Material supply method and system based on dynamic inventory prediction

Abstract

The invention relates to a material replenishment method and system based on dynamic inventory prediction. The method comprises the following steps: firstly, acquiring a historical inventory data set of multiple types of materials comprehensively covering warehouse-in, warehouse-out and inventory allowance records; secondly, in the inventory feature extraction link, material consumption rules and storage state features are innovatively extracted from time and space dimensions respectively; moreover, the dynamic inventory prediction model carries out combined processing on the two types of features, the limitation of a traditional prediction method is broken through, the complex association of materials in time and space can be fully considered, and the accuracy and reliability of inventory change prediction in a target time period are greatly improved. And finally, a supply trigger condition and a supply amount adjustment strategy are generated based on an accurate prediction result, and are fed back to a supply execution system, so that dynamic and accurate supply operation is realized, the inventory overstock or stockout phenomenon is effectively avoided, the inventory management efficiency is remarkably improved, the inventory cost is reduced, and the whole material inventory management process is comprehensively optimized.

Description

Technical Field

[0001] The present invention relates to the technical field of material data analysis, and more specifically, to a material replenishment method and system based on dynamic inventory forecasting. Background Art

[0002] In traditional material inventory management, most methods rely solely on single-dimensional data or simple empirical models to make inventory forecasts and replenishment decisions. For example, a common approach focuses solely on the number of materials entering and leaving the warehouse, lacking the systematic collection and analysis of data over a continuous time period, making it difficult to capture the complex patterns of material consumption and storage.

[0003] Research and analysis have revealed that existing technologies for analyzing inventory characteristics are often limited to surface data, failing to deeply explore material consumption patterns and storage status characteristics across time and space. This results in an incomplete and in-depth understanding of material behavior. Traditional inventory forecasting models are often static and isolated, unable to effectively integrate multi-dimensional information and adapting to the dynamic changes of materials in different scenarios, resulting in low forecast accuracy and reliability.

[0004] In addition, in the formulation of replenishment strategies, existing technologies usually use fixed trigger conditions and supply quantities, and lack the ability to make dynamic adjustments based on real-time inventory changes and forecast results, which can easily lead to inventory backlogs or out-of-stock situations, increasing the company's operating costs and production risks.

[0005] In view of the above-mentioned technical defects, how to achieve efficient and accurate material inventory management is a difficult problem that needs to be overcome. Summary of the Invention

[0006] In view of this, the present invention provides a material replenishment method and system based on dynamic inventory forecasting.

[0007] An embodiment of the present invention provides a material replenishment method based on dynamic inventory forecasting, which is applied to a material replenishment system. The method includes: Acquire a historical inventory data set containing multiple types of materials, wherein the historical inventory data set includes material entry records, exit records, and inventory balance records within a continuous time period; Performing inventory feature extraction processing on the historical inventory data set to obtain time series consumption features reflecting material consumption patterns and spatial storage features reflecting material storage status; Performing joint inventory forecasting on the time series consumption characteristics and the spatial storage characteristics through a dynamic inventory forecasting model to generate inventory change forecast results for multiple types of materials within a target time period; The replenishment trigger conditions and replenishment quantity adjustment strategies for the corresponding materials are generated according to the inventory change prediction results, and the replenishment trigger conditions and replenishment quantity adjustment strategies are fed back to the replenishment execution system to perform dynamic replenishment operations.

[0008] The present invention also provides a material replenishment system, comprising: a memory for storing program instructions and data; a processor for coupling with the memory and executing instructions in the memory to implement the above method.

[0009] The present invention also provides a computer storage medium comprising instructions, which implement the above method when executed on a processor.

[0010] In summary, the embodiment of the present invention constructs a comprehensive and intelligent material inventory management system as a whole. First, it obtains a set of historical inventory data of multiple types of materials that comprehensively covers the records of incoming, outgoing and inventory balances. Secondly, the inventory feature extraction link innovatively extracts the material consumption patterns and storage status characteristics from the time and space dimensions respectively. Thirdly, the dynamic inventory forecasting model jointly processes the above two types of features, breaking through the limitations of traditional forecasting methods, and can fully consider the complex relationship between materials in time and space, greatly improving the accuracy and reliability of inventory change forecasts within the target time period. Finally, based on accurate forecasting results, replenishment trigger conditions and replenishment quantity adjustment strategies are generated and fed back to the replenishment execution system, realizing dynamic and accurate replenishment operations, effectively avoiding inventory backlogs or out-of-stock phenomena, significantly improving inventory management efficiency, reducing inventory costs, and comprehensively optimizing the entire material inventory management process. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0012] Figure 1 A schematic flow chart of the steps of a material replenishment method based on dynamic inventory forecasting provided by an embodiment of the present invention.

[0013] Figure 2 This is a structural block diagram of a material replenishment system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0014] The technical solutions of the present invention will be described below in conjunction with the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementation methods described in the following exemplary embodiments do not represent all implementation methods consistent with the present invention. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present invention. It should be noted that the terms "first", "second", etc. in the specification of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0015] See also Figure 1 , Figure 1 This is a flow chart of a material replenishment method based on dynamic inventory forecasting provided by an embodiment of the present invention. The method is applied to a material replenishment system and may further include steps 110 to 140.

[0016] Step 110: Acquire a historical inventory data set containing multiple types of materials, wherein the historical inventory data set includes material entry records, exit records, and inventory balance records within a continuous time period.

[0017] In a power battery raw material factory, taking cobalt raw materials as an example, the historical inventory data set in the embodiment of the present invention records the relevant information of cobalt raw materials over a period of time. The factory's inventory management system continuously records the storage of cobalt raw materials, such as how much cobalt raw materials are purchased each time, and the specific storage time, thereby forming a storage record. The outbound record records in detail when the cobalt raw materials are taken out of the inventory for use in the production process, such as how much cobalt raw materials are delivered to the battery production workshop every day. The inventory balance record shows the amount of cobalt raw materials remaining in the warehouse in real time. By collecting this data, a historical inventory data set is formed that contains the complete storage, outbound and inventory balance of cobalt raw materials over a continuous time period.

[0018] Step 120: performing inventory feature extraction processing on the historical inventory data set to obtain temporal consumption features reflecting material consumption patterns and spatial storage features reflecting material storage status.

[0019] In this embodiment, inventory feature extraction is performed on a historical inventory data set of cobalt raw materials to obtain key features. This process aims to gain a deeper understanding of the consumption patterns of cobalt raw materials in the temporal dimension and their storage status in the spatial dimension. By analyzing and processing this data, the consumption patterns and storage characteristics of cobalt raw materials are discovered.

[0020] As an implementation method, the inventory feature extraction process is performed on the historical inventory data set to obtain the temporal consumption features reflecting the material consumption pattern and the spatial storage features reflecting the material storage status, including: Step 121: performing time dimension alignment processing on the historical inventory data set to generate a time series data unit of inventory entry, inventory exit, and balance of multiple types of materials having a continuous time series relationship.

[0021] In the cobalt raw material inventory management scenario, since data recording time points may differ, in order to more accurately analyze cobalt raw material inventory changes, it is necessary to align the historical inventory data set in the time dimension. For example, the cobalt raw material inbound, outbound, and inventory balance data recorded at different times of the day are uniformly sorted and aligned according to a fixed time point each day (such as midnight). This generates a cobalt raw material inbound-outbound-balance time series data unit with a continuous time series relationship. This is used to conduct a coherent analysis of the status of cobalt raw materials at various time points and further explore consumption and storage characteristics.

[0022] Step 122: Perform consumption pattern recognition processing on the time series data unit, and extract the change trend data of the material outbound quantity in adjacent time periods as the time series consumption feature, and the time series consumption feature includes the outbound quantity fluctuation frequency index and the outbound quantity change direction consistency data.

