Intelligent storage demand prediction scheduling method and system in combination with AIAgent

Abstract

The invention provides an intelligent storage demand prediction scheduling method and system combined with an AIAgent, and the method comprises the steps: obtaining a historical storage data set of a target storage scene, carrying out the demand feature extraction processing through the AIAgent, and generating a storage demand fluctuation feature and an inventory dynamic distribution feature; the method comprises the following steps: constructing an inventory-demand matching model based on warehousing demand fluctuation characteristics and preset market trend prediction data, calling a dynamic path optimization algorithm to generate a warehousing operation sequence according to inventory dynamic distribution characteristics and inventory adjustment priorities, and carrying out collaborative scheduling processing on the warehousing operation sequence based on replenishment trigger conditions, and generating an optimized warehousing operation instruction set, and feeding back the warehousing operation instruction set to the warehousing control system to execute inventory optimization operation. The method can effectively reduce the inventory redundancy caused by the demand prediction deviation and the operation cost caused by path repeated planning, and improves the real-time response capability of the warehousing system to the market fluctuation and the resource utilization rate.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence, and more specifically, to an intelligent warehouse demand forecasting and scheduling method and system that incorporates AI Agent. Background Technology

[0002] With the increasing demand for intelligent logistics warehousing, warehouse scheduling technology has become a core means of optimizing inventory management and operational efficiency. Existing warehouse scheduling technologies typically use historical inbound and outbound data for demand forecasting, combined with fixed rules for warehouse location allocation and route planning, such as predicting replenishment cycles through time-series statistical models or generating picking routes based on shortest path algorithms. However, such methods struggle to dynamically integrate market trend changes with real-time inventory status, leading to a disconnect between demand forecasting and operational execution: on the one hand, static forecasting models cannot capture sudden demand fluctuations, resulting in inventory backlogs or shortages; on the other hand, route planning and replenishment strategies are optimized independently, lacking a dynamic coordination mechanism based on inventory urgency, easily causing equipment movement conflicts and resource waste. Furthermore, traditional manual intervention adjustment methods suffer from response lag and subjective bias, making them difficult to adapt to high-frequency market changes. Summary of the Invention

[0003] This invention provides an intelligent warehouse demand forecasting and scheduling method and system that combines AI Agent.

[0004] In a first aspect, embodiments of the present invention provide an intelligent warehouse demand forecasting and scheduling method incorporating AI Agent, comprising the following steps:

[0005] Obtain a set of historical warehousing data for the target warehousing scenario. The set of historical warehousing data includes multiple historical warehousing operation sequences. Each historical warehousing operation sequence consists of at least one inbound operation record, one outbound operation record, and one inventory status update record.

[0006] The historical warehousing data set is processed by AIAgent to extract demand features and generate warehousing demand fluctuation features and inventory dynamic distribution features.

[0007] Based on the warehousing demand fluctuation characteristics and the preset market trend forecast data, an inventory-demand matching model is constructed. The inventory-demand matching model is used to output the inventory adjustment priority and replenishment trigger conditions.

[0008] Based on the dynamic distribution characteristics of the inventory and the priority of inventory adjustment, a dynamic path optimization algorithm is invoked to generate a warehousing operation sequence, which includes a storage location allocation strategy, picking route planning, and replenishment execution instructions.

[0009] Based on the replenishment triggering conditions, the warehousing operation sequence is coordinated and scheduled to generate an optimized warehousing operation instruction set, which is then fed back to the warehousing control system to perform inventory optimization operations.

[0010] In a second aspect, embodiments of the present invention provide a computer system, comprising:

[0011] A memory, wherein a computer program is stored;

[0012] A processor is used to load the computer program to implement the intelligent warehouse demand forecasting and scheduling method combined with AIAgent as described above.

[0013] This invention provides an intelligent warehouse demand forecasting and scheduling method that integrates AI agents. By fusing historical warehouse data with market trend forecasting data, it generates warehouse demand fluctuation characteristics and dynamic inventory distribution characteristics, breaking through the separation between traditional static demand forecasting and inventory management. This enables dynamic correlation modeling of demand fluctuation patterns and inventory spatial distribution. The inventory-demand matching model, built based on dynamic difference ratios and time window matching, can automatically generate replenishment priorities and triggering conditions according to real-time inventory status, solving the response lag problem caused by traditional reliance on manual experience adjustments. A dynamic path optimization algorithm deeply integrates inventory adjustment strategies with equipment movement paths, generating picking path planning and replenishment execution instructions based on multi-objective collaborative computation, eliminating the drawbacks of path conflicts and uneven resource allocation. Finally, a collaborative scheduling mechanism links warehouse allocation, path planning, and replenishment strategies for execution, forming an adaptive optimization chain from demand forecasting to operation execution. This effectively reduces the operational costs caused by inventory redundancy and redundant path planning due to demand forecasting deviations, while improving the warehouse system's real-time response capability to market fluctuations and resource utilization. Attached Figure Description

[0014] Figure 1 This is a flowchart of an intelligent warehouse demand forecasting and scheduling method that combines AIAgent, provided by an embodiment of the present invention.

[0015] Figure 2 This is a schematic diagram of the composition of a computer system provided in an embodiment of the present invention. Detailed Implementation

[0016] Please see Figure 1 , Figure 1 A flowchart of an intelligent warehouse demand forecasting and scheduling method combining AIAgent is provided for an embodiment of the present invention. The method is executed by a computer system and includes the following steps:

[0017] Step S100: Obtain the historical warehousing data set of the target warehousing scenario. The historical warehousing data set includes multiple historical warehousing operation sequences. Each historical warehousing operation sequence consists of at least one inbound operation record, one outbound operation record, and one inventory status update record.

[0018] The target warehousing scenario refers to the specific warehousing environment where demand forecasting and scheduling are to be performed. It has a pre-defined warehousing layout, operating model, and business requirements. The historical warehousing data set is the sum of warehousing operation-related data accumulated over a period of time within this target warehousing scenario. The historical warehousing operation sequence is a collection of warehousing operation records arranged chronologically. Inbound operation records refer to information related to goods entering the warehousing system, such as the name, quantity, inbound time, and supplier of the goods. Outbound operation records are records of goods leaving the warehousing system, including the name, quantity, outbound time, and customer of the goods. Inventory status update records are records that update the inventory quantity, location, and other statuses of goods within the warehouse after each inbound or outbound operation. Obtaining the historical warehousing data set of the target warehousing scenario can be achieved through a Warehouse Management System (WMS). The WMS records every warehousing operation; by organizing these records chronologically, the historical warehousing operation sequence can be obtained. For example, in the past month, an e-commerce warehouse has different numbers of goods entering and leaving the warehouse every day. The WMS will record every entry and exit operation in detail, as well as the changes in inventory after each operation. By summarizing the records for this month, a historical warehouse data set is formed.

[0019] Step S200: Extract demand features from the historical warehousing data set using AIAgent to generate warehousing demand fluctuation features and inventory dynamic distribution features.

[0020] AIAgent, or Artificial Intelligence Agent, is an intelligent program capable of autonomously executing tasks. It analyzes and processes input data according to pre-defined algorithms and rules. Demand feature extraction processing refers to identifying characteristic information from historical warehousing data sets that reflects the patterns of changes in warehousing demand and inventory status. Warehouse demand fluctuation characteristics describe the changes in warehousing demand over time, such as peak and trough periods and cyclical changes in demand. Inventory dynamic distribution characteristics refer to the changes in the distribution of inventory goods within the warehouse space over time, including the quantity and type of inventory in different storage areas. By using AIAgent to extract demand features from historical warehousing data sets, machine learning algorithms, such as neural network algorithms, can be employed. The historical warehousing data sets are used as input, and after calculation and learning by multiple layers of neural networks, the output warehousing demand fluctuation characteristics and inventory dynamic distribution characteristics are obtained.

[0021] As one implementation method, the historical warehousing data set includes multiple warehousing operation records. Each warehousing operation record contains an order execution timestamp, goods category identifier, inventory capacity change, and related parameters from the corresponding market trend forecast data. Based on this, step S200 may specifically include the following steps S210 to S250:

[0022] Step S210: Perform time-series alignment processing on the inbound and outbound operation records based on the order execution timestamp to generate a time-series warehousing feature sequence associated with the inventory status update record. The time-series warehousing feature sequence includes the total inventory change rate and goods turnover rate corresponding to each timestamp.

[0023] An order execution timestamp refers to the specific point in time when an order is executed, accurate to the year, month, day, hour, minute, and second, used to determine the sequence of order operations. Time-series alignment involves sorting inbound and outbound operation records according to their order execution timestamps, ensuring temporal consistency for better analysis of inventory status changes. A time-series warehouse feature sequence is a collection of warehouse features arranged chronologically. The inventory change rate refers to the percentage change in total inventory at each timestamp relative to the previous timestamp, reflecting inventory increases or decreases. The goods turnover rate refers to the speed at which goods flow through the warehouse system, expressed as the ratio of outbound volume to average inventory within a given time period. Time-series alignment of inbound and outbound operation records based on order execution timestamps can be achieved using sorting algorithms, such as quicksort, to sort the records in ascending order of timestamps. Then, by calculating the change in total inventory and goods turnover at each timestamp, a time-series warehouse feature sequence is generated. For example, in a logistics warehouse, there are multiple inbound and outbound orders every day. After sorting these orders according to their execution timestamps, the inventory change rate and cargo turnover rate at each point in time can be calculated to obtain a time-series warehousing characteristic sequence.

[0024] Step S220: Based on the cargo category identifier, classify and aggregate the time-series warehousing feature sequence to generate category demand sequences for different cargo categories. The category demand sequences are used to characterize the historical demand change trend of the same cargo category.

[0025] Goods category identifiers are used to distinguish different types of goods; for example, they can be product numbers or names. Categorization and aggregation processing involves grouping time-series warehousing feature sequences according to goods category identifiers, then performing statistical analysis on the data within each group to obtain the comprehensive characteristics of each goods category. Category demand sequences are demand data sequences for a specific goods category, reflecting historical demand changes for that category, such as growth or decline trends in demand, and seasonal variations. Categorization and aggregation processing of time-series warehousing feature sequences based on goods category identifiers can use hash table algorithms, storing the goods category identifier as the key and the corresponding time-series warehousing feature sequence as the value in the hash table. Then, statistical operations such as summation and averaging are performed on the data corresponding to each goods category to generate the category demand sequence. For example, in a large supermarket warehouse, goods categories include food, daily necessities, and appliances. Through categorization and aggregation processing, the demand sequence for each category can be obtained, thereby understanding the demand trends of different categories of goods.

[0026] Step S230: Extract periodic demand patterns and sudden demand offsets from the category demand sequence. The periodic demand patterns are identified by the sliding time window algorithm to identify recurring demand peak intervals. The sudden demand offsets are obtained by calculating the dynamic deviation between the category demand sequence and the correlation parameters in the market trend forecast data.

