetwork. The multimodal feature encoding subnetwork uses a parallel temporal convolutional network structure to extract features from the multimodal time-series quality feature data and fuse them to form a high-dimensional hidden layer feature vector. The physical information constraint subnetwork embeds a postharvest physiological differential equation model for agricultural products, using the high-dimensional hidden layer feature vector as the initial state and parameters for time integration to deduce the theoretical trajectory of quality changes. The quality prediction subnetwork outputs the remaining high-quality sales period and dynamic quality decay curve. The loss function of the physical information fusion neural network is a weighted sum of the data loss term between the predicted and measured values, and the physical constraint loss term between the theoretical trajectory and the predicted trajectory.

[0026] S3. Effective Inventory Measurement Step: Based on the dynamic quality decay curve, the book inventory quantity of agricultural products is dynamically converted into standardized effective inventory equivalent, realizing the integrated calculation of inventory quantity and inventory quality. The formula for calculating standardized effective inventory equivalent is: ,in For the first The quality degradation curve values ​​of each batch at time t. This is a conversion function based on a preset sellable quality threshold; when When the quality is greater than or equal to the marketable quality threshold, ;when When it is below the marketable quality threshold Furthermore, this batch was marked as an object for downgrading or processing. Specifically, , The value range is [0, 1], and the output is achieved by a physical information fusion neural network; 1 represents fresh out of the factory, and 0 represents the complete loss of commodity value. This is a preset threshold for marketable quality, such as 0.6. In terms of quality, this can include the degree of aging of agricultural products, the risk of mold, and the degree of oxidative rancidity. Monitoring indicators can correspond to fatty acid value and starch aging degree, moldy gas concentration and water activity, as well as peroxide value for edible oils. This transforms the slow quality deterioration process into a clear binary state of being marketable or unmarketable.

[0027] S4. Procurement Decision-Making Steps: Generate a procurement plan based on standardized effective inventory equivalents, future sales demand forecasts, and historical supply quality data.

[0028] S41. Procurement demand forecasting steps: Based on the current standardized effective inventory level... Preset safety stock equivalent threshold and future sales demand forecast Calculate the recommended purchase quantity ,in This represents the equivalent of in-transit inventory. In the warehousing management of pre-packaged rice, flour, and cooking oil, all inventory variables have undergone quality discounting. This is the actual inventory after removing unsellable portions. The system takes into account potential quality loss during transportation. When effective inventory decreases rapidly (e.g., due to accelerated aging caused by environmental changes), the system automatically triggers replenishment to prevent sales disruptions caused by quality deterioration. For example, in practical applications, the system forecasts sales volume for flour warehouses over the next seven days. 1000 bags; safety stock 200 bags; current effective inventory The equivalent of 800 bags (i.e., 1000 bags on paper, but 200 bags have aged); the equivalent of effective inventory in transit. The quantity is 150 bags (i.e., 200 bags are in transit, with an estimated arrival quality of 0.75); therefore, If so, replenishment is required.

[0029] S42. Supplier Quality Assessment Steps: Construct a supplier quality index based on the dynamic quality degradation curves of historical batches after warehousing. ,in, This refers to the number of historical supply batches from the supplier. For the first Theoretical marketable lifespan of the batch; For the first The batch dynamic quality decay curve and the supplier quality index are used to assess the storability and aging rate of agricultural products from different suppliers. The dynamic quality decay curve is a function curve that changes over time. The vertical axis of the function curve is the standardized commodity value, ranging from 0 to 1, and the horizontal axis is the storage time. The function curve is dynamically updated based on real-time sensing data to respond to the impact of changes in the storage environment on the quality of agricultural products. This represents the area under the quality curve. For pre-packaged rice, flour, and oil products, different suppliers' raw materials, processing techniques, and packaging methods can lead to significant differences in aging rates: simply comparing the number of days available for sale... While other indices cannot reflect the product's quality maintenance level throughout its lifecycle, this index calculates the ratio of the number of days each batch can be sold to the area of ​​the quality curve integral. A smaller ratio indicates that the batch has maintained higher quality over a longer period (i.e., a larger integral area), suggesting better supplier quality. By averaging the index across multiple historical batches, the system creates a dynamically updated supplier quality profile. This provides objective and multi-dimensional decision-making support for purchase order allocation, safety stock adjustment, and supplier tiered management, ensuring the overall shelf life of inventory from the source. S43. Purchase order generation steps: Combine the suggested purchase quantity, supplier quality index, and time and price information to generate a purchase order that includes the quantity of supplied goods and the expected delivery time.