[0023] This step performs consumption pattern recognition processing on the time series data units of cobalt raw materials. In this process, the focus is on the changes in the cobalt raw material outbound volume within adjacent time periods to extract the time series consumption characteristics. For example, within a week, the daily cobalt raw material outbound volume is observed, the difference in outbound volume between two adjacent days is compared, and the trend of these differences is analyzed to determine the outbound volume fluctuation frequency index and the consistency data of the outbound volume change direction. These two data, as important components of the time series consumption characteristics, can characterize the consumption change pattern of cobalt raw materials in the time dimension.

[0024] In a preferred embodiment, the processing of consumption pattern recognition on the time series data unit and extracting the change trend data of the material outbound quantity in adjacent time periods as the time series consumption feature includes: Step 1221: extract the difference in the shipment quantity between adjacent time periods in the time series data unit, and calculate the sign change frequency of the shipment quantity difference in the continuous time period as the shipment quantity fluctuation frequency index.

[0025] Taking cobalt raw materials as an example, we extract the cobalt raw material delivery quantities from adjacent time periods (e.g., two consecutive days) from the time series data unit and calculate the difference between them. We then count the frequency of changes in the sign (positive or negative) of these differences over a continuous period (e.g., one month). If the sign of the cobalt raw material delivery quantity difference changes frequently within a month, it indicates a high delivery quantity fluctuation frequency; conversely, if the sign changes less frequently, it indicates a low delivery quantity fluctuation frequency. The delivery quantity fluctuation frequency indicator calculated in this way can quantify the degree of fluctuation in cobalt raw material delivery quantities.

[0026] Step 1222: Analyze the absolute value change trend of the difference in the outbound quantity, and determine the duration of the change direction of the outbound quantity as the consistency data of the change direction of the outbound quantity.

[0027] Continue to analyze the difference in the outbound volume of cobalt raw materials, and pay attention to the changing trends in the absolute values ​​of these differences. For example, observe whether the absolute value of the outbound volume difference gradually increases, gradually decreases, or remains relatively stable over a period of time. Based on this changing trend, determine how long the change in the outbound volume direction can last. If the absolute value of the outbound volume difference maintains an increasing or decreasing trend for a long time, it means that the direction of the change in the outbound volume is more persistent and lasts for a long time; conversely, if the absolute value changes are unstable, it means that the persistence is weak and the duration is short. The consistency data of the change direction of the outbound volume further improves the understanding of the consumption pattern of cobalt raw materials.

[0028] Step 1223: construct a time series smoothing model for the material outbound quantity, perform sliding window smoothing on the outbound quantity of the time series data unit, and generate an outbound quantity smoothing sequence that eliminates random fluctuations.

[0029] To more accurately capture trends in cobalt raw material shipments, a time series smoothing model can be constructed. This model uses a sliding window smoothing method to process cobalt raw material shipment data. For example, a sliding window size of five days is set, and the shipment data for these five days is averaged to obtain a smoothed shipment value. The window is then slid back one day, and a new five-day average is calculated, and so on. This method generates a smoothed series of cobalt raw material shipments that eliminates random fluctuations, providing a more comprehensive view of shipment trends.

[0030] Step 1224: Calculate the average outbound rate data of the material within a preset time window using the outbound quantity smoothing sequence. The average outbound rate data is used to represent the stable consumption level of the material.

[0031] Optionally, the generated smoothed sequence of cobalt raw material outbound quantities is used to calculate the average outbound rate data within a preset time window. For example, if the preset time window is one week, the smoothed sequence values ​​of the outbound quantities within that week are summed and then divided by the number of days in the week to obtain the average daily outbound rate. This average outbound rate data can reflect the stable consumption level of cobalt raw materials during this period, thereby clarifying the normal usage rate of cobalt raw materials.

[0032] Step 1225: Perform periodic fluctuation detection processing on the smoothed sequence of the outbound quantity, and extract the periodic fluctuation cycle length and fluctuation amplitude data of the material outbound quantity as a supplementary time series consumption feature dimension.

[0033] Further analysis is conducted on the smoothed series of cobalt raw material delivery volumes to detect cyclical fluctuations. For example, Fourier transforms are used to identify cyclical patterns in delivery volumes. If cyclical fluctuations are detected, the length of the fluctuation cycle is determined, e.g., a peak fluctuation every 15 days. The amplitude of the fluctuation is also calculated, which is the difference between the peak fluctuation and the average value. These cyclical fluctuation cycle length and amplitude data serve as supplementary time series consumption feature dimensions, further improving the description of cobalt raw material consumption patterns.

[0034] Step 123: Perform storage status analysis on the time series data unit to generate storage quantity distribution characteristics of materials in different storage areas as spatial storage characteristics. The spatial storage characteristics include storage quantity difference data between storage areas and storage quantity stability indicators within a storage area.

[0035] In cobalt raw material inventory management, in addition to focusing on consumption patterns, it's also necessary to analyze its storage status. We perform storage status analysis on the inventory balance records in time series data units to generate spatial storage characteristics that reflect the distribution of cobalt raw material in different storage areas. By analyzing these characteristics, we can further explore the distribution differences of cobalt raw material in various storage areas and the stability of storage levels.

[0036] In another preferred embodiment, performing storage status analysis on the time series data units to generate storage quantity distribution characteristics of materials in different storage areas as spatial storage characteristics includes: Step 1231: performing storage area division processing on the inventory balance records in the time series data unit to obtain storage amount time series sub-units of multiple types of materials in each storage area.

[0037] In a cobalt raw material factory's warehouse, inventory balance records are divided into different storage areas. For example, the warehouse may be divided into three storage areas: A, B, and C. The cobalt raw material inventory balance data at each point in time is mapped to these three storage areas, forming a time series sub-unit of the cobalt raw material storage volume in storage areas A, B, and C. This allows for detailed analysis of the changes in the cobalt raw material storage volume in each storage area.

[0038] Step 1232: Calculate the standard deviation of the storage capacity of each storage area in a continuous time period as the storage capacity stability index within the storage area. The smaller the stability index value, the smaller the storage capacity fluctuation.

[0039] As you can understand, for each storage area's cobalt raw material storage time series subunit, the standard deviation of its storage level over a continuous time period (e.g., one month) is calculated. The standard deviation calculation logic first calculates the square of the difference between the storage level at each time point and the average storage level over that period, sums these squared values, divides them by the number of time periods, and then takes the square root of the result. For example, for storage area A, a small calculated standard deviation indicates that the cobalt raw material storage level in storage area A has low fluctuations and high stability. Conversely, a large standard deviation indicates high fluctuations and low stability.

[0040] Step 1233: Analyze the storage capacity differences between different storage areas during the same time period, and calculate the average value of the storage capacity differences as storage capacity difference data between the storage areas.

[0041] In an embodiment of the present invention, the differences in the storage amounts of cobalt raw materials in different storage areas (e.g., storage areas A, B, and C) over the same time period (e.g., daily) are compared. For example, the daily storage amount differences between storage areas A and B, between A and C, and between B and C are calculated. These differences are then averaged over a period of time (e.g., a week) to obtain storage amount difference data between storage areas. This data can reflect the degree of difference in cobalt raw material storage amounts between different storage areas, thereby indicating the distribution balance of cobalt raw materials across the storage areas.

[0042] Step 1234: extract the change trend of the proportion of the storage capacity of each storage area to the total storage capacity, and generate a storage area proportion change curve as a dynamic feature description of the storage area storage capacity distribution.

[0043] In an embodiment of the present invention, the proportion of the cobalt raw material storage capacity of each storage area to the total storage capacity is calculated, and the changing trend of this proportion over a period of time is observed. For example, the proportion of the cobalt raw material storage capacity of storage areas A, B, and C to the total storage capacity is calculated daily, and these proportions are plotted over time to form a storage area proportion change curve. This curve can intuitively demonstrate the changes in the relative importance of the storage capacity of each storage area at different points in time and can also serve as dynamic reference information for inventory management and resource allocation.