[0027] Periodic demand patterns refer to the regular changes in demand for goods within a certain time frame, such as peak demand for certain commodities at fixed times each week, month, or year. The sliding window algorithm is an algorithm for processing time series data. It analyzes and processes the data within a fixed-size window that slides across the time series. Sudden demand shifts refer to the deviation between normal and sudden demand caused by unexpected factors, such as sudden events or promotional activities that cause a sudden increase or decrease in demand. Correlation parameters in market trend forecasting data are indicators related to market trends, such as market growth rate and consumer confidence index. To extract periodic demand patterns from categorical demand sequences, the sliding window algorithm is used. A suitable window size is set, and the window slides across the categorical demand sequence. The peak demand within each window is counted, and the recurring peak intervals are identified as periodic demand patterns. Sudden demand shifts are calculated by comparing each data point in the categorical demand sequence with the correlation parameters in the market trend forecasting data. The difference between them is the dynamic deviation. These dynamic deviations are then summarized and analyzed to obtain the sudden demand shift. For example, for a seasonal beverage, a sliding time window algorithm can be used to discover that there is a demand peak every summer, which is a cyclical demand pattern; and if demand suddenly increases during a non-summer period due to promotional activities, the sudden demand offset can be obtained by calculating the deviation from the market trend forecast data.

[0028] Step S240: Aggregate the changes in inventory capacity by storage location area to generate a capacity change sequence for each storage location area, and perform spatial clustering on the capacity change sequence to determine the dynamic distribution characteristics of inventory in different storage locations.

[0029] Inventory capacity change refers to the change in inventory capacity of a storage location relative to the previous value after each inbound or outbound operation. Aggregation by storage location involves summarizing and organizing all inventory capacity changes related to a specific storage location to form a capacity change sequence for that location. Spatial clustering is an algorithm that groups spatial data, classifying data points into different categories based on spatial distance and similarity. The dynamic distribution characteristics of inventory in different storage locations refer to the changes in inventory capacity of each storage location over time and the inventory distribution relationships between different storage locations. Aggregating inventory capacity changes by storage location can be done using a dictionary data structure, storing the storage location as the key and the corresponding inventory capacity change as the value. Then, the inventory capacity changes for each storage location are sorted to generate a capacity change sequence. Spatial clustering of the capacity change sequence can be performed using the DBSCAN (density-based spatial clustering application) algorithm, which divides the storage location into different clusters based on the spatial location and inventory capacity changes, thereby determining the dynamic distribution characteristics of inventory in different storage locations. For example, in a large warehouse, by dividing the warehouse into multiple storage areas and aggregating and clustering the changes in inventory capacity of each storage area, we can understand the dynamic distribution of inventory in different storage areas.

[0030] Step S250: Overlay the cyclical demand pattern with the market trend forecast data to generate the overlaid warehouse demand fluctuation characteristics. At the same time, correct the warehouse demand fluctuation characteristics according to the sudden demand offset to generate the final corrected warehouse demand fluctuation characteristics.

[0031] Trend overlay combines cyclical demand patterns and market trend forecasts, comprehensively considering their influence to obtain a more complete demand fluctuation characteristic. Market trend forecasts reflect the overall market trend, while cyclical demand patterns reflect the cyclical changes in demand for goods themselves. Overlaying them can more accurately describe the fluctuations in warehousing demand. Correcting the warehousing demand fluctuation characteristic based on sudden demand offsets takes into account the impact of unforeseen factors on demand, adjusting the fluctuation characteristics to better reflect reality. Trend overlaying cyclical demand patterns and market trend forecasts can be done using a linear overlay method, adding the values ​​of the cyclical demand patterns and market trend forecasts at the same time point to obtain the overlaid warehousing demand fluctuation characteristic. Then, based on sudden demand offsets, the overlaid warehousing demand fluctuation characteristic is adjusted. For example, if the sudden demand offset is positive, indicating a sudden increase in demand, the value of the warehousing demand fluctuation characteristic is increased at the corresponding time point; if the sudden demand offset is negative, indicating a sudden decrease in demand, the value of the warehousing demand fluctuation characteristic is decreased at the corresponding time point, thus generating the final corrected warehousing demand fluctuation characteristic. For example, for an electronic product, cyclical demand patterns show peak demand during holidays, and market trend forecasts indicate an overall upward trend in market demand. Superimposing these two data points yields a preliminary characteristic of warehousing demand fluctuations. If, during a certain period, a sudden increase in demand occurs due to the release of a new product, the preliminary characteristic of warehousing demand fluctuations can be adjusted based on the offset of this sudden demand, resulting in a more accurate final characteristic of warehousing demand fluctuations.

[0032] As one implementation method, step S200 involves extracting demand features from the historical warehousing data set using AIAgent, which can be specifically implemented as follows: S201 to S205:

[0033] Step S201: Jointly analyze the final corrected warehousing demand fluctuation characteristics and inventory dynamic distribution characteristics to generate a multi-dimensional warehousing characteristic set, which includes demand time series correlation degree and storage location spatial distribution density.

[0034] Joint analysis combines the finalized characteristics of warehouse demand fluctuations with the dynamic distribution characteristics of inventory, considering the interrelationships and influences between the two. A multi-dimensional warehouse feature set is a collection of features across multiple dimensions, providing a more comprehensive description of the warehouse system's state. Demand temporal correlation refers to the degree of correlation between changes in warehouse demand over time and changes in inventory status, reflecting the temporal synchronicity between demand and inventory. Storage location spatial distribution density refers to the density of goods distributed within a storage location area, reflecting the utilization of storage locations. Joint analysis of the finalized characteristics of warehouse demand fluctuations and the dynamic distribution characteristics of inventory can be performed using correlation analysis algorithms, such as the Pearson correlation coefficient algorithm, to calculate the demand temporal correlation. Storage location spatial distribution density can be obtained by calculating the ratio of the quantity of goods within each storage location area to the area of ​​that area. For example, in a warehouse, analyzing the correlation between demand fluctuations for certain commodities and inventory changes in different storage location areas, as well as the distribution density of goods in each storage location area, can generate a multi-dimensional warehouse feature set.

[0035] Step S202: Perform sliding window segmentation on the demand time-series correlation in the multi-dimensional warehouse feature set to generate multiple local demand trend segments, and extract the peak duration and demand change rate from each local demand trend segment.

[0036] Sliding window segmentation involves sliding a fixed-size window across the demand time-series correlation data, dividing the data into multiple local segments for individual analysis. A local demand trend segment refers to the changing trend of demand time-series correlation within a certain time range. Peak duration refers to the duration of the peak demand within a local demand trend segment; the rate of demand change refers to the speed at which demand changes over time within a local demand trend segment. Sliding window segmentation of demand time-series correlation in a multi-dimensional warehousing feature set can be performed using a sliding window algorithm. By setting an appropriate window size and sliding step, the window slides across the demand time-series correlation data, treating each window as a local demand trend segment. Peak duration can be extracted from each local demand trend segment by finding the maximum value within the segment and determining the time period containing that maximum value. The rate of demand change can be extracted by calculating the ratio of the demand difference between adjacent time points within the segment to the time interval. For example, by using sliding window segmentation to obtain multiple local demand trend segments for the demand time series data of a certain product in a warehouse, and then analyzing the peak duration and demand change rate of each segment, we can understand the demand changes of the product in different time periods.

[0037] As one implementation method, step S202 involves performing sliding window segmentation on the demand time-series correlation degree in the multi-dimensional warehouse feature set, which may specifically include the following steps S2021 to S2025:

[0038] Step S2021: Divide the demand time sequence correlation into equally spaced time sequence windows according to the order execution timestamp, with each time sequence window covering a continuous time period of a preset length.

[0039] Order execution timestamps are used to determine the temporal order of demand time-series correlation data. Dividing demand time-series correlation data into equally spaced time-series windows based on order execution timestamps is to group the data by time, facilitating subsequent analysis. An equally spaced time-series window means that each window has the same duration; the preset length is determined based on actual needs and analysis objectives. Dividing demand time-series correlation data into equally spaced time-series windows based on order execution timestamps can be achieved using a loop to distribute the demand time-series correlation data sequentially into different time-series windows according to the order of order execution timestamps. For example, demand time-series correlation data can be divided into time-series windows of one hour each, with each window covering a continuous one-hour time period.

[0040] Step S2022: Perform a stationarity test on the demand time series correlation degree within each time series window. If the test result indicates non-stationarity, perform difference processing until a stationary demand series is generated.

[0041] The stationarity test is a method to determine whether time series data is stationary. Stationarity means that the statistical characteristics of a time series do not change over time. Difference processing is a method to transform a non-stationary time series into a stationary one. It eliminates trends and seasonal variations in the time series by calculating the differences between adjacent data points. The stationarity test for demand correlation within each time window can be performed using the ADF (Augmented Dickey-Fuller) test algorithm, which determines stationarity by checking if the autoregressive coefficient of the time series is 1. If the test result indicates non-stationarity, difference processing is performed, i.e., the differences between adjacent data points are calculated to obtain a new series. Then, the stationarity test is performed again on the new series until a stationary demand series is obtained. For example, for demand correlation data containing multiple time windows, the stationarity test is performed on the data within each window. If the data within a certain window is non-stationary, difference processing is used to transform it into a stationary demand series.

[0042] Step S2023: Perform trend decomposition on the stationary demand series using an autoregressive model to separate the set of long-term trend components, periodic components, and residual components.

[0043] An autoregressive model is a statistical model used to describe time series data, predicting future values ​​based on historical data. Trend decomposition breaks down a stationary demand series into different components to better understand its patterns of change. The long-term trend component refers to the overall trend of the time series over a longer period, reflecting its long-term direction; the periodic component refers to the periodically changing part of the time series, reflecting its periodic fluctuations; and the residual component refers to the part of the time series that cannot be explained by the long-term trend and periodic components, reflecting its random fluctuations. To perform trend decomposition on a stationary demand series using an autoregressive model, an ARIMA (Autoregressive Integral Moving Average) model can be used. This model automatically identifies the autoregressive order, differencing order, and moving average order of the time series. Inputting a stationary demand series into an ARIMA model outputs a set of long-term trend components, periodic components, and residual components. For example, for a stationary demand series that has undergone differencing, using an ARIMA model for trend decomposition yields the series' long-term trend, periodic fluctuations, and random fluctuations.

[0044] Step S2024: Dynamically align the long-term trend component with the market trend forecast data to generate trend matching weights, and perform weighted correction on the periodic component based on the trend matching weights.

[0045] Dynamic alignment matches long-term trend components and market trend forecast data in terms of time and value, making them comparable. Trend matching weights are weighted values ​​generated based on the degree of matching between the long-term trend components and market trend forecast data; they measure the correlation between the two. Weighting the cyclical components based on trend matching weights makes them more consistent with market trend changes. Dynamic alignment of long-term trend components and market trend forecast data can use time alignment algorithms to match the data at the same points in time. Then, the similarity between the long-term trend components and market trend forecast data at each point in time is calculated, and trend matching weights are generated based on this similarity. Finally, the trend matching weights are multiplied by the cyclical components to obtain the weighted and corrected cyclical components. For example, for the long-term trend components and market trend forecast data of a certain commodity in a warehouse, trend matching weights are generated through dynamic alignment, and then the cyclical components of that commodity are weighted and corrected to better reflect the actual market situation.

[0046] Step S2025: Superimpose the corrected periodic component with the residual component to generate a local demand trend segment, and determine the maximum value range in the local demand trend segment as the peak duration.

[0047] Superimposing the corrected periodic component with the residual component combines their effects to obtain a more complete segment of the local demand trend. The maximum value interval refers to the time period with the highest value in the local demand trend segment; defining this as the peak duration accurately describes the duration of the demand peak. Superimposing the corrected periodic component with the residual component can be done using addition, summing their values ​​at the same time point to obtain the local demand trend segment. Then, by traversing the local demand trend segment, the maximum value is found, and the time period containing the maximum value is determined, which is the peak duration. For example, for a weighted corrected periodic component and residual component, superimposing them yields a local demand trend segment; analyzing the maximum value interval of this segment determines the duration of the demand peak.