[0030] S5. Inventory Decision-Making Steps: Based on standardized effective inventory equivalents and individual theoretical aging cycles, generate differentiated inventory disposal instructions and adaptively adjust inventory management parameters.

[0031] S51. Based on the individual theoretical aging cycle, all batches are sorted from shortest to longest, and the highest aging risk priority delivery instruction is automatically generated. The batch with the shortest remaining sales time is prioritized for matching to short-term sales channels or processing channels.

[0032] S52. Based on the real-time fluctuations of standardized effective inventory equivalent, the safety stock threshold is dynamically adjusted. When the rate of decline of standardized effective inventory equivalent exceeds the preset threshold, the safety stock coefficient is automatically increased.

[0033] S53. Based on the individual theoretical aging cycle, according to the preset diversion rule engine, the unified agricultural product batches are diverted to different disposal channels, including at least long-term storage channels, regular sales channels and processing and disposal channels.

[0034] S6. Feedback Optimization Steps: Collect actual sales data and actual loss data of agricultural products after they leave the warehouse, as well as actual quality data of purchase orders after they arrive. Input the actual sales data, actual loss data, and actual quality data as feedback signals into the physical information fusion neural network and the procurement decision model, and iteratively optimize the network parameters and model parameters.

[0035] Example 2 A big data analytics-based agricultural product sales forecasting and inventory management system, used to implement the method steps of any one of Embodiment 1, refer to... Figure 2 It includes a multimodal sensing terminal, an edge computing gateway, a physical information fusion prediction server, a depreciation inventory calculation engine, an inventory decision engine, and a procurement decision engine.

[0036] Multimodal sensing terminals are deployed in the storage unit, including near-infrared spectrometers, gas chromatographs, and temperature and humidity sensor networks, to collect real-time data on the quality characteristics and storage environment of pre-packaged rice, flour, and oil products. The near-infrared spectroscopy module is deployed at sampling and testing stations or automatic sampling devices in the storage unit, using a near-infrared spectrometer in the 900-1700nm band. It uses average spectral data from at least 32 scans per sample to provide feedback on the product's moisture content, fatty acid value, protein content, and starch retrogradation degree. The gas chromatography module is embedded in the warehouse ventilation ducts or the top space of the shelves as a miniature sensor array. It uses microelectromechanical systems integrated gas sensors to collect the concentration of musty odor markers in the headspace gas in real-time at a preset sampling frequency, including aflatoxin metabolites, ethanol, and acetic acid. The temperature and humidity sensor network is distributed as wireless sensor nodes between shelf layers and inside the warehouse to monitor ambient temperature and relative humidity in real-time and predict the risk of condensation and mold growth.

[0037] The edge computing gateway communicates with the multimodal sensing terminal to perform data cleaning and feature extraction on visual quality, respiratory metabolism and texture decay data, forming multimodal time-series feature data.

[0038] The physical information fusion prediction server communicates with the edge computing gateway and deploys a time-series convolutional and physical information fusion neural network. The physical information fusion neural network is embedded with a dynamic model of agricultural product aging and mold growth as physical constraints, which is used to receive multimodal time-series feature data and generate dynamic quality decay curves.

[0039] The effective inventory calculation engine communicates with the physical information fusion prediction server to convert book inventory into standardized effective inventory equivalent based on the dynamic quality decay curve.

[0040] The inventory decision engine communicates with this smaller inventory calculation engine to generate outbound, replenishment, and diversion instructions based on standardized effective inventory equivalents and individual theoretical aging cycles. The inventory decision engine includes an outbound optimization module, a replenishment parameter adaptive module, and a multi-channel diversion module. The outbound optimization module generates priority outbound instructions based on individual theoretical aging cycles, prioritizing those with the highest aging risk. The replenishment parameter adaptive module dynamically adjusts the safety stock threshold and capture time based on real-time fluctuations in standardized effective inventory equivalents. The multi-channel diversion module diverts agricultural products to differentiated disposal channels according to preset rules based on individual theoretical aging cycles.