[0044] Step 1235: Perform inflection point detection on the storage area ratio change curve to identify key time nodes where storage capacity is transferred between different areas as characteristic markers of storage state changes.

[0045] Optionally, the storage area share curve can be used to detect inflection points based on slope changes. For example, a significant change in the slope of the curve is identified as an inflection point. These inflection points represent key time points when cobalt raw material storage shifts between different areas. By identifying these points, the factory can understand when significant changes in storage status occur, allowing it to adjust its inventory management strategy in a timely manner.

[0046] Optionally, the method further includes: Step 124: Input the temporal consumption feature and the spatial storage feature into a feature association analysis network for collaborative verification processing to obtain a set of associated features with a unified time base.

[0047] As you can understand, the previously extracted temporal consumption and spatial storage characteristics of the cobalt raw material are fed into the feature association analysis network. The network then performs collaborative verification on these two types of characteristics to ensure their consistency across the time dimension. For example, it checks whether the time periods in the temporal consumption characteristics match the time periods in the spatial storage characteristics. Through a series of analyses and processing, these two types of characteristics are integrated to form a set of associated features with a unified time base.

[0048] Step 125: Based on the correlation calculation results of the different dimensional features in the associated feature set, dynamically adjust the fusion weight data of the temporal consumption feature and the spatial storage feature in the joint inventory forecasting process.

[0049] For example, we analyze the correlation between the frequency of fluctuations in outbound volumes and the difference in storage volumes between storage areas. Based on these correlations, we dynamically adjust the weights for integrating temporal consumption and spatial storage features in the combined inventory forecast. If a high correlation is found between the frequency of fluctuations in outbound volumes and the difference in storage volumes between storage areas, we appropriately increase the weights for these two features in the combined inventory forecast, ensuring that the forecast more accurately reflects actual changes in cobalt raw material inventory.

[0050] Step 130: Perform joint inventory forecasting processing on the temporal consumption characteristics and the spatial storage characteristics through a dynamic inventory forecasting model to generate inventory change forecast results for multiple types of materials within a target time period.

[0051] Furthermore, a dynamic inventory forecasting model combines the temporal consumption characteristics and spatial storage characteristics of cobalt raw materials. The model comprehensively considers the information contained in these two types of characteristics to predict cobalt raw material inventory changes within a target time period (e.g., the next week). This model can generate inventory change forecasts that include cobalt raw material inventory balance trends and delivery volume forecasts.

[0052] In an alternative embodiment, the dynamic inventory forecasting model is used to perform a joint inventory forecasting process on the temporal consumption characteristics and the spatial storage characteristics to generate inventory change forecast results for multiple types of materials within a target time period, including: Step 131: Input the temporal consumption features and the spatial storage features into the feature fusion layer of the dynamic inventory forecasting model, perform association modeling processing in combination with the fusion weight data, and generate a fusion feature vector with spatiotemporal consistency constraints.

[0053] In this embodiment of the present invention, the temporal consumption characteristics and spatial storage characteristics of the cobalt raw material are input into the feature fusion layer of the dynamic inventory forecasting model. In this layer, the two types of features are modeled in a correlation process, combining the previously dynamically adjusted fusion weights. For example, different features are weighted and combined based on their weights, combining the consumption characteristics in the temporal dimension with the storage characteristics in the spatial dimension. This generates a fused feature vector with spatiotemporal consistency constraints, which integrates the temporal and spatial characteristic information of the cobalt raw material.

[0054] Step 132: The fused feature vector is subjected to collaborative extraction processing of consumption detail features and storage trend information by the multi-scale perception layer of the dynamic inventory forecasting model to obtain a multi-scale perception feature set.

[0055] The fused feature vectors are then analyzed in depth using the dynamic inventory forecasting model's multi-scale perception layer, which collaboratively extracts detailed consumption features and storage trend information from different time scales and feature dimensions. For example, by analyzing cobalt raw material consumption and storage from both short-term and long-term time scales, key features at different scales are extracted, forming a multi-scale perception feature set that contains rich, detailed information on cobalt raw material consumption and storage.

[0056] As a preferred embodiment, the multi-scale perception layer of the dynamic inventory forecasting model performs collaborative extraction processing of consumption detail features and storage trend information on the fused feature vector to obtain a multi-scale perception feature set, including: Step 1321: Input the fused feature vector into the feature segmentation sublayer of the multi-scale perception layer, evenly divide the time dimension of the fused feature vector according to a preset time window segmentation ratio, and generate a two-way parallel feature stream including a first period feature branch and a second period feature branch.

[0057] Specifically, the fused feature vector is input into the feature segmentation sublayer of the multi-scale perception layer. The time dimension of the fused feature vector is evenly divided according to a preset time window segmentation ratio (e.g., dividing the time dimension into two equal parts). This generates a dual-path parallel feature stream consisting of a first-period feature branch and a second-period feature branch. For example, the first-period feature branch may contain cobalt raw material feature information for a shorter time period, while the second-period feature branch contains feature information for a longer period.

[0058] Step 1322: In the first periodic feature branch, sliding window convolution kernels with different time window lengths are used to perform multi-level time resolution expansion processing on the first periodic feature branch to extract consumption detail feature maps at different time scales, and the time dimension of the consumption detail feature map is aligned with the time dimension of the second periodic feature branch.

[0059] In the first-period feature branch, sliding window convolution kernels with different time window lengths are applied. For example, convolution kernels with time window lengths of 3 days and 5 days are used to process the first-period feature branch. Through this multi-level time resolution expansion process, detailed consumption features of cobalt raw materials are extracted at different time scales, forming consumption detail feature maps. The time dimensions of these feature maps are ensured to be consistent with the time dimensions of the second-period feature branch.

[0060] Step 1323: In the second periodic feature branch, adaptive time average pooling processing is performed on the second periodic feature branch to generate a storage trend feature vector with accumulated statistical information, and the storage trend feature vector is reshaped in the time dimension to obtain a storage accumulation trend feature graph that matches the time length of the consumption detail feature graph.

[0061] Adaptive time-averaged pooling is performed on the second-period feature branch to generate a storage trend feature vector containing accumulated statistical information. For example, this vector is formed by calculating statistical information such as the average storage amount over a period of time. This vector is then reshaped over time to match the time length of the previously generated consumption detail feature map. This results in a storage accumulation trend feature map, which allows for the extraction of cobalt raw material consumption and storage characteristics from different perspectives.

[0062] Step 1324: Input the consumption detail feature map and the storage accumulation trend feature map into the feature fusion subnetwork, dynamically generate the fusion weight of each time step based on the temporal attention mechanism, perform cross-scale information fusion processing on the consumption detail feature map and the storage accumulation trend feature map through weighted fusion operation, and generate a multi-level fusion feature map set.

[0063] Furthermore, the consumption detail feature map and the storage accumulation trend feature map are input into the feature fusion subnetwork. Within this network, a temporal attention mechanism dynamically generates fusion weights based on the importance of features at each time step. For example, if the consumption detail feature is more important at a given time step, a higher weight is assigned to the consumption detail feature map. A weighted fusion operation then fuses these two feature maps across scales to generate a multi-level fused feature map set. This set integrates information from different time scales and feature dimensions, providing a more comprehensive reflection of the cobalt raw material inventory characteristics.

[0064] Step 1325: normalize the multi-level fusion feature map set in the time dimension to eliminate the dimensional differences between features of different time scales, and splice the normalized feature maps along the time dimension to generate a multi-scale perceptual feature set with a unified time reference.

[0065] Furthermore, the generated multi-level fused feature map set is normalized along the time dimension to ensure that features at different time scales have the same dimension. For example, the feature values ​​at different time scales are scaled or transformed to maintain dimension consistency. The normalized feature maps are then concatenated along the time dimension to form a multi-scale perceptual feature set with a unified time base. This set integrates information about cobalt raw material consumption details and storage trends at different time scales.