[0048] Step S203: Dynamically correct the deviation between the peak duration and the correlation parameters in the market trend forecast data to generate the corrected local demand trend, and superimpose the demand change rate with the storage location spatial distribution density to generate demand-spatial coupling characteristics.

[0049] Dynamic deviation correction compares the peak duration with the correlation parameters in the market trend forecast data, calculates the deviation between the two, and adjusts the peak duration accordingly to better reflect actual market conditions. The corrected local demand trend, after dynamic deviation correction, better reflects the true changes in market demand. Demand-spatial coupling characteristics combine the rate of demand change with the spatial distribution density of storage locations, reflecting the relationship between demand and storage space. Dynamic deviation correction of the peak duration with the correlation parameters in the market trend forecast data calculates the difference between the two at each time point, and adjusts the peak duration based on the difference. Superimposing the rate of demand change with the spatial distribution density of storage locations, using addition or multiplication operations, combines the values ​​at the same time point to obtain the demand-spatial coupling characteristic. For example, for the peak duration of a certain commodity in a warehouse and the correlation parameters in the market trend forecast data, dynamic deviation correction yields the corrected local demand trend; simultaneously, superimposing the rate of demand change for that commodity with the spatial distribution density of storage locations generates the demand-spatial coupling characteristic.

[0050] Step S204: Use a pre-set clustering algorithm to perform pattern recognition on the demand-spatial coupling features, determine the high-frequency demand storage area and the low-frequency demand storage area, and calculate the demand-inventory matching deviation coefficient for each storage area.

[0051] Pre-defined clustering algorithms are pre-set algorithms for clustering analysis of data, which can classify data into different categories based on the similarity between data points. Pattern recognition refers to finding patterns with similar characteristics from demand-spatial coupling features, thereby identifying high-frequency and low-frequency demand storage areas. High-frequency demand storage areas refer to storage areas with frequent demand, while low-frequency demand storage areas refer to storage areas with less frequent demand. The demand-inventory matching deviation coefficient is an indicator that measures the degree of matching between demand and inventory in a storage area; it reflects whether the inventory in the storage area can meet demand. Using pre-defined clustering algorithms to perform pattern recognition on demand-spatial coupling features, the K-Means clustering algorithm can be used. This algorithm iteratively divides data points into K different categories. Taking demand-spatial coupling features as input and setting an appropriate K value, the K-Means algorithm will divide storage areas into different categories, determining high-frequency and low-frequency demand storage areas based on these categories. Calculating the demand-inventory matching deviation coefficient for each storage location area involves calculating the difference between demand and inventory in that area, then standardizing the difference to obtain the coefficient. For example, considering the demand-spatial coupling characteristics of a warehouse, K-Means clustering can be used for pattern recognition to identify high-frequency and low-frequency demand storage locations, and the demand-inventory matching deviation coefficient for each location area can be calculated.

[0052] Step S205: Map and match the high-frequency demand storage area and the low-frequency demand storage area with the capacity change sequence in the dynamic distribution characteristics of inventory to generate a set of dynamic feature vectors for input to the inventory-demand matching model.

[0053] Mapping and matching involves mapping high-frequency and low-frequency demand storage areas to capacity change sequences in the dynamic distribution characteristics of inventory, identifying the correlations between them. The dynamic feature vector set is a collection of vectors containing multiple dynamic features. It can serve as input to the inventory-demand matching model for predicting inventory adjustment and replenishment strategies. Mapping and matching high-frequency and low-frequency demand storage areas to capacity change sequences in the dynamic distribution characteristics of inventory can be done using a dictionary data structure, storing storage areas as keys and their corresponding capacity change sequences as values. Then, the capacity change sequences corresponding to the high-frequency and low-frequency demand storage areas are organized and combined to generate the dynamic feature vector set. For example, for high-frequency and low-frequency demand storage areas in a warehouse, mapping and matching them to the capacity change sequences in the dynamic distribution characteristics of inventory generates a dynamic feature vector set for input to the inventory-demand matching model.

[0054] Step S300: Based on the characteristics of warehousing demand fluctuations and preset market trend forecast data, construct an inventory-demand matching model. The inventory-demand matching model is used to output the inventory adjustment priority and replenishment trigger conditions.

[0055] Warehouse demand fluctuations reflect changes in warehousing demand over time. Pre-set market trend forecasts are data related to market trends, such as market growth rates and changes in consumer demand. An inventory-demand matching model is used to solve the matching problem between inventory and demand. It analyzes warehouse demand fluctuations and market trend forecasts to output inventory adjustment priorities and replenishment trigger conditions. Inventory adjustment priorities refer to the order and importance of inventory adjustments in different storage areas; replenishment trigger conditions refer to the conditions that trigger replenishment operations, such as inventory levels falling below a certain threshold or demand reaching a certain quantity. Building an inventory-demand matching model based on warehouse demand fluctuations and pre-set market trend forecasts can utilize regression models from machine learning, such as linear regression or neural network regression models. By using warehouse demand fluctuations and pre-set market trend forecasts as inputs, and inventory adjustment priorities and replenishment trigger conditions as outputs, the regression model is trained to obtain the inventory-demand matching model. For example, given the warehousing demand fluctuation characteristics of a warehouse and the preset market trend forecast data, a linear regression model is used to construct an inventory-demand matching model. By training the model, it can output inventory adjustment priorities and replenishment trigger conditions based on the input data.

[0056] As one implementation method, step S300 involves constructing an inventory-demand matching model based on the characteristics of warehousing demand fluctuations and preset market trend forecast data. This may specifically include the following steps S310 to S350:

[0057] Step S310: Input the final corrected warehousing demand fluctuation characteristics into the pre-trained demand prediction network to generate a demand prediction distribution for a future preset time period. The demand prediction distribution includes multiple demand time intervals and the predicted demand for each demand time interval.

[0058] A pre-trained demand forecasting network is a neural network model pre-trained to predict demand. It can predict demand over a future period based on input warehouse demand fluctuation characteristics. The future preset time period refers to the time range for demand forecasting, and the demand forecast distribution refers to the distribution of demand over time within this preset time period, including multiple demand time intervals and the corresponding predicted demand volume for each interval. The final corrected warehouse demand fluctuation characteristics are input into the pre-trained demand forecasting network. Deep learning frameworks such as TensorFlow or PyTorch can be used, treating the warehouse demand fluctuation characteristics as input data. After computation and processing by the network, the network outputs the demand forecast distribution for the future preset time period. For example, the final corrected warehouse demand fluctuation characteristics of a warehouse, when input into the pre-trained demand forecasting network, yield the demand forecast distribution for each week of the next month.

[0059] Step S320: Based on the capacity change sequence of different storage areas in the dynamic distribution characteristics of inventory, calculate the real-time inventory matching degree of each storage area. The real-time inventory matching degree is the dynamic difference ratio between the current inventory and the predicted demand.

[0060] The capacity change sequence of different storage locations in the dynamic inventory distribution characteristics reflects the change in inventory capacity of each storage location over time. Real-time inventory matching degree is an indicator that measures the degree of matching between current inventory and predicted demand. It is obtained by calculating the dynamic difference ratio between current inventory and predicted demand. To calculate the real-time inventory matching degree for each storage location based on the capacity change sequence in the dynamic inventory distribution characteristics, we can first obtain the current inventory of each storage location from the capacity change sequence, then compare it with the predicted demand for the corresponding time interval in the demand forecast distribution, calculate the difference between the two, and then divide the difference by the predicted demand to obtain the dynamic difference ratio, which is the real-time inventory matching degree. For example, for different storage locations of a warehouse, we can obtain the current inventory based on its capacity change sequence and compare it with the demand forecast distribution for the next week to calculate the real-time inventory matching degree for each storage location.

[0061] Step S330: Sort multiple demand time intervals based on the dynamic difference ratio to generate inventory adjustment priorities for each storage location area. The inventory adjustment priorities are used to indicate the replenishment order and replenishment quantity allocation weight of the storage location area.

[0062] The sorting process ranks multiple demand time intervals based on their dynamic difference ratios, prioritizing those with larger ratios and delegating them to those with smaller ratios. Inventory adjustment priority is generated based on this ranking and indicates the order in which inventory adjustments are made and the weighting of replenishment quantities for different storage areas. Ranking multiple demand time intervals based on dynamic difference ratios can be done using sorting algorithms such as bubble sort or quicksort. The process involves using the dynamic difference ratio as the basis for ranking the demand time intervals, and then generating an inventory adjustment priority for each storage area based on the ranking results. For example, for different storage areas within a warehouse, the demand time intervals are ranked according to their dynamic difference ratios to generate an inventory adjustment priority for each storage area, determining the replenishment order and the weighting of replenishment quantities.

[0063] Step S340: Match the demand time interval with the preset replenishment cycle to determine the replenishment trigger conditions for each storage location area. The replenishment trigger conditions include the replenishment start time threshold and the replenishment quantity threshold.

[0064] Time window matching compares the demand time interval with a preset replenishment cycle to find the matching relationship between the two. The replenishment trigger condition refers to the condition that triggers the replenishment operation; the replenishment start time threshold is the time point at which the replenishment operation begins; and the replenishment quantity threshold is the quantity that needs to be replenished. Time window matching between the demand time interval and the preset replenishment cycle can be performed by iterating through the demand time interval and the replenishment cycle to find the overlapping portion. Based on the overlapping portion, the replenishment start time threshold and the replenishment quantity threshold are determined. For example, for the demand time interval and the preset replenishment cycle of a warehouse, time window matching can determine the replenishment trigger condition for each storage location area, such as when to start replenishment and what quantity to replenish.

[0065] Step S350: Bind the inventory adjustment priority to the replenishment trigger condition by rules to generate an inventory-demand matching model that includes dynamic priority adjustment logic and replenishment condition judgment logic. The inventory-demand matching model is used to dynamically output replenishment strategies and storage location allocation instructions based on real-time inventory data.

[0066] Rule binding links inventory adjustment priorities and replenishment trigger conditions, establishing a rule-based relationship between them. Dynamic priority adjustment logic refers to the logic of dynamically adjusting inventory adjustment priorities based on real-time inventory data and demand changes; replenishment condition judgment logic refers to the logic of determining whether replenishment is necessary based on real-time inventory data and replenishment trigger conditions. Binding inventory adjustment priorities and replenishment trigger conditions to rules can be achieved using a rule engine. These priorities and trigger conditions are used as the conditions and actions of rules, and the rule engine's reasoning and execution generate an inventory-demand matching model. This model can dynamically adjust inventory adjustment priorities and determine replenishment conditions based on real-time inventory data, outputting replenishment strategies and storage location allocation instructions. For example, for the inventory adjustment priorities and replenishment trigger conditions of a warehouse, using a rule engine to bind rules generates an inventory-demand matching model that dynamically outputs replenishment strategies and storage location allocation instructions based on real-time inventory data.

[0067] Step S400: Based on the dynamic distribution characteristics of inventory and the priority of inventory adjustment, call the dynamic path optimization algorithm to generate a warehousing operation sequence. The warehousing operation sequence includes the storage location allocation strategy, picking route planning and replenishment execution instructions.