[0041] The procurement decision engine communicates with the depreciation inventory calculation engine to generate procurement plans based on standardized effective inventory equivalents and individual theoretical aging cycles. The procurement decision engine includes a procurement demand forecasting module, a supplier quality assessment module, and a purchase order generation module. The procurement demand module calculates the recommended procurement quantity based on the current standardized effective inventory equivalent, a preset safety stock equivalent threshold, and future sales demand forecasts. The supplier quality assessment module constructs a supplier quality index based on the dynamic quality decay curves of historical batches after warehousing. The purchase order generation module generates purchase orders by combining the recommended procurement quantity, the supplier quality index, and market price information.

[0042] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for agricultural product sales forecasting and inventory management based on big data analysis, characterized in that, Includes the following steps: Data collection steps: Real-time collection of multimodal time-series quality characteristic data of agricultural products is achieved through a composite sensing array deployed in the storage unit; The composite sensing array includes at least a near-infrared spectroscopy module for monitoring changes in product water activity and fatty acid value, a gas chromatography module for monitoring characteristic gases of mold growth in the storage environment, and a temperature and humidity sensing network for monitoring the temperature and humidity distribution of agricultural products. Analysis steps: Input the multimodal time-series quality feature data into a pre-trained physical information fusion neural network; the physical information fusion neural network is a deep learning model that embeds a dynamic model of agricultural product aging and mold growth as a physical constraint, and outputs the theoretical aging cycle and dynamic quality decay curve of an individual batch. Effective inventory measurement step: Based on the dynamic quality decay curve, the book inventory quantity of agricultural products is dynamically converted into standardized effective inventory equivalent, realizing the integrated calculation of inventory quantity and inventory quality; Procurement decision-making steps: Based on the standardized effective inventory equivalent, future sales demand forecasts, and historical supply quality data, a procurement plan is generated; Specifically, this includes: the procurement demand forecasting step: based on standardized effective inventory equivalents. Preset safety stock equivalent threshold and future sales demand forecast Calculate the recommended purchase quantity ,in Equivalent to in-transit effective inventory; Standardized effective inventory equivalent This represents the true inventory after removing unsellable portions; future sales demand forecast. Forecasted sales volume within a predetermined future timeframe; Supplier quality assessment steps: Construct a supplier quality index based on the dynamic quality degradation curves of historical batches after warehousing. ,in, This refers to the number of historical supply batches from the supplier. For the first Theoretical marketable lifespan of the batch; For the first The batch dynamic quality degradation curve, wherein the supplier quality index is used to evaluate the storability and aging rate of agricultural products from different suppliers; Purchase order generation steps: Combine the suggested purchase quantity, the supplier quality index, and time and price information to generate a purchase order that includes the quantity of supplied goods and the expected delivery time; Inventory decision-making steps: Based on the standardized effective inventory equivalent and individual theoretical aging cycle, generate differentiated inventory disposal instructions and adaptively adjust inventory management parameters.

2. The big data analytics based agri-product sales forecasting and inventory management method as claimed in claim 1, wherein, The near-infrared spectroscopy module is deployed at the sampling and testing station or automatic sampling device of the storage unit. It uses a near-infrared spectrometer in the 900-1700nm band and uses average spectral data from no less than 32 scans per sample to provide feedback on the product's moisture content, fatty acid value, protein content and starch aging degree. The gas chromatography module uses a microelectromechanical system integrated gas sensor, which is embedded in the warehouse ventilation duct or the top space of the shelf to collect the concentration of musty odor markers in the head air in real time, including aflatoxin metabolites, ethanol and acetic acid. The temperature and humidity sensing network is distributed in the form of wireless sensor nodes between shelf layers and inside the warehouse to monitor ambient temperature and relative humidity in real time and predict the risk of condensation and mold growth. 3.The big data analysis based agricultural product sales forecasting and inventory management method according to claim 1, wherein, The physical information fusion neural network includes: The multimodal feature encoding subnetwork adopts a parallel temporal convolutional network structure to extract features from the multimodal temporal quality feature data and fuse them to form a high-dimensional hidden layer feature vector. The physical information constrained subnetwork contains a differential equation model of the aging and mold growth dynamics of agricultural products. Using the high-dimensional hidden layer feature vector as the initial state and parameters, time integration is performed to deduce the theoretical trajectory of quality changes. The quality prediction subnetwork outputs the theoretical aging period and dynamic quality decay curve. The loss function of the physical information fusion neural network is composed of a weighted average of the data loss term between the predicted value and the measured value, and the physical constraint loss term between the theoretical trajectory and the predicted trajectory.