[0066] Step 133: Use the time attention layer of the dynamic inventory forecasting model to perform temporal dependency modeling on the multi-scale perceptual feature set to generate an attention weight distribution map in the time dimension.

[0067] In the dynamic inventory forecasting model, the temporal attention layer performs temporal dependency modeling on the generated multi-scale perceptual feature set for cobalt raw materials. This process analyzes the dependencies between features at different time steps within the set. For example, the model examines how the consumption and storage characteristics of cobalt raw materials interact with each other over time. This analysis allows the model to determine which time step features are most important for predicting future inventory changes. Based on these analysis results, a time-based attention weight distribution graph is generated. In this graph, each time step is assigned a weight; larger weights indicate greater importance of the features at that time step in the forecast. This allows the dynamic inventory forecasting model to focus more closely on the features at key time steps, improving forecast accuracy.

[0068] Step 134: Perform dynamic weighted aggregation processing on the multi-scale perceptual feature set according to the temporal attention weight distribution map to obtain a temporal enhancement feature vector that strengthens temporal correlation.

[0069] Optionally, the multi-scale perceptual feature set is dynamically weighted and aggregated based on the generated time attention weight distribution map. Specifically, for each feature vector in the multi-scale perceptual feature set, it is weighted according to the weight value of its corresponding time step in the attention weight distribution map. For example, if the feature vector of a certain time step corresponds to a higher weight value, then the contribution of the feature vector will be greater during the aggregation process. By means of the above-mentioned dynamic weighting method, the feature vectors of different time steps are aggregated, so that the temporal correlation is enhanced in the aggregated vector. Finally, a temporal enhanced feature vector with enhanced temporal correlation is obtained, which better reflects the continuity and correlation of the cobalt raw material inventory characteristics in the time dimension.

[0070] Step 135: Perform inventory change probability prediction processing on the time series enhanced feature vector through the prediction output layer of the dynamic inventory prediction model to generate an inventory change prediction result including an inventory balance change curve and an outbound quantity prediction sequence. The inventory balance change curve is used to represent the inventory balance prediction value of different time periods, and the outbound quantity prediction sequence is used to represent the outbound quantity prediction value of different time periods.

[0071] As you can understand, the prediction output layer of the dynamic inventory forecasting model receives the time-series enhanced feature vector and processes it to predict inventory change probabilities. This layer analyzes and calculates the information contained in the vector. For example, it combines historical cobalt raw material consumption patterns, storage status, and current market demand to predict inventory balances and outbound shipments for different time periods. Ultimately, it generates an inventory change forecast consisting of an inventory balance change curve and an outbound shipment forecast sequence. The inventory balance change curve graphically displays the predicted changes in cobalt raw material inventory balances over different time periods (e.g., daily within the next week); the outbound shipment forecast sequence clearly lists the forecast outbound shipment values ​​for each time period.

[0072] Step 140: Generate replenishment trigger conditions and replenishment quantity adjustment strategies for corresponding materials based on the inventory change prediction results, and feed the replenishment trigger conditions and replenishment quantity adjustment strategies back to the replenishment execution system to execute dynamic replenishment operations.

[0073] In an embodiment of the present invention, based on the inventory change forecast results of cobalt raw materials, the factory needs to formulate corresponding replenishment strategies to ensure smooth production. This process includes determining the replenishment trigger conditions and adjusting the replenishment quantity, and conveying these strategies to the replenishment execution system for actual operation.

[0074] In an optional embodiment, generating a replenishment trigger condition and a replenishment quantity adjustment strategy for a corresponding material based on the inventory change forecast result includes: Step 141: Analyze the inventory balance change curve in the inventory change prediction result, and detect the time point when the inventory balance is lower than a preset safety threshold as the replenishment trigger time point.

[0075] Optionally, the inventory balance change curve generated in the inventory change forecast results is analyzed in detail. A safety threshold is preset based on the factory's production needs and safety stock standards. For example, the safety stock threshold for cobalt raw materials is set at 500 kg. On the inventory balance change curve, the system searches for points where the inventory balance falls below 500 kg. These points are potential triggers for replenishment. This method allows for timely detection of insufficient inventory.

[0076] As an implementation, analyzing the inventory balance change curve in the inventory change prediction result and detecting the time point when the inventory balance is lower than a preset safety threshold as the replenishment trigger time point includes: Step 1411: performing smoothing and filtering processing on the inventory balance change curve to eliminate high-frequency noise interference in the prediction result and generate a smoothed inventory balance change curve.

[0077] In practical applications, the inventory balance curve may be affected by various factors, resulting in high-frequency noise. This noise can interfere with the determination of the actual inventory trend. Therefore, smoothing the inventory balance curve is necessary. For example, a moving average filter can be used, with an appropriate window size (e.g., three days) set to process the data points on the curve. The data points within the window are averaged, and the average value is used to replace the data point at the center of the window. This process is repeated for the entire curve. This method eliminates high-frequency noise in the forecast results, generating a smoother inventory balance curve that more accurately reflects the actual inventory balance trend.

[0078] Step 1412: Traverse all time points on the smoothed inventory balance change curve and extract a set of time points where the inventory balance value is less than or equal to a preset safety threshold.

[0079] On the smoothed inventory balance curve, starting from the starting time point, each time point is traversed in sequence. The corresponding inventory balance value at each time point is checked to see if it is less than or equal to the preset safety threshold. For example, if the preset safety threshold is 500 kg, when traversing to a time point with an inventory balance value of 480 kg, that time point is extracted and added to the set of time points where the inventory balance value is less than or equal to the preset safety threshold. This comprehensive traversal and extraction process accurately determines all time points where inventory balance is insufficient.

[0080] Step 1413: Perform continuity analysis on the set of time points, and identify the first time point in the continuous time points as the initial trigger candidate point.

[0081] For example, a set of time points might include multiple consecutive time points where inventory levels fell below the safety threshold, such as on days 5, 6, and 7. In this case, the first time point in this set of consecutive time points, day 5, is identified as the candidate initial trigger point. This continuity analysis can determine the earliest point at which inventory levels fell below the safety threshold, providing critical information for timely replenishment measures.

[0082] Step 1414: Extract the inventory balance value at the time point before the initial trigger candidate point, and determine whether the inventory balance value is greater than a preset safety threshold. If so, retain the initial trigger candidate point as the replenishment trigger time point; otherwise, continue to traverse backward.

[0083] For the determined initial trigger candidate point, the inventory balance value at the previous time point is extracted. For example, if the initial trigger candidate point is the 5th day, the inventory balance value on the 4th day is extracted. This value is then determined to be greater than the preset safety threshold. If the inventory balance value on the 4th day is 520 kg, which is greater than the preset safety threshold of 500 kg, the 5th day is retained as the replenishment trigger time point. If the inventory balance value on the 4th day is also less than or equal to 500 kg, the system continues to traverse the time point set backward to find the next qualifying time point as the replenishment trigger time point. This judgment mechanism ensures that the determined replenishment trigger time point is when the inventory balance just falls below the safety threshold, avoiding triggering the replenishment operation too early or too late.

[0084] Step 1415: Perform end time verification on the last time point in the time point set so that the inventory balance value of the subsequent time point of the last time point recovers to above the safety threshold after replenishment, thereby generating a final replenishment trigger time point set.

[0085] Optionally, an end time verification is performed for the last time point in the time point set. For example, if the last time point in the time point set is the 7th day, it is necessary to verify whether the inventory balance value on the 8th day and later can be restored to above the safety threshold after the replenishment operation. Through a prediction model or other relevant analysis methods, determine whether the subsequent inventory balance can meet the safety requirements if replenishment is performed on the 7th day. If so, these verified time points are used as the final replenishment trigger time point set. Through this end time verification, it is ensured that the replenishment operation can effectively restore the inventory balance to a safe level and ensure the normal progress of production.