[0068] The dynamic distribution characteristics of inventory reflect the changes in the distribution of inventory goods within the storage space over time, while inventory adjustment priority indicates the order and importance of inventory adjustments in different storage areas. The dynamic path optimization algorithm is an algorithm for optimizing path planning; it can generate the optimal warehousing operation path based on the dynamic distribution characteristics of inventory and the inventory adjustment priority. A warehousing operation sequence is a sequence containing multiple warehousing operation steps. The storage location allocation strategy refers to how to allocate goods to different storage areas; the picking path planning refers to selecting the optimal picking path within the storage space; and the replenishment execution instruction refers to the specific instructions for executing the replenishment operation, such as the quantity, time, and path of replenishment. Based on the dynamic distribution characteristics of inventory and the inventory adjustment priority, the dynamic path optimization algorithm is invoked to generate the warehousing operation sequence. The dynamic distribution characteristics of inventory and the inventory adjustment priority can be used as inputs to the dynamic path optimization algorithm. The algorithm analyzes the storage space and inventory status to generate the storage location allocation strategy, picking path planning, and replenishment execution instructions, integrating them into the warehousing operation sequence. For example, given the dynamic distribution characteristics and inventory adjustment priorities of a warehouse's inventory, a dynamic path optimization algorithm is invoked to generate a warehouse operation sequence that includes location allocation strategies, picking route planning, and replenishment execution instructions.

[0069] As one implementation method, step S400 involves generating a warehousing operation sequence by invoking a dynamic path optimization algorithm based on the dynamic distribution characteristics of inventory and the priority of inventory adjustments. This may specifically include the following steps S410–S450:

[0070] Step S410: Based on the inventory adjustment priority, filter the target storage area to be adjusted from the inventory dynamic distribution characteristics, and obtain the real-time inventory data and spatial location data of the target storage area.

[0071] Inventory adjustment priority indicates the order and importance of inventory adjustments in different storage areas. Based on the inventory adjustment priority, target storage areas requiring adjustment are selected from the dynamic inventory distribution characteristics to determine which areas need adjustment. Real-time inventory data refers to information such as the current inventory quantity and type in the target storage area; spatial location data refers to the specific coordinates of the target storage area within the warehouse space. Selecting target storage areas based on inventory adjustment priority and dynamic inventory distribution characteristics involves iterating through the inventory adjustment priorities to identify higher-priority areas as target storage areas. Then, the real-time inventory data and spatial location data of the target storage areas are obtained from the dynamic inventory distribution characteristics. For example, given the inventory adjustment priority and dynamic inventory distribution characteristics of a warehouse, target storage areas requiring adjustment are selected based on priority, and their real-time inventory data and spatial location data are obtained.

[0072] Step S420: Match real-time inventory data with replenishment trigger conditions to determine the replenishment execution instruction for the target storage area. The replenishment execution instruction includes the replenishment quantity and the starting point of the replenishment path.

[0073] Real-time inventory data reflects the current inventory status of the target storage area. Replenishment trigger conditions refer to the conditions that trigger a replenishment operation, including the replenishment start time threshold and the replenishment quantity threshold. Matching real-time inventory data with replenishment trigger conditions involves comparing the parameters in the real-time inventory data with those parameters to determine whether replenishment is needed and the specific replenishment details. When the inventory level in the real-time inventory data is lower than the replenishment quantity threshold in the replenishment trigger conditions, a replenishment operation is required. In this case, the replenishment quantity is determined based on the difference between the real-time inventory data and the replenishment quantity threshold. The starting point of the replenishment path is usually determined based on the location of the storage area in the inventory dynamic distribution characteristics and the storage location of the replenished goods. For example, if the real-time inventory level of a target storage area is 50 units and the replenishment quantity threshold in the replenishment trigger conditions is 100 units, then the replenishment quantity is 50 units. If the commonly used replenishment storage area for this good is located in the southeast corner of the warehouse, then a specific storage location in that area can be used as the starting point of the replenishment path. Conditional statements can be used to determine the replenishment execution instruction for the target storage area. First, read the real-time inventory data and replenishment trigger conditions. Then, determine whether the real-time inventory is lower than the replenishment quantity threshold. If it is lower, calculate the replenishment quantity and determine the starting point of the replenishment path based on the dynamic distribution characteristics of the inventory. Integrate this information into a replenishment execution instruction.

[0074] Step S430: Based on spatial location data and preset picking equipment movement constraints, generate an initial picking path from the starting point of the replenishment path to the target storage location.

[0075] Spatial location data refers to the specific coordinates of the target storage area and the starting point of the replenishment route within the warehouse space. Preset picking equipment movement constraints refer to the limitations imposed on the picking equipment when moving within the warehouse space, such as aisle width, turning radius, and speed limits. An initial picking route is generated based on the spatial location data and the preset picking equipment movement constraints using a path planning algorithm, such as the A* algorithm. The A* algorithm is a heuristic search algorithm that finds the optimal path by evaluating the cost from the starting point to the target point, including the actual cost already traversed and the estimated cost to reach the target point. In this process, the spatial location data needs to be converted into map data, which includes the layout of the warehouse space and aisle information. Simultaneously, the preset picking equipment movement constraints are used as constraint parameters for the algorithm. For example, if the aisle width is limited to 2 meters, the algorithm will avoid selecting aisles with a width less than 2 meters during path planning. The A* algorithm searches the map, starting from the starting point of the replenishment route and gradually expanding the nodes until the target storage location is found, generating the initial picking route.

[0076] Step S440: Perform dynamic weight optimization on the initial picking path. The dynamic weight optimization includes multi-objective normalization calculation of path length weight, equipment energy consumption weight and time cost weight, and generates an optimized picking path plan.

[0077] The initial picking path is a preliminary path generated based on spatial location and movement constraints, but it may not be optimal. Dynamic weight optimization further optimizes the initial picking path, considering multiple objective factors, including path length, equipment energy consumption, and time cost. Path length weight refers to the degree to which the actual length of the path affects the overall path quality; equipment energy consumption weight refers to the impact of the energy consumed by the picking equipment on the path selection; and time cost weight refers to the impact of the time required to complete the path selection. Multi-objective normalization calculation unifies these three objective factors, giving them the same dimensions and range for comprehensive evaluation. For example, linear normalization can be used to normalize path length, equipment energy consumption, and time cost separately, mapping their values ​​to the [0,1] interval. Then, a weight coefficient is assigned to each objective factor, for example, a path length weight coefficient of 0.4, an equipment energy consumption weight coefficient of 0.3, and a time cost weight coefficient of 0.3. The normalized objective factor values ​​are multiplied by their corresponding weight coefficients and then summed to obtain the comprehensive evaluation value for each path. The path with the lowest overall evaluation value is selected as the optimized picking path plan.

[0078] Step S450: Configure the storage location allocation strategy as a storage location goods distribution adjustment instruction based on the optimized picking path planning, and integrate the storage location allocation strategy, picking path planning and replenishment execution instructions into a warehousing operation sequence.

[0079] The original warehouse location allocation strategy concerned how to distribute goods to different warehouse location areas. Now, it is configured as a warehouse location goods distribution adjustment instruction based on the optimized picking path plan. This means adjusting the distribution of goods in warehouse locations according to the optimized picking path to improve the efficiency of warehousing operations. For example, if the optimized picking path passes through certain warehouse location areas, the goods in those areas can be adjusted to make them easier to pick. Integrating the warehouse location allocation strategy, picking path planning, and replenishment execution instructions into a warehousing operation sequence can use data structures such as lists or dictionaries. The warehouse location allocation strategy, picking path planning, and replenishment execution instructions are each treated as elements in a list or dictionary, arranged sequentially to form the warehousing operation sequence. In this way, warehouse personnel can execute each operation sequentially according to this sequence to complete inventory adjustments and replenishment tasks.

[0080] As one implementation method, the dynamic path optimization algorithm achieves path planning through the following steps S401 to S406:

[0081] Step S401: Construct a warehouse topology network based on the spatial location data of the warehouse location area in the dynamic distribution characteristics of inventory. The nodes in the warehouse topology network correspond to the coordinates of the warehouse location area, and the edges correspond to the movable paths between warehouse location areas.

[0082] The spatial location data of storage locations in the dynamic distribution characteristics of inventory records the specific coordinates of each storage location within the storage space. A storage topology network is a graph structure used to represent the structure of a storage space, where nodes represent the coordinates of storage locations and edges represent movable paths between them. Constructing a storage topology network based on the spatial location data of storage locations can utilize graph construction algorithms from graph theory. First, the coordinates of each storage location are used as nodes in the graph. Then, based on the layout and access information of the storage space, it is determined which storage locations have movable paths between them, and these paths are used as edges in the graph. For example, if two storage locations are connected by a passageway, an edge is added between the corresponding nodes. During the construction process, weights can be added to the edges, representing information such as path length and traversal difficulty. In this way, the storage space is abstracted into a topology network, facilitating subsequent path planning algorithms.

[0083] Step S402: Configure real-time inventory status weights for nodes in the warehouse topology network according to the inventory adjustment priority. The real-time inventory status weights are proportional to the inventory adjustment priority.

[0084] Inventory adjustment priority indicates the order and importance of inventory adjustments in different storage areas. Real-time inventory status weight is a weight value assigned to nodes in the warehouse topology network, reflecting the importance of the inventory status of the corresponding storage area in path planning. Assigning real-time inventory status weights to nodes in the warehouse topology network based on inventory adjustment priority can be done using a linear mapping method. First, determine the range of inventory adjustment priority values, for example, from 1 to 10, where 1 represents the lowest priority and 10 represents the highest priority. Then, map this range to the range of real-time inventory status weight values, for example, from 0.1 to 1. For each node, determine the real-time inventory status weight according to the mapping relationship based on the inventory adjustment priority of its corresponding storage area. For example, if the inventory adjustment priority of a storage area is 8, after mapping, its corresponding real-time inventory status weight might be 0.8. Thus, in subsequent path planning, nodes corresponding to storage areas with higher inventory adjustment priorities will have higher weights in path selection and are more likely to be selected first.

[0085] Step S403: Call the path search algorithm to traverse from the starting node of the warehouse topology network and calculate the set of shortest paths to each target storage location node. Each path in the shortest path set includes the path length and equipment movement energy consumption parameters.

[0086] Pathfinding algorithms are used to find the shortest path in a graph. In this scenario, starting from the starting node of the warehouse topology network, the algorithm traverses the network to find the shortest path to each target storage area node. Common pathfinding algorithms include Dijkstra's algorithm and A* algorithm. Taking Dijkstra's algorithm as an example, it maintains a distance table recording the shortest distance from the starting node to each node. First, the distance to the starting node is set to 0, and the distances to other nodes are set to infinity. Then, the node with the smallest distance is selected from the distance table and set as the current node, updating the distances to its adjacent nodes. This process is repeated until all nodes have been traversed or the target node has been found. In calculating the shortest path, not only the path length but also the energy consumption parameters of the moving equipment must be considered. An energy consumption weight can be assigned to each edge, and the edge length and energy consumption weight are considered together when calculating the path length. For example, if a path is 10 meters long and the edge energy consumption weight is 1.2, then the total cost of this path is 10 × 1.2 = 12. The path search algorithm can be used to obtain the set of shortest paths to each target storage area node. Each path includes path length and equipment movement energy consumption parameters.

[0087] As one implementation method, the execution process of the path search algorithm may specifically include the following steps S4031 to S4036:

[0088] Step S4031: Obtain the default length weight and real-time device movement direction constraints for each path segment in the warehouse topology network.