4. The big data analytics based agri-product sales forecasting and inventory management method as claimed in claim 1, wherein, In the aforementioned cost-effectiveness measurement step: The formula for calculating the standardized effective inventory equivalent is as follows: ,in For the first The quality degradation curve values ​​of each batch at time t. This is a conversion function based on a preset sellable quality threshold; When greater than or equal to a salable quality threshold, ; when less than a salable quality threshold, and marking the batch for downgraded processing or disposal. 5.The method of predicting sales of agricultural products and managing inventory based on big data analysis according to claim 1, wherein, The inventory decision-making steps include: Based on the individual theoretical aging cycle, all batches are sorted from shortest to longest, and the highest aging risk priority delivery instruction is automatically generated, and the batch with the shortest aging cycle is prioritized for matching to short-term sales channels or processing channels. Based on the real-time fluctuations of the standardized effective inventory equivalent, the safety stock threshold is dynamically adjusted. When the rate of decline of the standardized effective inventory equivalent exceeds the preset threshold, the safety stock coefficient is automatically increased. Based on the individual theoretical aging cycle, and according to the preset diversion rule engine, the same batch of agricultural products will be diverted to different disposal channels, which include at least long-term storage channels, regular sales channels and processing and disposal channels. 6.The big data analysis based agricultural product sales forecasting and inventory management method according to claim 1, wherein, It also includes feedback optimization steps: The actual sales data and actual loss data of agricultural products after they leave the warehouse, as well as the actual quality data after the purchase order arrives, are collected. The actual sales data, actual loss data, and actual quality data are used as feedback signals and input into the physical information fusion neural network and the procurement decision model to iteratively optimize the network parameters and model parameters.

7. A big data analysis based agricultural product sales forecasting and inventory management system for implementing the method steps of any one of claims 1-6, characterized in that, include: Multimodal sensing terminals, deployed in storage units, include near-infrared spectrometers, gas chromatographs, and temperature and humidity sensor networks, used to collect real-time data on the quality characteristics of agricultural products and storage environment. An edge computing gateway is communicatively connected to the multimodal sensing terminal and is used to collect data, perform data cleaning and feature extraction, and form multimodal time-series feature data. The physical information fusion prediction server is communicatively connected to the edge computing gateway and is deployed with a time-series convolutional and physical information fusion neural network. The physical information fusion neural network is embedded with a dynamic model of agricultural product aging and mold growth as a physical constraint, which is used to receive the multimodal time-series feature data and generate dynamic quality decay curves. The effective inventory calculation engine is connected to the physical information fusion prediction server and is used to convert the book inventory into standardized effective inventory equivalent based on the dynamic quality decay curve. The procurement decision engine communicates with the depreciated inventory calculation engine and is used to generate outbound instructions, replenishment instructions, and diversion instructions based on the standardized effective inventory equivalent and individual theoretical aging cycle.

8. The big data analytics based agri-product sales forecasting and inventory management system as claimed in claim 7, wherein, The inventory decision engine includes an outbound optimization module, a replenishment parameter adaptive module, and a multi-channel distribution module. The outbound optimization module is used to generate outbound instructions with the highest aging risk based on the individual theoretical aging cycle. The replenishment parameter adaptive module is used to dynamically adjust the safety stock threshold and capture time point based on the real-time fluctuation of the standardized effective inventory equivalent. The multi-channel diversion module is used to divert agricultural products to differentiated disposal channels according to preset rules based on individual theoretical aging cycles.

9. The big data analytics based agri-product sales forecasting and inventory management system as claimed in claim 7, wherein, The procurement decision engine includes a procurement demand forecasting module, a supplier quality assessment module, and a purchase order generation module. The procurement demand forecasting module is used to calculate the recommended procurement quantity based on the current standardized effective inventory equivalent, the preset safety stock equivalent threshold, and the future sales demand forecast; the supplier quality assessment module is used to construct a supplier quality index based on the dynamic quality decay curve after historical batches are put into storage; the purchase order generation module is used to generate a purchase order by combining the recommended procurement quantity, the supplier quality index, and market price information.

Images