[0086] Step 142: extract the inventory balance change curve before the replenishment trigger time point's stock-out quantity forecast sequence, and calculate the cumulative stock-out quantity from the current time to the replenishment trigger time point as the material quantity data to be replenished.

[0087] After determining the replenishment trigger time, the forecast sequence of outbound shipments before the replenishment trigger time is extracted from the inventory change forecast results. For example, if the replenishment trigger time is the 5th day, the forecast sequence of outbound shipments from the current time (e.g., day 1) to day 5 is extracted. All forecast outbound shipments in this sequence are then accumulated. For example, the forecast outbound shipments from day 1 to day 5 are 100 kg, 120 kg, 110 kg, 90 kg, and 130 kg, respectively. These values ​​are added together to obtain the cumulative outbound shipments from the current time to the replenishment trigger time. This value serves as the material quantity data to be replenished, clearly indicating the amount of cobalt raw materials that need to be replenished.

[0088] Step 143: Analyze the inventory balance decrease rate of the inventory balance change curve after the replenishment trigger time point, and determine the minimum increment and maximum increment range of the replenishment quantity adjustment in combination with the material's storage cycle data.

[0089] Analyze the portion of the inventory balance curve after the replenishment trigger point to calculate the rate of inventory balance decline. For example, by observing the inventory balance changes at several time points (such as days 6 and 7) after the replenishment trigger point (day 5), calculate the daily inventory balance decline and thus the rate of decline. Furthermore, consider the cobalt raw material incoming inventory cycle data and the time required for procurement and transportation of cobalt raw materials. If the incoming inventory cycle is three days, the replenishment adjustment range must ensure that the inventory balance does not fall too low during the incoming inventory cycle. Based on this analysis, determine the minimum and maximum increment ranges for replenishment adjustments. For example, the minimum increment could be 200 kg to ensure timely replenishment of consumption, while the maximum increment could be 500 kg to avoid over-replenishment and inventory backlogs.

[0090] Step 144: Construct a supply and demand balance model for material replenishment, take the data on the quantity of materials to be replenished and the replenishment quantity adjustment range as input, and generate a replenishment quantity adjustment strategy that meets the inventory safety threshold requirement.

[0091] This step aims to build a supply and demand balance model for material replenishment. The purpose of this supply and demand balance model is to develop a replenishment quantity adjustment strategy that meets the inventory safety threshold requirements based on the amount of material to be replenished and the replenishment quantity adjustment range.

[0092] In one embodiment, the supply and demand balance model for material replenishment is constructed, and the data of the amount of material to be replenished and the replenishment amount adjustment range are used as input to generate a replenishment amount adjustment strategy that meets the inventory safety threshold requirement, including: Step 1441: Set the input variables of the supply and demand balance model to be the amount of material to be replenished, the minimum increment and the maximum increment of the supply amount adjustment range, and the output variable to be the supply amount adjustment data.

[0093] This step defines the input and output variables of the supply-demand balance model. The previously calculated amount of material to be replenished, along with the minimum and maximum increments within the defined replenishment adjustment range, serve as input variables. The model's output variable is the specific replenishment adjustment data, which determines the actual amount of cobalt raw material required.

[0094] Step 1442: Setting the constraint condition of the supply-demand balance model to the inventory balance after replenishment remains above a safety threshold for at least a preset time period after the replenishment trigger time point.

[0095] Constraints are set for the supply-demand balance model to ensure that replenishment operations can effectively safeguard inventory. The inventory balance after replenishment is required to remain above a safety threshold for at least a preset period of time (e.g., one week) after the replenishment trigger. This constraint ensures that the replenishment adjustment strategy can meet the basic requirements for cobalt raw material inventory during production, preventing production from being impacted by insufficient inventory.

[0096] Step 1443: Based on the shipment quantity prediction sequence in the inventory change prediction result, calculate the cumulative value of the shipment quantity at each time point after the replenishment trigger time point.

[0097] Based on the forecasted shipment sequence from the inventory change forecast results, calculate the cumulative shipment values ​​at various points in time after the replenishment trigger. For example, if the replenishment trigger is day 5, and the forecasted shipment sequence indicates 100 kg shipped on day 6, 110 kg on day 7, and 90 kg on day 8, the cumulative shipment values ​​on day 6 are 100 kg, 100 kg + 110 kg = 210 kg on day 7, and 100 kg + 110 kg + 90 kg = 300 kg on day 8. By calculating these cumulative values, we can clearly understand the consumption of cobalt raw materials at different points in time after replenishment.

[0098] Step 1444: summing the cumulative value of the shipment quantity and the replenishment quantity adjustment data to obtain the inventory balance forecast value at each time point after replenishment.

[0099] Optionally, the calculated cumulative shipment value is summed with the replenishment adjustment data. For example, if the replenishment adjustment data is 300 kg, and the inventory balance at the replenishment trigger time (day 5) is 480 kg, then the predicted inventory balance on day 6 is 480 + 300 - 100 = 680 kg, the predicted value on day 7 is 680 - 110 = 570 kg, and the predicted value on day 8 is 570 - 90 = 480 kg. This calculation method generates predicted inventory balance values ​​at various time points after replenishment, allowing evaluation of inventory balance changes under different replenishment adjustment strategies.

[0100] Step 1445: Through the optimization solution network of the supply and demand balance model, find the minimum supply quantity adjustment data within the supply quantity adjustment range so that the inventory balance forecast value at all time points is not lower than the safety threshold, as the final supply quantity adjustment strategy.

[0101] As you can understand, the optimization network of the supply-demand balance model searches within a set supply adjustment range (e.g., 200 kg to 500 kg). The goal is to find the minimum supply adjustment that ensures the predicted inventory balance at all points within the preset time period remains above a safety threshold (e.g., 500 kg). By continuously trying different supply adjustment data and evaluating them against the previously calculated inventory balance forecast, the minimum supply adjustment is ultimately determined as the final supply adjustment strategy.

[0102] Step 145: Generate a replenishment trigger condition set including timestamp alignment according to the replenishment trigger time point and the replenishment quantity adjustment strategy. The replenishment trigger condition set includes a trigger time point, a triggering inventory balance threshold, and corresponding replenishment quantity adjustment data.

[0103] Optionally, a replenishment trigger condition set is generated based on the determined replenishment trigger time point and replenishment quantity adjustment strategy. The information in this set is presented in a timestamp-aligned manner, which facilitates the replenishment execution system to accurately execute operations. For example, the replenishment trigger time point is the 5th day, the inventory balance threshold at the time of triggering is 500 kilograms, and the corresponding replenishment quantity adjustment data is 350 kilograms. The generated replenishment trigger condition set then contains this information, clarifying that on the 5th day, when the inventory balance reaches 500 kilograms, a replenishment operation needs to be executed, and the replenishment quantity is 350 kilograms. This clear information set ensures that the replenishment execution system can perform dynamic replenishment operations accurately.

[0104] In an optional design approach, feeding back the replenishment triggering condition and the replenishment amount adjustment strategy to the replenishment execution system to perform a dynamic replenishment operation includes: Step 146: Standardize the data format of the trigger time point, the inventory balance threshold at the time of triggering, and the corresponding replenishment quantity adjustment data in the replenishment trigger condition to generate a replenishment instruction data unit that complies with the interface specification of the replenishment execution system; extract the trigger time point in the replenishment instruction data unit and perform time synchronization calibration with the current system time.

[0105] The generated replenishment trigger conditions undergo data format standardization to conform to the interface specifications of the replenishment execution system. For example, the trigger time is formatted according to the set date and time format, and the trigger inventory threshold and replenishment quantity adjustment data are converted to the specified data type and length. After processing is complete, a replenishment instruction data unit is generated. The trigger time is then extracted from this data unit and synchronized with the current system time. For example, if the current system time is accurate real-time, the trigger time in the replenishment instruction data unit is compared and adjusted with the current system time to ensure the accuracy of the trigger time, allowing the replenishment execution system to execute the operation at the correct time.