[0089] In a warehouse topology network, the default length weight of each path segment is pre-defined, representing the basic importance of that path segment's length in path planning. Real-time equipment movement direction constraints refer to the restrictions on the movement direction of picking equipment at the current moment. For example, some aisles may only allow one-way traffic, or equipment may only be able to turn in a predetermined direction at a certain location. The default length weight and real-time equipment movement direction constraints for each path segment in the warehouse topology network can be obtained by reading relevant data from the warehouse management system. The warehouse management system records basic information for each path segment, including the default length weight, and also monitors the movement direction of equipment in real time to obtain real-time equipment movement direction constraints. For example, in a large warehouse, different aisles have different lengths and passage rules. The warehouse management system sets a default length weight for each aisle and provides equipment movement direction constraints based on the real-time location and status of the equipment.

[0090] Step S4032: Calculate the inventory urgency of each target storage location node based on the real-time inventory status weight. The inventory urgency is related to the replenishment order in the inventory adjustment priority.

[0091] Real-time inventory status weights reflect the importance of the inventory status of the corresponding storage area in path planning. Inventory urgency is an indicator that measures the urgency of inventory demand in the target storage area, and it is related to the replenishment order in inventory adjustment priority. The inventory urgency of each target storage area node is calculated based on its real-time inventory status weight using a linear function. The real-time inventory status weight is used as input, and the inventory urgency is calculated through a predefined linear function. For example, inventory urgency = real-time inventory status weight × a constant coefficient + an offset. This constant coefficient and offset can be adjusted according to actual conditions to match the inventory urgency with the replenishment order in inventory adjustment priority. Storage areas with higher inventory adjustment priority will also have higher inventory urgency. Thus, in path planning, storage areas with higher inventory urgency will receive more attention and are more likely to be accessed first.

[0092] Step S4033: Map the inventory urgency to the priority weight of the path segment, and linearly superimpose it with the default length weight to generate a dynamically adjusted path segment weight.

[0093] Inventory urgency is an indicator of the urgency of inventory demand in a target storage area. Mapping it to priority weights for path segments transfers this information to the path segments, giving higher priority to path segments associated with storage areas of high urgency. The mapping process can use a simple linear mapping function to map the inventory urgency value to a range of path segment priority weights. For example, mapping an inventory urgency range of 0 to 1 to a path segment priority weight range of 0.1 to 1. Then, the mapped path segment priority weights are linearly superimposed with the default length weight, i.e., the dynamically adjusted path segment weight = path segment priority weight × one weight coefficient + default length weight × another weight coefficient. These two weight coefficients can be adjusted according to actual conditions to balance the impact of inventory urgency and path length in path planning. In this way, dynamically adjusted path segment weights are generated, making path planning more rational and prioritizing storage areas with high inventory urgency.

[0094] Step S4034: Update the edge weights of the warehouse topology network based on the dynamically adjusted path segment weights, and recalculate the shortest path from the starting node to each target storage area node.

[0095] The dynamically adjusted path segment weights reflect the importance of path segments after considering inventory urgency and path length. Updating the edge weights of the warehouse topology network based on these dynamically adjusted weights involves assigning the weights to the corresponding edges in the network. Then, a path search algorithm, such as Dijkstra's algorithm or A* algorithm, is used to recalculate the shortest paths from the starting node to each target storage area node. Because the edge weights have changed, the recalculated shortest paths may differ from the previous ones. For example, previously, only path length was considered, leading to storage areas with lower inventory urgency. After updating the edge weights, a slightly longer path leading to storage areas with higher inventory urgency might be chosen to meet inventory adjustment needs. By updating edge weights and recalculating the shortest paths, the path planning becomes more aligned with the actual inventory status and requirements.

[0096] Step S4035: If the load of a path node is detected to exceed a preset threshold, a temporary path detour mark is made for the high-load node in the warehouse topology network, and a set of alternative paths to bypass the high-load node is generated.

[0097] Path node load refers to a comprehensive indicator of factors such as equipment traffic and cargo storage volume at a particular node. A preset threshold is a pre-defined limit; when the path node load exceeds this threshold, it indicates potential congestion or other problems at that node, affecting path efficiency. Temporarily marking high-load nodes in the warehouse topology network involves identifying these nodes and prompting the path planning algorithm to avoid them in subsequent planning. Generating a set of alternative paths to bypass high-load nodes can be done using path search algorithms from graph theory. During the search, high-load nodes are excluded from accessibility, and alternative paths to each target storage area node are searched starting from the originating node. For example, depth-first search or breadth-first search algorithms can be used to traverse the warehouse topology network and generate a set of alternative paths to bypass high-load nodes. This way, when a node experiences excessive load, alternative paths can be selected, ensuring smooth warehousing operations.

[0098] Step S4036: Compare the cost of the alternative path set with the shortest path, and select the path that satisfies the inventory urgency and has the lowest energy consumption for equipment movement as the picking path plan after conflict resolution.

[0099] The alternative path set is a set of paths that bypass high-load nodes, while the shortest path is the shortest path calculated earlier from the starting node to each target storage area node. Comparing the cost of the alternative path set with the shortest path requires considering multiple factors, including path length, equipment movement energy consumption, and inventory urgency. For each path, its comprehensive cost is calculated, which can be a weighted sum of path length, equipment movement energy consumption, and inventory urgency. For example, comprehensive cost = path length × length weight + equipment movement energy consumption × energy consumption weight + inventory urgency × urgency weight. The length weight, energy consumption weight, and urgency weight can be adjusted according to actual conditions. Then, the comprehensive costs of the alternative path set and the shortest path are compared, and the path that satisfies inventory urgency and has the lowest equipment movement energy consumption is selected as the picking route plan after conflict resolution. This ensures that inventory adjustment needs are met while reducing equipment energy consumption and improving warehousing efficiency.

[0100] Step S404: Detect whether there are path conflict conditions in the shortest path set. Path conflict conditions include conflict in the device movement direction or the load of the path node exceeds a preset threshold.

[0101] The shortest path set is the set of shortest paths from the starting node to each target storage area node, calculated using a path search algorithm. Path conflict conditions include conflicts in equipment movement directions and path node loads exceeding preset thresholds. Detecting the existence of path conflict conditions in the shortest path set can be done using conditional judgment methods. For equipment movement direction conflicts, it can be checked whether there are contradictions in the equipment movement directions of different paths in the shortest path set, such as two paths moving in opposite directions at a certain node. For path node loads exceeding preset thresholds, the load status of path nodes can be monitored in real time and compared with the preset threshold. For example, in a warehouse management system, sensors can be set up to monitor the equipment flow and goods storage volume at nodes; when the load of a node exceeds the preset threshold, a path conflict condition is determined to exist. By detecting path conflict conditions, problems that may affect path passage can be identified in a timely manner, providing a basis for subsequent conflict resolution.

[0102] Step S405: If path conflict conditions exist, a multi-objective optimization algorithm is used to resolve the conflict in the shortest path set to generate an optimized path set. Each path in the optimized path set includes the movement direction adjustment parameters and energy consumption balance coefficient after conflict resolution.

[0103] Multi-objective optimization algorithms are used to solve problems involving conflicting objectives. In this scenario, they are used to resolve conflicts among sets of shortest paths. When a path conflict is detected, a multi-objective optimization algorithm, such as a genetic algorithm or simulated annealing, is employed. Taking a genetic algorithm as an example, the set of shortest paths is first used as the initial population, with each path representing an individual. Then, a fitness function is defined, which considers multiple objectives, such as path length, equipment movement energy consumption, and conflict resolution. Through genetic operations such as selection, crossover, and mutation, the population is continuously evolved until an optimal set of paths that meets the requirements is found. During conflict resolution, the movement direction of the paths needs to be adjusted to avoid conflicts in equipment movement directions. Simultaneously, energy balance must be considered to ensure a more reasonable energy distribution throughout the operation. Each path in the optimized path set includes movement direction adjustment parameters after conflict resolution and an energy balance coefficient. These parameters guide the actual operation of the picking equipment, ensuring smooth path passage and efficient equipment operation.

[0104] Step S406: Select the path with the lowest overall path cost from the optimized path set as the picking path plan. The overall path cost is determined by the path length, equipment movement energy consumption parameters and inventory status weight.

[0105] The optimized path set is the set of paths obtained after conflict resolution using a multi-objective optimization algorithm. The comprehensive path cost is determined by jointly calculating the path length, equipment movement energy consumption parameters, and inventory status weights, comprehensively considering the importance of path length, equipment energy consumption, and inventory status. The path with the lowest comprehensive path cost from the optimized path set is selected as the picking route plan. This can be done by traversing each path in the optimized path set and calculating the comprehensive path cost for each path according to a predefined formula. For example, comprehensive path cost = path length × length weight + equipment movement energy consumption parameter × energy consumption weight + inventory status weight × status weight. The length weight, energy consumption weight, and status weight can be adjusted according to actual conditions. Then, the comprehensive path costs of each path are compared, and the path with the lowest cost is selected as the picking route plan. This minimizes the comprehensive cost of the path while meeting inventory adjustment requirements, improving the efficiency and economy of warehousing operations.

[0106] Step S500: Based on the replenishment triggering conditions, perform collaborative scheduling processing on the warehousing operation sequence to generate an optimized warehousing operation instruction set, and feed the warehousing operation instruction set back to the warehousing control system to execute inventory optimization operations.

[0107] Replenishment trigger conditions refer to the conditions that trigger a replenishment operation, including the replenishment start time threshold and the replenishment quantity threshold. Coordinated scheduling of the warehousing operation sequence involves coordinating and arranging the various operations within the sequence based on the replenishment trigger conditions to ensure smooth operation. Generating an optimized warehousing operation instruction set requires considering multiple factors, such as the order of operations, equipment availability, and inventory status. Scheduling algorithms, such as heuristic scheduling algorithms or rule-based scheduling algorithms, can be used. Taking heuristic scheduling algorithms as an example, they adjust the warehousing operation sequence by defining heuristic rules, such as prioritizing urgent replenishment tasks and rationally arranging the operation sequence of equipment. During the adjustment process, the execution time and order of each operation are determined based on the replenishment trigger conditions, while also considering equipment movement paths and operation times to avoid conflicts and delays. The generated optimized warehousing operation instruction set contains detailed operation instructions, such as location allocation, picking routes, replenishment quantity, and time. The warehouse operation instruction set is fed back to the warehouse control system. The warehouse control system will control picking equipment and other warehouse facilities according to these instructions to perform inventory optimization operations, such as replenishment and storage location adjustment, in order to meet warehousing needs and improve inventory management efficiency.

[0108] As one implementation method, step S500 involves coordinating and scheduling the warehousing operation sequence based on replenishment trigger conditions to generate an optimized warehousing operation instruction set. This may specifically include the following steps S510–S550:

[0109] Step S510: Based on the replenishment strategy and storage location allocation instructions output by the inventory-demand matching model, detect the path conflict conditions between the storage location allocation strategy and the replenishment execution instructions in the warehousing operation sequence. The path conflict conditions include the spatial overlap rate between the picking path planning and the replenishment path in the same storage location area and the conflict indicator of the equipment movement direction.

[0110] The replenishment strategy and location allocation instructions output by the inventory-demand matching model are generated based on warehousing demand and inventory status, guiding warehousing operations. Detecting path conflicts between the location allocation strategy and replenishment execution instructions in the warehousing operation sequence requires detailed analysis of the picking route planning in the location allocation strategy and the replenishment route in the replenishment execution instructions. The spatial overlap rate between the picking route planning and replenishment route in the same location area refers to the proportion of the overlapping portion of the two routes within the same location area to the total path length. Excessive overlap can lead to equipment congestion and low operational efficiency. Equipment movement direction conflict indicators indicate whether the equipment movement directions of two paths are opposite at a certain node or segment. If they are opposite, it will lead to equipment collisions and conflicts. Detecting path conflict conditions can be done using path analysis algorithms to simulate the picking route planning and replenishment route in the warehousing topology network, calculating the spatial overlap rate and determining whether equipment movement directions conflict. For example, by comparing the coordinates and direction information of the two paths at each node and segment, it can be determined whether path conflict conditions exist.

[0111] Step S520: If a path conflict condition is detected, the replenishment quantity threshold of the target storage area is recalculated based on the real-time inventory matching degree in the dynamic inventory distribution characteristics, and an adjusted replenishment execution instruction is generated based on the replenishment quantity threshold.

[0112] When path conflicts are detected, replenishment execution instructions need to be adjusted to avoid them. The replenishment threshold for the target storage area is recalculated based on the real-time inventory matching degree in the dynamic inventory distribution characteristics. The real-time inventory matching degree is the dynamic difference ratio between the current inventory level and the predicted demand, reflecting the degree of matching between inventory and demand. A function can be used, taking the real-time inventory matching degree as input, to calculate the new replenishment threshold. For example, a high real-time inventory matching degree indicates relatively sufficient inventory, and the replenishment threshold can be appropriately lowered; a low real-time inventory matching degree indicates relatively insufficient inventory, and the replenishment threshold needs to be appropriately increased. Adjusted replenishment execution instructions are generated based on the new replenishment threshold, including updated replenishment quantity and possible replenishment path origin information. For example, if the replenishment threshold is lowered, the replenishment quantity in the replenishment execution instruction will also be reduced accordingly. By recalculating the replenishment threshold and generating adjusted replenishment execution instructions, path conflicts can be avoided while ensuring that inventory meets demand.

[0113] Step S530: Perform dynamic weight matching between the starting point of the replenishment path in the adjusted replenishment execution instruction and the current position of the equipment in the picking path planning to generate a path node sequence after conflict resolution. The dynamic weight matching includes a weighted calculation of the movement cost between path nodes and the urgency of inventory.

[0114] The replenishment path start point in the adjusted replenishment execution instruction is the newly determined replenishment starting position, and the current equipment position in the picking path planning is the current position of the picking equipment. Dynamically weighting the replenishment path start point and the current equipment position requires considering the movement costs between path nodes and inventory urgency. Movement costs between path nodes refer to the cost required for equipment to move from one node to another, including path length, equipment energy consumption, etc. Inventory urgency is an indicator of the urgency of inventory demand in the target storage area. Dynamic weighting involves a weighted calculation of movement costs between path nodes and inventory urgency. For example, the movement costs between path nodes can be multiplied by a cost weight, and the inventory urgency can be multiplied by an urgency weight, then the two can be added together to obtain a comprehensive weight. By traversing all possible path node sequences, calculating the comprehensive weight of each sequence, and selecting the sequence with the smallest comprehensive weight as the conflict-resolved path node sequence, this reduces equipment movement costs and improves operational efficiency while ensuring that inventory urgency is met.

[0115] As one implementation method, the process of generating the path node sequence after conflict resolution may specifically include the following steps S531 to S535:

[0116] Step S531: Obtain the upper limit of the inventory capacity and the current inventory quantity of the target storage area, and calculate the remaining capacity difference between the replenishment quantity in the adjusted replenishment execution instruction and the upper limit of the inventory capacity.

[0117] The maximum inventory capacity of the target storage area is the maximum quantity of goods that the area can hold, while the current inventory level is the actual quantity of goods currently stored in that area. The replenishment quantity in the adjusted replenishment order is the quantity of goods that needs to be replenished after the adjustment. To calculate the remaining capacity difference between the replenishment quantity in the adjusted replenishment order and the maximum inventory capacity, simply subtract the sum of the current inventory and the replenishment quantity from the maximum inventory capacity. For example, if the maximum inventory capacity of the target storage area is 200 units, the current inventory is 100 units, and the adjusted replenishment quantity is 50 units, then the remaining capacity difference is 200 - (100 + 50) = 50 units. By calculating the remaining capacity difference, we can understand the remaining storage capacity of the storage area, providing a reference for subsequent replenishment operations.

[0118] Step S532: If the remaining capacity difference is less than the preset safety buffer, the replenishment threshold is dynamically expanded according to the capacity change sequence in the inventory dynamic distribution characteristics, and a replenishment execution instruction after secondary adjustment is generated.

[0119] The preset safety buffer is a pre-defined capacity value used to ensure that the storage area still has safe storage space after replenishment. When the remaining capacity difference is less than the preset safety buffer, it indicates that the storage area may not be able to accommodate the adjusted replenishment quantity, and the replenishment quantity threshold needs to be expanded. Dynamically expanding the replenishment quantity threshold based on the capacity change sequence in the dynamic distribution characteristics of inventory allows analysis of the trend and pattern of the capacity change sequence, predicting the capacity changes of the storage area in the future. For example, if the capacity change sequence shows that the capacity of the storage area is likely to increase in the future, the replenishment quantity threshold can be appropriately expanded. A second-adjusted replenishment execution instruction is generated, updating the replenishment quantity and possible replenishment path starting points based on the expanded replenishment quantity threshold. By dynamically expanding the replenishment quantity threshold and generating a second-adjusted replenishment execution instruction, congestion and safety issues caused by excessive replenishment in the storage area can be avoided.

[0120] Step S533: Calculate the path cost between the starting point of the replenishment path in the second-adjusted replenishment execution instruction and the current position of the equipment. The path cost calculation includes multi-objective normalization processing of equipment turning times, path segment length, and inventory matching degree.

[0121] The replenishment path starting point in the second-adjusted replenishment execution instruction is the replenishment starting position determined after the second adjustment, and the current equipment position is the position of the picking equipment at the current moment. Path cost calculation includes multi-objective normalization processing of equipment turning times, path segment length, and inventory matching degree. Equipment turning times refer to the number of times the equipment needs to turn during its movement from its current position to the replenishment path starting point; too many turns increase equipment energy consumption and operation time. Path segment length refers to the length of each segment into which the path is divided; the longer the path segment length, the higher the equipment movement cost. Inventory matching degree is the degree of matching between the current inventory and the predicted demand; the higher the inventory matching degree, the better the inventory can meet the demand. Multi-objective normalization processing unifies the equipment turning times, path segment length, and inventory matching degree, giving them the same dimensions and range. For example, linear normalization methods can be used to normalize the equipment turning times, path segment length, and inventory matching degree separately, mapping their values ​​to the [0,1] interval. Then, a weight coefficient is assigned to each target factor. The normalized target factor value is multiplied by the corresponding weight coefficient and then summed to obtain the path cost. By calculating the path cost, the path with the minimum path cost can be selected, thereby improving the efficiency of equipment movement and the effectiveness of inventory management.

[0122] Step S534: Prioritize the path node sequence according to the normalized path cost, generate a candidate path node set, and select the node sequence with the lowest comprehensive cost from the candidate path node set as the path node sequence after conflict resolution.

[0123] The normalized path cost is obtained after multi-objective normalization, comprehensively considering factors such as the number of equipment turns, path segment length, and inventory matching degree. The path node sequence is prioritized based on the normalized path cost using sorting algorithms such as bubble sort or quicksort. The path node sequences are then sorted in ascending order of normalized path cost to generate a candidate path node set. Finally, the node sequence with the lowest overall cost is selected from the candidate set as the conflict-resolved path node sequence. This ensures that the selected path node sequence has the lowest overall cost while meeting inventory requirements, improving the efficiency and economy of warehousing operations.

[0124] Step S535: Align the conflict-resolved path node sequence with the replenishment triggering conditions in the inventory-demand matching model within a time window to generate the final optimized warehouse operation instruction set.

[0125] The conflict-resolved path node sequence is the optimal sequence obtained after conflict resolution and path cost calculation. The replenishment triggering conditions in the inventory-demand matching model include a replenishment start time threshold and a replenishment quantity threshold. Aligning the conflict-resolved path node sequence with the replenishment triggering conditions within a time window means matching the execution time of the path node sequence with the time threshold in the replenishment triggering conditions, ensuring that the replenishment operation is performed at the appropriate time. For example, if the replenishment start time threshold in the replenishment triggering conditions is a specific moment, then the execution time of the path node sequence needs to be adjusted so that it starts execution at that moment. Through time window alignment, a final optimized warehouse operation instruction set is generated. This instruction set contains detailed operation instructions, such as the path node sequence, execution time, and replenishment quantity, which can guide warehouse equipment to efficiently complete inventory optimization operations.

[0126] Step S540: Perform dynamic remapping processing on the picking path plan based on the path node sequence to generate an updated picking path plan, and then integrate the updated picking path plan with the storage location allocation strategy.

[0127] The path node sequence is the optimal sequence of path nodes after conflict resolution. Dynamic remapping of the picking route plan involves adjusting and updating the picking route plan based on the path node sequence. For example, if the path node sequence contains new nodes or path segments, this information needs to be integrated into the picking route plan. This generates an updated picking route plan that better matches actual operational needs and path conflict resolution. Integrating the updated picking route plan with the storage location allocation strategy involves combining the path information from the picking route plan with the storage location information from the storage location allocation strategy to form a unified operational instruction. For example, the instruction explicitly specifies which storage location the equipment should pick from and along which path to move to the target storage location. Through dynamic remapping and instruction integration, the coordination and efficiency of warehousing operations can be improved, ensuring the smooth operation of picking and storage of goods.

[0128] Step S550: Based on the replenishment triggering conditions in the inventory-demand matching model, verify the timeliness of the warehousing operation sequence after instruction fusion, and generate an optimized warehousing operation instruction set that meets the conditions of inventory capacity constraints and path non-conflict.

[0129] The replenishment trigger conditions in the inventory-demand matching model include replenishment start time thresholds and replenishment quantity thresholds. Timeliness verification of the warehousing operation sequence after instruction fusion involves checking whether each operation in the sequence is performed within the time range specified by the replenishment trigger conditions, and whether it meets inventory capacity constraints and path conflict-free conditions. For example, it checks whether the replenishment operation starts after the replenishment start time threshold, whether the replenishment quantity does not exceed the inventory capacity limit, and whether there are no path conflicts. If some operations in the sequence do not meet these conditions, the sequence needs to be adjusted and optimized. Through timeliness verification, an optimized warehousing operation instruction set that meets inventory capacity constraints and path conflict-free conditions is generated. This instruction set ensures the smooth operation of warehousing and improves the efficiency and accuracy of inventory management.

[0130] As one implementation, after step S500, where the warehousing operation instruction set is fed back to the warehousing control system to perform inventory optimization operations, the method provided in this embodiment of the invention may further include the following derivative steps S600 to S1000:

[0131] Step S600: Collect inventory status update data and equipment operation logs in real time after the warehouse control system executes the optimized warehouse operation instruction set, and generate an execution feedback dataset. The execution feedback dataset includes the actual inventory quantity of the storage location, the execution time of the picking path, and the completion status of the replenishment operation.