[0106] Step 147: performing a matching and verification process on the supply quantity adjustment data in the supply instruction data unit and the currently available supply resources of the supply execution system to generate a resource matching status identifier.

[0107] Optionally, the replenishment quantity adjustment data in the replenishment instruction data unit is compared and verified with the currently available replenishment resources in the replenishment execution system. For example, if the replenishment quantity adjustment data requires the replenishment of 350 kilograms of cobalt raw material, the replenishment execution system checks whether the amount of cobalt raw material currently available in the warehouse is sufficient. If the available resources can meet the replenishment requirement of 350 kilograms, a resource matching status indicator "match successful" is generated; if the available resources are insufficient, such as only 200 kilograms of cobalt raw material, a resource matching status indicator "match failed" is generated. This matching verification process ensures that the replenishment operation is carried out when resources are available.

[0108] Step 148: If the resource matching status is marked as a successful match, the supply instruction data unit is sent to the execution control node of the supply execution system to execute the supply operation; if the resource matching status is marked as a failed match, a supply resource scheduling request is generated based on the difference between the supply quantity adjustment data and the available supply resources, and the scheduling request is sent to the resource management system to coordinate the replenishment of the available supply resources before executing the supply operation.

[0109] When the resource matching status is marked as "matched successfully," the replenishment instruction data unit is sent to the execution control node of the replenishment execution system. Upon receiving the instruction, the execution control node executes the replenishment operation according to the instruction information. For example, on the fifth day, when the inventory balance reaches 500 kg, the replenishment operation of 350 kg of cobalt raw material is executed. If the resource matching status is marked as "matched failed," the difference between the replenishment quantity adjustment data and the available replenishment resources is calculated. For example, if the replenishment quantity adjustment data is 350 kg and the available resources are 200 kg, the difference is 150 kg. Based on this difference, a replenishment resource scheduling request is generated and sent to the resource management system. Upon receiving the request, the resource management system coordinates the replenishment of available replenishment resources, such as allocating 150 kg of cobalt raw material from another warehouse to the current warehouse. After the resource replenishment is completed, the resource matching verification is performed again. If the match is successful, the replenishment instruction data unit is sent to the execution control node of the replenishment execution system to execute the replenishment operation.

[0110] As an independent embodiment, the method further includes: Obtain the actual replenishment operation records and real-time inventory data after execution returned by the replenishment execution system; Perform time series alignment and matching on real-time inventory data and inventory change forecasts generated by a dynamic inventory forecasting model. Extract the deviation feature set between the forecasted and actual values ​​for the corresponding time period. The deviation feature set includes inventory deviation trend data, outbound quantity deviation fluctuation data, and replenishment quantity deviation direction data. A replenishment execution effect evaluation model is constructed based on the deviation feature set. The execution effect evaluation model is used to perform multi-dimensional evaluation processing on the accuracy of dynamic inventory forecasts, the effectiveness of replenishment trigger conditions, and the rationality of replenishment quantity adjustment strategies. A replenishment execution effect evaluation report is generated, which includes the confidence level of forecast deviations, the compliance rate of trigger conditions, and the adaptability of strategy adjustments. The replenishment execution effect evaluation report is used for iterative optimization of the dynamic inventory forecast model and adaptive adjustment of the replenishment strategy generation logic.

[0111] In the inventory management process for cobalt raw materials, the replenishment execution system returns a record of the actual replenishment operation after completing the replenishment operation. For example, the record shows that 350 kg of cobalt raw materials were replenished according to the replenishment instruction on the 5th day. It also provides real-time inventory data after execution, such as the inventory balance after execution was 830 kg. This real-time inventory data is time-aligned with the inventory change forecast results generated by the dynamic inventory forecasting model. If the forecasting model predicts that the inventory balance after the replenishment on the 5th day will be 850 kg, the deviation between the forecast and the actual value for the corresponding time period is extracted through comparison.

[0112] For inventory balance deviation trend data, analyze the difference between the predicted and actual inventory balance over time. If this difference shows a gradual increase or decrease over multiple time periods, you can identify an inventory balance deviation trend. For example, if after several consecutive replenishments, the predicted inventory balance is consistently higher than the actual value by a certain amount, and this value tends to increase, this indicates an inventory balance deviation trend.

[0113] Shipment volume deviation and fluctuation data focuses on the fluctuations between the forecasted and actual shipment volumes. For example, if the forecasted daily shipment volume is relatively stable within a certain time period, but the actual shipment volume fluctuates significantly, the shipment volume deviation and fluctuation data can be obtained by statistically analyzing these fluctuations.

[0114] The supply deviation direction data determines the direction of the difference between the supply forecast and the actual supply. If the actual supply is greater than the forecast, the supply deviation direction is positive; otherwise, it is negative.

[0115] Based on this set of deviation features, a replenishment execution effectiveness evaluation model was constructed. This model performs a multi-dimensional assessment of the accuracy of dynamic inventory forecasts, the effectiveness of replenishment trigger conditions, and the rationality of replenishment quantity adjustment strategies. For example, the model calculates the confidence level of forecast deviations by statistically analyzing the deviations between a large number of predicted and actual values ​​to assess the reliability of the forecast results. The model also calculates the degree of compliance between the actual and expected trigger conditions of replenishment trigger conditions to obtain the trigger condition compliance rate. Furthermore, the model analyzes whether the replenishment quantity adjustment strategy achieves the expected results in actual execution to determine the adaptability of the strategy adjustment.

[0116] Ultimately, a replenishment execution performance evaluation report, including these evaluation metrics, is generated. This report is crucial for the iterative optimization of the dynamic inventory forecasting model. If the forecast deviation confidence level is low, the model may need parameter adjustments or structural improvements. If the trigger condition compliance rate is low, the replenishment trigger setting may need to be reviewed. If the strategy adjustment is poorly adapted, the replenishment quantity adjustment strategy may need to be optimized. Through these feedback and adjustments, the accuracy and effectiveness of inventory management are continuously improved.

[0117] As another independent embodiment, the method further includes: Collect complete inventory operation data for each time period after replenishment is executed, and perform time dimension alignment processing according to the time period division rules of the historical inventory data set to generate inventory data units for the newly added time period; The inventory data units of the newly added time period are spliced ​​with the original historical inventory data set in time series to form an updated historical inventory data set that includes the extended time span. The updated historical inventory data set maintains the integrity of the inventory-in, inventory-out, and balance records within the continuous time period; Perform data quality verification on the updated historical inventory data set, through time series continuity detection, outlier identification and missing value filling operations, to ensure the time series consistency and feature extractability of the updated historical inventory data set; The updated historical inventory dataset is used as input data for subsequent inventory feature extraction to capture the latest changing trends in material consumption patterns and storage status.

[0118] After each replenishment, complete inventory data for each time period is collected. For example, for cobalt raw materials, detailed records are kept of incoming and outgoing quantities, as well as inventory balances, for each time period. Based on the time period division rules previously established for the historical inventory data set, for example, daily as a time period, the new data is time-aligned. Ensure that the time information in the new data is consistent with the original time period divisions, and generate inventory data units for the newly added time period.

[0119] The inventory data units for the newly added time period are spliced ​​with the original historical inventory data set in time series. For example, if the original historical inventory data set records up to the 30th day, the newly added data begins on the 31st day. These new data are then chronologically spliced ​​onto the end of the original data set to form an updated historical inventory data set covering the extended time span. This process ensures the integrity of the inventory inflow, outflow, and balance records within the continuous time period, so that the entire data set fully reflects the changes in cobalt raw material inventory over time.