[0132] After a warehouse control system executes an optimized set of warehouse operation instructions, it impacts inventory status and equipment operation. Real-time collection of inventory status updates and equipment operation logs is crucial. Inventory status updates include changes in the actual inventory level in each storage location area, while equipment operation logs record detailed information about the equipment during instruction execution, such as the execution time of picking routes and the completion status of replenishment operations. An execution feedback dataset is generated, summarizing and organizing the inventory status updates and equipment operation logs. Data acquisition devices and sensors can be used to acquire real-time inventory status updates; for example, inventory sensors can monitor the quantity of goods in storage locations. Equipment operation logs can be obtained through the equipment's own recording system. For instance, a data acquisition module can be set up in the warehouse control system to periodically collect inventory status updates and equipment operation logs, storing this data in a database to form an execution feedback dataset. By collecting this dataset, the effectiveness of the warehouse operation instruction set can be understood, providing a basis for subsequent model optimization and strategy adjustments.

[0133] Step S700: Compare the actual inventory quantity of the storage location in the execution feedback dataset with the predicted demand quantity in the inventory-demand matching model to generate inventory optimization deviation parameters. The inventory optimization deviation parameters include the inventory quantity difference ratio and the demand response delay time.

[0134] The actual inventory level in the execution feedback dataset represents the actual quantity of goods stored in each storage location area after the warehouse control system executes the optimized warehousing operation instruction set. Conversely, the predicted demand in the inventory-demand matching model is the quantity of goods that each storage location area should have within a set time period, predicted by the model based on historical data and market trends. Comparing the discrepancies between the two provides a clear indication of the gap between the actual effect and the expected results of inventory optimization operations.

[0135] Generate inventory optimization deviation parameters, where the inventory difference ratio is obtained by calculating the ratio of the difference between the actual inventory level and the predicted demand at the storage location to the predicted demand. The specific calculation method is as follows: first, calculate the difference between the two; if the predicted demand is Q... 预测 The actual inventory is Q. 实际 The difference is |Q 预测 -Q 实际 Then divide the difference by the predicted demand Q. 预测 ,Right now This ratio reflects the degree of deviation between inventory quantity and forecast; the larger the ratio, the greater the gap between actual inventory and forecasted demand.

[0136] Demand response delay refers to the time interval between the predicted demand and the actual inventory level meeting that demand. For example, if an inventory-demand matching model predicts that a certain storage area needs to reach a set inventory level on Monday to meet demand, but the storage area doesn't reach the corresponding inventory level until Wednesday, then the demand response delay is two days. By analyzing the time information in the execution feedback dataset, the predicted demand time and the actual inventory level meeting the demand are determined; subtracting these two values ​​yields the demand response delay.

[0137] For example, in an electronics warehouse, an inventory-demand matching model predicts a demand of 500 units for a certain model of mobile phone in a specific storage area over a weekend. However, after executing optimized warehousing operation instructions, the actual inventory in that area is only 450 units. The inventory discrepancy is 10%. If the predicted demand is for Friday, and the actual inventory reaches the required level by Sunday, the demand response delay is two days. Through such deviation comparisons and parameter generation, the differences between the inventory optimization operation and the expected quantity and timing can be clearly understood.

[0138] Step S800: Dynamically correct the warehousing demand fluctuation characteristics based on the inventory optimization deviation parameters to generate updated warehousing demand fluctuation characteristics, and input the updated warehousing demand fluctuation characteristics into the inventory-demand matching model for iterative optimization of model parameters.

[0139] Inventory optimization deviation parameters reflect the difference between actual inventory levels and model predictions. Dynamically correcting warehousing demand fluctuations based on these parameters can make these fluctuations more consistent with actual warehousing demand. This dynamic correction can employ a weighted correction method, assigning different weights based on the proportion of inventory discrepancies and the duration of demand response delays. For example, a higher weight can be given to cases with larger inventory discrepancies to more significantly adjust warehousing demand fluctuations.

[0140] Specifically, the amplitude of demand fluctuations can be adjusted based on the inventory level difference ratio. If the inventory level difference ratio is large and the actual inventory level is less than the predicted demand level, it indicates that demand may be underestimated, and the peak demand in the warehousing demand fluctuation characteristics can be appropriately increased. Conversely, if the actual inventory level is greater than the predicted demand level, it indicates that demand may be overestimated, and the peak demand level can be appropriately decreased. Regarding the demand response delay, the time distribution of demand fluctuations can be adjusted. If the demand response delay is long, it indicates that the timeliness of demand is not being well met, and the demand time in the demand fluctuation characteristics can be advanced to improve the timeliness of demand response.

[0141] After generating updated warehouse demand fluctuation characteristics, these are input into the inventory-demand matching model for iterative parameter optimization. Inventory-demand matching models are typically built using machine learning algorithms, such as neural network models. During iterative optimization, the updated warehouse demand fluctuation characteristics are used as input, and the model's outputs, such as inventory adjustment priorities and replenishment trigger conditions, are compared with actual execution feedback data to calculate the loss function. A common loss function is the mean squared error (MSE), which calculates the average of the squared errors between the model's output and the actual value. By continuously adjusting the model's parameters, such as the weights and biases in the neural network, the value of the loss function gradually decreases, thereby achieving iterative optimization of the model parameters. This allows the model to more accurately predict inventory demand and formulate reasonable replenishment strategies.

[0142] For example, in a clothing warehouse, deviation comparison reveals a significant discrepancy in the inventory of a certain type of summer clothing, coupled with delayed demand response. Actual inventory is far below predicted demand, and replenishment is not timely. To dynamically correct the warehousing demand fluctuation characteristics, the peak demand for this type of clothing in summer is increased, and the demand timing is appropriately advanced. The updated warehousing demand fluctuation characteristics are then input into the inventory-demand matching model. Through multiple iterations to optimize the model parameters, the model can subsequently more accurately predict the demand for this type of clothing and schedule replenishment.

[0143] Step S900: Recalculate the inventory adjustment priority and replenishment triggering conditions based on the iteratively optimized inventory-demand matching model, and generate the adjusted inventory-demand matching strategy.

[0144] After the parameter update in step S800, the iteratively optimized inventory-demand matching model can more accurately reflect the actual warehousing demand and inventory status. Recalculating inventory adjustment priorities and replenishment trigger conditions based on this optimized model is to make inventory management strategies more aligned with actual needs, thereby improving the efficiency and accuracy of inventory management.

[0145] When recalculating inventory adjustment priorities, the model comprehensively considers factors such as updated warehousing demand fluctuation characteristics and dynamic inventory distribution characteristics. For example, for storage areas with large demand fluctuations and frequent inventory changes, their inventory adjustment priority may be increased accordingly. The model will reassess and rank the importance of each storage area based on demand forecast distribution and real-time inventory matching degree, and determine the new replenishment order and replenishment quantity allocation weights.

[0146] The recalculation of replenishment trigger conditions is also based on the optimized model. The model combines the demand time interval and the preset replenishment cycle to more accurately determine the replenishment initiation time threshold and replenishment quantity threshold for each storage location. For example, if demand forecasts for a certain storage location indicate a significant increase in demand in the near future, and the dynamic inventory distribution characteristics suggest that inventory in that area is being depleted quickly, the model will appropriately lower the replenishment initiation time threshold, i.e., trigger the replenishment operation earlier, while simultaneously increasing the replenishment quantity threshold to meet future demand growth.

[0147] An adjusted inventory-demand matching strategy is generated, integrating the recalculated inventory adjustment priorities and replenishment trigger conditions. This strategy clarifies the inventory adjustment and replenishment arrangements for each storage location area at different times, providing more scientific and reasonable guidance for subsequent warehousing operations. For example, in a food warehouse, the iteratively optimized model discovered that the demand for a certain type of snack increases significantly on weekends, and the inventory depletion rate in a certain storage location area accelerates. Therefore, after recalculation, the inventory adjustment priority of that storage location area is increased, replenishment operations are triggered earlier, and the replenishment volume is increased, forming an inventory-demand matching strategy more suitable for the actual situation.

[0148] Step S1000: Call the dynamic path optimization algorithm to replan the path of the warehousing operation sequence based on the adjusted inventory-demand matching strategy, generate a secondary optimized warehousing operation instruction set, and feed the secondary optimized warehousing operation instruction set back to the warehousing control system to trigger dynamic inventory adjustment operation.

[0149] The dynamic path optimization algorithm has already been used to generate the warehousing operation sequence in the previous steps. Now, based on the adjusted inventory-demand matching strategy, the path replanning of the warehousing operation sequence is performed to further optimize the path arrangement of warehousing operations, improve operational efficiency and reduce costs.

[0150] The revised inventory-demand matching strategy includes new inventory adjustment priorities and replenishment trigger conditions, which will affect the routing planning of warehouse operations. For example, changes in inventory adjustment priorities may lead to changes in the storage locations that need to be prioritized, and adjustments to replenishment trigger conditions may affect the timing and quantity of replenishment, thereby affecting the selection of replenishment routes.

[0151] When using the dynamic path optimization algorithm for path replanning, the algorithm reconstructs the warehouse topology network based on the adjusted strategy, assigning new real-time inventory status weights to nodes. For example, if the inventory adjustment priority of a certain storage area is increased, the real-time inventory status weight of that node will also increase accordingly. Then, the algorithm recalculates the set of shortest paths from the starting node to each target storage area node, considering factors such as path length, equipment movement energy consumption parameters, and inventory status weights. During the calculation process, path conflict conditions are also detected, such as equipment movement direction conflicts or path node loads exceeding preset thresholds, and multi-objective optimization algorithms are used to resolve these conflicts.

[0152] A second-optimized set of warehouse operation instructions is generated, which includes redesigned location allocation strategies, picking route planning, and replenishment execution instructions. This optimized set of instructions is then fed back to the warehouse control system, which triggers dynamic inventory adjustments based on these instructions. For example, it controls picking equipment to pick goods according to the new picking routes and performs replenishment based on the new replenishment execution instructions. This dynamic adjustment allows warehouse operations to better adapt to actual inventory needs and status changes, improving the overall efficiency of warehouse management.

[0153] It is understood that the various algorithms involved in the above descriptions of the embodiments of the present invention, such as the cosine distance algorithm and correlation analysis algorithms, such as the Pearson correlation coefficient algorithm, can all be obtained from relevant content in the prior art. To save space, they will not be elaborated on in the embodiments of the present invention. In addition, those skilled in the art can supplement the details based on common knowledge in the art when implementing the solution of the present invention. For example, they can use normalization to eliminate dimensional conflicts before feature fusion, use interpolation to eliminate dimensional differences, reasonably set thresholds based on historical data, experience or business scenario requirements, train the model based on a general model training method, set the number of layers in the model structure based on actual needs, select activation functions, etc. The present invention will not provide redundant descriptions of overly detailed implementation processes.

[0154] Please see Figure 2 , Figure 2This is a schematic diagram of a computer system provided in an embodiment of the present invention. The computer system includes at least a processor 101, a communication interface 102, and a memory 103. The processor 101, communication interface 102, and memory 103 can be connected via a bus or other means. The processor 101 (or Central Processing Unit, CPU) is the computing and control core of the computer system, capable of parsing various instructions and processing various data within the computer system. The communication interface 102 may optionally include a standard wired interface or a wireless interface (such as Wi-Fi, mobile communication interface, etc.), and can be used to send and receive data under the control of the processor 101; the communication interface 102 can also be used for data transmission and interaction within the computer system. The memory 103 is a storage device in the computer system used to store programs and data. It is understood that the memory 103 here can include the computer system's built-in memory, or it can include extended memory supported by the computer system. The memory 103 provides storage space, which stores the computer system's operating system; this invention does not limit this storage space. In one embodiment, the processor 101 executes the intelligent warehouse demand forecasting and scheduling method combined with AIAgent provided above by running a computer program in the memory 103.

Claims

1. A method for intelligent warehouse demand forecasting and scheduling combined with AI Agent, characterized in that, Includes the following steps: Obtain a set of historical warehousing data for the target warehousing scenario. The set of historical warehousing data includes multiple historical warehousing operation sequences. Each historical warehousing operation sequence consists of at least one inbound operation record, one outbound operation record, and one inventory status update record. The historical warehousing data set is processed by AIAgent to extract demand features and generate warehousing demand fluctuation features and inventory dynamic distribution features. Based on the warehousing demand fluctuation characteristics and the preset market trend forecast data, an inventory-demand matching model is constructed. The inventory-demand matching model is used to output the inventory adjustment priority and replenishment trigger conditions. Based on the dynamic distribution characteristics of the inventory and the priority of inventory adjustment, a dynamic path optimization algorithm is invoked to generate a warehousing operation sequence, which includes a storage location allocation strategy, picking route planning, and replenishment execution instructions. Based on the replenishment triggering conditions, the warehousing operation sequence is coordinated and scheduled to generate an optimized warehousing operation instruction set, which is then fed back to the warehousing control system to perform inventory optimization operations.

2. The method according to claim 1, characterized in that, The historical warehousing data set includes multiple warehousing operation records. Each warehousing operation record contains an order execution timestamp, goods category identifier, inventory capacity change, and related parameters from the corresponding market trend forecast data. The process of extracting demand features from the historical warehouse data set using AIAgent includes: Based on the order execution timestamp, the inbound and outbound operation records are time-series aligned to generate a time-series warehousing feature sequence associated with the inventory status update record. The time-series warehousing feature sequence includes the total inventory change rate and goods turnover rate corresponding to each timestamp. Based on the cargo category identifier, the time-series warehousing feature sequence is classified and aggregated to generate category demand sequences for different cargo categories. The category demand sequences are used to characterize the historical demand change trend of the same cargo category. Periodic demand patterns and sudden demand offsets are extracted from the category demand sequence. The periodic demand patterns are identified by a sliding time window algorithm to identify recurring peak demand intervals. The sudden demand offsets are obtained by calculating the dynamic deviation between the category demand sequence and the correlation parameters in the market trend forecast data. The changes in inventory capacity are aggregated by storage location area to generate a capacity change sequence for each storage location area. Spatial clustering is then performed on the capacity change sequences to determine the dynamic distribution characteristics of inventory in different storage locations. The cyclical demand pattern is overlaid with the market trend forecast data to generate an overlaid warehouse demand fluctuation feature. At the same time, the warehouse demand fluctuation feature is corrected according to the sudden demand offset to generate a final corrected warehouse demand fluctuation feature.

3. The method according to claim 2, characterized in that, The process of constructing an inventory-demand matching model based on the characteristics of warehousing demand fluctuations and preset market trend forecast data includes: The final corrected warehousing demand fluctuation characteristics are input into a pre-trained demand prediction network to generate a demand prediction distribution within a preset time period in the future. The demand prediction distribution includes multiple demand time intervals and the predicted demand quantity corresponding to each demand time interval. Based on the capacity change sequence of different storage locations in the dynamic distribution characteristics of inventory, the real-time inventory matching degree of each storage location is calculated, where the real-time inventory matching degree is the dynamic difference ratio between the current inventory and the predicted demand. Based on the dynamic difference ratio, the multiple demand time intervals are sorted to generate the inventory adjustment priority of each storage area. The inventory adjustment priority is used to indicate the replenishment order and replenishment quantity allocation weight of the storage area. The demand time interval is matched with the preset replenishment cycle to determine the replenishment triggering conditions for each storage location area. The replenishment triggering conditions include a replenishment start time threshold and a replenishment quantity threshold. The inventory adjustment priority is bound to the replenishment triggering condition by rules to generate an inventory-demand matching model that includes dynamic priority adjustment logic and replenishment condition judgment logic. The inventory-demand matching model is used to dynamically output replenishment strategies and storage location allocation instructions based on real-time inventory data.

4. The method according to claim 3, characterized in that, The step of generating a warehousing operation sequence by invoking a dynamic path optimization algorithm based on the dynamic distribution characteristics of the inventory and the inventory adjustment priority includes: Based on the inventory adjustment priority, target storage areas that need to be adjusted are selected from the inventory dynamic distribution characteristics, and real-time inventory data and spatial location data of the target storage areas are obtained. The real-time inventory data is matched with the replenishment triggering conditions to determine the replenishment execution instruction for the target storage area. The replenishment execution instruction includes the replenishment quantity and the starting point of the replenishment path. Based on the spatial location data and the preset picking equipment movement constraints, an initial picking path is generated from the starting point of the replenishment path to the target storage location. The initial picking path is subjected to dynamic weight optimization processing, which includes multi-objective normalization calculation of path length weight, equipment energy consumption weight and time cost weight, to generate an optimized picking path plan. The storage location allocation strategy is configured as a storage location goods distribution adjustment instruction based on the optimized picking path planning, and the storage location allocation strategy, the picking path planning and the replenishment execution instruction are integrated into the warehousing operation sequence.

5. The method according to claim 4, characterized in that, The step of coordinating and scheduling the warehousing operation sequence based on the replenishment trigger condition to generate an optimized warehousing operation instruction set includes: Based on the replenishment strategy and storage location allocation instructions output by the inventory-demand matching model, the path conflict conditions between the storage location allocation strategy and the replenishment execution instructions in the warehousing operation sequence are detected. The path conflict conditions include the spatial overlap rate between the picking path planning and the replenishment path in the same storage location area and the conflict indicator of the equipment movement direction. If the path conflict condition is detected, the replenishment quantity threshold of the target storage location area is recalculated based on the real-time inventory matching degree in the dynamic inventory distribution characteristics, and an adjusted replenishment execution instruction is generated based on the replenishment quantity threshold. The starting point of the replenishment path in the adjusted replenishment execution instruction is dynamically weighted and matched with the current position of the equipment in the picking path planning to generate a path node sequence after conflict resolution. The dynamic weight matching includes a weighted calculation of the movement cost between path nodes and the urgency of inventory. The picking route plan is dynamically remapped based on the path node sequence to generate an updated picking route plan, and the updated picking route plan is then fused with the storage location allocation strategy. Based on the replenishment triggering conditions in the inventory-demand matching model, the timeliness of the warehousing operation sequence after instruction fusion is verified, and an optimized warehousing operation instruction set that meets the conditions of inventory capacity constraints and path non-conflict is generated.

6. The method according to claim 5, characterized in that, The process of generating the conflict-resolved path node sequence includes: Obtain the upper limit of the inventory capacity and the current inventory of the target storage area, and calculate the remaining capacity difference between the replenishment quantity in the adjusted replenishment execution instruction and the upper limit of the inventory capacity; If the remaining capacity difference is less than the preset safety buffer, the replenishment quantity threshold is dynamically expanded according to the capacity change sequence in the inventory dynamic distribution characteristics, and a replenishment execution instruction after secondary adjustment is generated. The path cost is calculated between the starting point of the replenishment path in the second-adjusted replenishment execution instruction and the current position of the equipment. The path cost calculation includes multi-objective normalization processing of equipment turning times, path segment length and inventory matching degree. The path node sequence is prioritized according to the normalized path cost to generate a candidate path node set, and the node sequence with the lowest comprehensive cost is selected from the candidate path node set as the conflict-resolved path node sequence. The path node sequence after conflict resolution is aligned with the replenishment triggering conditions in the inventory-demand matching model within a time window to generate the final optimized warehousing operation instruction set.

7. The method according to claim 2, characterized in that, The process of extracting demand features from the historical warehouse data set using AIAgent includes the following steps: The final corrected warehousing demand fluctuation characteristics are jointly analyzed with the inventory dynamic distribution characteristics to generate a multi-dimensional warehousing characteristic set, which includes demand time series correlation degree and storage location spatial distribution density. The demand time-series correlation degree in the multi-dimensional warehouse feature set is processed by sliding window segmentation to generate multiple local demand trend segments, and the peak duration and demand change rate are extracted from each local demand trend segment. The peak duration is dynamically corrected for deviation with the correlation parameters in the market trend prediction data to generate a corrected local demand trend. The demand change rate is then superimposed with the storage location spatial distribution density to generate a demand-spatial coupling feature. The demand-spatial coupling features are pattern-recognized using a pre-set clustering algorithm to determine high-frequency demand storage areas and low-frequency demand storage areas, and the demand-inventory matching deviation coefficient for each storage area is calculated. The high-frequency demand storage area and the low-frequency demand storage area are mapped and matched with the capacity change sequence in the dynamic distribution characteristics of the inventory to generate a set of dynamic feature vectors for input to the inventory-demand matching model.

8. The method according to claim 7, characterized in that, The sliding window segmentation process for the demand time-series correlation degree in the multi-dimensional warehousing feature set includes the following steps: The demand time sequence correlation is divided into equally spaced time sequence windows based on the order execution timestamp, and each time sequence window covers a continuous time period of a preset length; The stationarity of the demand time series correlation within each time window is tested. If the test result indicates non-stationarity, difference processing is performed until a stationary demand series is generated. The stationary demand sequence is decomposed using an autoregressive model to separate the set of long-term trend components, periodic components, and residual components. The long-term trend component is dynamically aligned with the market trend prediction data to generate trend matching weights, and the periodic component is weighted and corrected based on the trend matching weights. The corrected periodic component is superimposed with the residual component to generate a local demand trend segment, and the maximum value range in the local demand trend segment is determined as the peak duration.

9. The method according to claim 4, characterized in that, The dynamic path optimization algorithm implements path planning through the following steps: A warehouse topology network is constructed based on the spatial location data of the warehouse location areas in the dynamic distribution characteristics of the inventory. The nodes in the warehouse topology network correspond to the coordinates of the warehouse location areas, and the edges correspond to the movable paths between the warehouse location areas. Based on the inventory adjustment priority, a real-time inventory status weight is configured for each node in the warehouse topology network, and the real-time inventory status weight is proportional to the inventory adjustment priority. The path search algorithm is invoked to traverse the warehouse topology network starting from the starting node and calculate the set of shortest paths to each target storage area node. Each path in the set of shortest paths includes the path length and equipment movement energy consumption parameters. Detect whether there are path conflict conditions in the shortest path set. The path conflict conditions include conflict in the device movement direction or the load of the path node exceeds a preset threshold. If path conflict conditions exist, a multi-objective optimization algorithm is used to resolve the conflict in the shortest path set to generate an optimized path set. Each path in the optimized path set includes movement direction adjustment parameters and energy consumption balance coefficient after conflict resolution. The path with the lowest overall path cost is selected from the optimized path set as the picking path plan. The overall path cost is determined by a combination of path length, equipment movement energy consumption parameters, and inventory status weight.

10. A computer system, characterized in that, include: A memory, wherein a computer program is stored; A processor for loading the computer program to implement the intelligent warehouse demand forecasting and scheduling method incorporating AIAgent as described in any one of claims 1-9.

Images