[0120] Perform data quality verification on the updated historical inventory data set. First, perform a time series continuity check to verify that the data is continuous across the time dimension and to identify any gaps or jumps. For example, check to see if any data for a particular day is missing, causing a break in the time series. If any discontinuity is found, address it accordingly.

[0121] Next, outlier identification is performed, using statistical analysis and other methods to identify data values ​​that significantly deviate from the normal range. For example, if the amount of cobalt raw materials shipped out of the warehouse on a certain day is suddenly many times greater than usual, this may be due to data recording errors or other abnormal reasons and needs to be marked and processed.

[0122] For missing values, appropriate methods are used. For example, interpolation can be performed based on data from previous and subsequent time periods, or missing values ​​can be predicted using statistical models. These operations ensure temporal consistency and feature extraction within the updated historical inventory data set.

[0123] Finally, the updated historical inventory dataset is used as input for subsequent inventory feature extraction. Because this new data includes the latest inventory status, it helps capture the latest trends in cobalt raw material consumption and storage conditions. For example, new trends such as a recent acceleration in cobalt raw material consumption or increased material transfers between storage areas may be discovered.

[0124] As another independent embodiment, the method further includes: Obtain the forecast deviation feature set in the replenishment execution effect evaluation report and the newly added inventory data units in the updated historical inventory data set and use them as input information for model optimization; Through the parameter self-adjustment branch of the dynamic inventory forecasting model, based on the balance deviation trend data and the inventory deviation fluctuation data in the forecast deviation feature set, the fusion weight data of the time series consumption characteristics and the spatial storage characteristics is corrected and processed to generate the adjusted fusion weight that adapts to the latest consumption-storage association relationship; The new inventory data units are used to calibrate the feature response of the multi-scale perception layer of the dynamic inventory forecasting model. By comparing the output distribution differences of the multi-scale perception feature set on new and old data, the time window segmentation ratio and the feature extraction preference of the sliding window convolution kernel in the multi-scale perception layer are adjusted. The adjusted fusion weights and calibrated multi-scale perception layer parameters are injected into the dynamic inventory prediction model.

[0125] In this embodiment, the forecast deviation feature set from the replenishment execution effectiveness evaluation report and the newly added inventory data units from the updated historical inventory data set are obtained. The balance deviation trend data and the shipment deviation fluctuation data in the forecast deviation feature set reflect inaccuracies in the dynamic inventory forecasting model during the previous forecasting process. For example, the balance deviation trend indicates that the predicted inventory balance value is consistently higher than the actual value, while the shipment deviation fluctuation indicates that the actual shipment volume fluctuates more than the predicted value.

[0126] The dynamic inventory forecasting model's parameter self-adjustment branch uses this deviation data to modify the fusion weights of the temporal consumption and spatial storage features. For example, if the remaining inventory deviation is primarily due to overweighting the temporal consumption feature, the weight of the temporal consumption feature is appropriately lowered, while the weight of the spatial storage feature is adjusted accordingly. This generates adjusted fusion weights that adapt to the latest consumption-storage relationship. This allows the model to better balance information in the temporal and spatial dimensions in subsequent forecasts, improving forecast accuracy.

[0127] The multi-scale perception layer of the dynamic inventory forecasting model is calibrated for feature response using newly added inventory data units. The output distribution of the multi-scale perception feature set on the old and new data is compared. For example, on the old data, the consumption detail feature map extracted under a certain time window segmentation ratio can better reflect the consumption of cobalt raw materials, but on the new data, the output distribution of the feature map has changed, which may mean that the current time window segmentation ratio is no longer applicable. By analyzing these differences, the time window segmentation ratio and the feature extraction preference of the sliding window convolution kernel in the multi-scale perception layer are adjusted. For example, the time window segmentation ratio can be appropriately increased or decreased, or the size, step size and other parameters of the sliding window convolution kernel can be adjusted, so that the multi-scale perception layer can better extract key features from the new data.

[0128] Finally, the adjusted fusion weights and calibrated multi-scale perception layer parameters are injected into the dynamic inventory forecasting model. In this way, the dynamic inventory forecasting model can use these optimized parameters to more accurately process the temporal consumption characteristics and spatial storage characteristics of cobalt raw materials in subsequent inventory forecasts, thereby improving the accuracy and reliability of the forecast, thereby providing stronger support for the factory's inventory management and replenishment decisions, continuously optimizing the entire inventory management process, and ensuring the smooth progress of production and the rational use of resources.

[0129] In an embodiment of the present invention, the time series analysis method, deep learning framework and optimization algorithm in the prior art can be used to supplement the further optimization of feature extraction, model construction and strategy generation.

[0130] For the extraction of time-series consumption features, such as the calculation of the frequency index of fluctuations in outbound volume and the consistency data of change direction, the sliding average or exponential smoothing method (such as the Holt-Winters method) can be used to process the outbound volume data to fully reveal the consumption pattern; by utilizing the standardized implementation of Fourier transform to detect periodic fluctuations, the adequacy of the extraction of periodic dimensions can be ensured.

[0131] In the analysis of spatial storage characteristics, the calculation of standard deviation and the generation of storage capacity variability data can be efficiently implemented through the statistical analysis toolkit of Python or MATLAB, avoiding dimensional inconsistency and ensuring the clarity of storage capacity stability indicators.

[0132] The multi-scale perception layer and temporal attention layer of the dynamic inventory forecasting model can be built based on the TensorFlow or PyTorch framework, using the LSTM or Transformer architecture for temporal dependency modeling. By adaptively adjusting the size and step size of the sliding window convolution kernel, consumption details and storage trends can be captured, making the feature fusion and weighted aggregation processes operational. At the same time, the optimization solution network of the supply and demand balance model can use the Gurobi or CPLEX solver to handle the search for the minimum supply quantity under constraints, ensuring the rationality of the supply quantity adjustment strategy.

[0133] During the replenishment triggering and resource matching phase, real-time data stream APIs (such as RESTful services) are used to automatically align the trigger time with the inventory balance threshold. This is then combined with the inventory API in the resource management system to verify resource matching, ensuring consistency in dynamic replenishment execution. This design improves forecast accuracy and replenishment efficiency.

[0134] In summary, the embodiment of the present invention constructs a comprehensive and intelligent material inventory management system as a whole. First, it obtains a set of historical inventory data of multiple types of materials that comprehensively covers the records of incoming, outgoing and inventory balances. Secondly, the inventory feature extraction link innovatively extracts the material consumption patterns and storage status characteristics from the time and space dimensions respectively. Thirdly, the dynamic inventory forecasting model jointly processes the above two types of features, breaking through the limitations of traditional forecasting methods, and can fully consider the complex relationship between materials in time and space, greatly improving the accuracy and reliability of inventory change forecasts within the target time period. Finally, based on accurate forecasting results, replenishment trigger conditions and replenishment quantity adjustment strategies are generated and fed back to the replenishment execution system, realizing dynamic and accurate replenishment operations, effectively avoiding inventory backlogs or out-of-stock phenomena, significantly improving inventory management efficiency, reducing inventory costs, and comprehensively optimizing the entire material inventory management process.

[0135] Further, Figure 2 The structural block diagram of the material replenishment system 300 is shown, which includes: a memory 310 for storing program instructions and data; a processor 320 for coupling with the memory 310 and executing the instructions in the memory 310 to implement the above method.

[0136] Furthermore, a computer storage medium is provided, comprising instructions, which implement the above method when executed on a processor.

[0137] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A material replenishment method based on dynamic inventory forecasting, characterized in that: The method comprises: Acquire a historical inventory data set containing multiple types of materials, wherein the historical inventory data set includes material entry records, exit records, and inventory balance records within a continuous time period; Performing inventory feature extraction processing on the historical inventory data set to obtain time series consumption features reflecting material consumption patterns and spatial storage features reflecting material storage status; Performing joint inventory forecasting on the time series consumption characteristics and the spatial storage characteristics through a dynamic inventory forecasting model to generate inventory change forecast results for multiple types of materials within a target time period; The replenishment trigger conditions and replenishment quantity adjustment strategies for the corresponding materials are generated according to the inventory change prediction results, and the replenishment trigger conditions and replenishment quantity adjustment strategies are fed back to the replenishment execution system to perform dynamic replenishment operations.

2. The method according to claim 1, characterized in that The performing of inventory feature extraction processing on the historical inventory data set to obtain time series consumption features reflecting material consumption patterns and spatial storage features reflecting material storage status includes: Performing time dimension alignment processing on the historical inventory data set to generate a time series data unit of inbound-outbound-balance of multiple types of materials with a continuous time series relationship; Perform consumption pattern recognition processing on the time series data unit, extract the change trend data of the material outbound quantity in adjacent time periods as the time series consumption feature, and the time series consumption feature includes the outbound quantity fluctuation frequency index and the outbound quantity change direction consistency data; Performing storage status analysis on the time series data units to generate storage quantity distribution characteristics of materials in different storage areas as spatial storage characteristics, wherein the spatial storage characteristics include storage quantity difference data between storage areas and storage quantity stability indicators within storage areas; The method further comprises: Inputting the temporal consumption feature and the spatial storage feature into a feature association analysis network for collaborative verification processing to obtain a set of associated features with a unified time base; Based on the correlation calculation results of the different dimensional features in the associated feature set, the fusion weight data of the temporal consumption feature and the spatial storage feature in the joint inventory forecasting process is dynamically adjusted.

3. The method according to claim 2, characterized in that The performing consumption pattern recognition processing on the time series data unit and extracting the change trend data of the material outbound quantity in adjacent time periods as the time series consumption feature includes: Extracting the difference in the shipment quantity between adjacent time periods in the time series data unit, and calculating the sign change frequency of the shipment quantity difference in the continuous time period as the shipment quantity fluctuation frequency index; Analyze the absolute value change trend of the difference in the outbound quantity, and determine the duration of the change direction of the outbound quantity as the consistency data of the change direction of the outbound quantity; Constructing a time series smoothing model for material outbound quantities, performing sliding window smoothing on the outbound quantities of the time series data units, and generating an outbound quantity smoothing sequence that eliminates random fluctuations; Calculating the average outbound rate data of the material within a preset time window through the outbound quantity smoothing sequence, wherein the average outbound rate data is used to represent the stable consumption level of the material; The periodic fluctuation detection process is performed on the smoothed sequence of the outbound quantity, and the periodic fluctuation cycle length and fluctuation amplitude data of the outbound quantity of materials are extracted as a supplementary time series consumption feature dimension.

4. The method according to claim 2, characterized in that The performing storage status analysis on the time series data unit to generate storage quantity distribution characteristics of materials in different storage areas as spatial storage characteristics includes: Performing storage area division processing on the inventory balance records in the time series data unit to obtain storage amount time series sub-units of multiple types of materials in each storage area; Calculating the standard deviation of the storage capacity of each storage area over a continuous time period as a storage capacity stability index within the storage area, wherein a smaller value of the stability index indicates smaller fluctuation of the storage capacity; Analyze the storage capacity differences between different storage areas during the same time period, and calculate the average value of the storage capacity differences as the storage capacity difference data between the storage areas; Extract the changing trend of the proportion of storage capacity in each storage area to the total storage capacity, and generate a storage area proportion change curve as a dynamic feature description of the storage capacity distribution of the storage area; An inflection point detection process is performed on the storage area ratio change curve to identify key time nodes for storage capacity transfer between different areas as characteristic markers of storage state changes.

5. The method according to claim 2, characterized in that The dynamic inventory forecasting model is used to perform joint inventory forecasting on the time series consumption characteristics and the spatial storage characteristics to generate inventory change forecast results for multiple types of materials within a target time period, including: Inputting the temporal consumption features and the spatial storage features into the feature fusion layer of the dynamic inventory forecasting model, performing association modeling processing in combination with the fusion weight data, and generating a fusion feature vector with spatiotemporal consistency constraints; Performing collaborative extraction processing of consumption detail features and storage trend information on the fused feature vector through the multi-scale perception layer of the dynamic inventory forecasting model to obtain a multi-scale perception feature set; Using the temporal attention layer of the dynamic inventory forecasting model to perform temporal dependency modeling on the multi-scale perceptual feature set, and generating an attention weight distribution map in the time dimension; Performing dynamic weighted aggregation processing on the multi-scale perceptual feature set according to the temporal attention weight distribution map to obtain a temporal enhancement feature vector that strengthens temporal correlation; The prediction output layer of the dynamic inventory prediction model is used to perform inventory change probability prediction processing on the time series enhanced feature vector to generate an inventory change prediction result including an inventory balance change curve and an outbound quantity prediction sequence. The inventory balance change curve is used to represent the inventory balance prediction value of different time periods, and the outbound quantity prediction sequence is used to represent the outbound quantity prediction value of different time periods.

6. The method according to claim 5, characterized in that The multi-scale perception layer of the dynamic inventory forecasting model performs collaborative extraction processing of consumption detail features and storage trend information on the fused feature vector to obtain a multi-scale perception feature set, including: Inputting the fused feature vector into the feature segmentation sublayer of the multi-scale perception layer, uniformly dividing the time dimension of the fused feature vector according to a preset time window segmentation ratio, and generating a two-way parallel feature stream including a first period feature branch and a second period feature branch; In the first periodic feature branch, sliding window convolution kernels with different time window lengths are used to perform multi-level time resolution expansion processing on the first periodic feature branch to extract consumption detail feature maps at different time scales, where the time dimension of the consumption detail feature map is aligned with the time dimension of the second periodic feature branch; In the second periodic feature branch, adaptive time average pooling processing is performed on the second periodic feature branch to generate a storage trend feature vector having accumulated statistical information, and time dimension reshaping processing is performed on the storage trend feature vector to obtain a storage accumulation trend feature graph that matches the time length of the consumption detail feature graph; Input the consumption detail feature map and the storage accumulation trend feature map into the feature fusion sub-network, dynamically generate the fusion weight of each time step based on the temporal attention mechanism, perform cross-scale information fusion processing on the consumption detail feature map and the storage accumulation trend feature map through weighted fusion operation, and generate a multi-level fusion feature map set; The multi-level fusion feature map set is normalized in the time dimension to eliminate the dimensional difference between features of different time scales, and the normalized feature maps are spliced ​​along the time dimension to generate a multi-scale perception feature set with a unified time reference.

7. The method according to claim 1, characterized in that Generating the replenishment trigger conditions and replenishment quantity adjustment strategies for corresponding materials according to the inventory change prediction results includes: Analyze the inventory balance change curve in the inventory change prediction result, and detect the time point when the inventory balance falls below a preset safety threshold as the replenishment trigger time point; Extract the inventory balance change curve before the replenishment trigger time point to predict the outbound quantity, and calculate the cumulative outbound quantity from the current time to the replenishment trigger time point as the material quantity data to be replenished; Analyze the inventory balance decrease rate of the inventory balance change curve after the replenishment trigger time point, and determine the minimum and maximum increment ranges of the replenishment quantity adjustment in combination with the material's storage cycle data; Constructing a supply and demand balance model for material replenishment, taking the material quantity data to be replenished and the replenishment quantity adjustment range as input, and generating a replenishment quantity adjustment strategy that meets the inventory safety threshold requirements; generating a replenishment trigger condition set including timestamp alignment according to the replenishment trigger time point and the replenishment quantity adjustment strategy, wherein the replenishment trigger condition set includes a trigger time point, a triggering inventory balance threshold, and corresponding replenishment quantity adjustment data; The analyzing the inventory balance change curve in the inventory change prediction result and detecting the time point when the inventory balance is lower than a preset safety threshold as the replenishment trigger time point includes: Performing smoothing and filtering processing on the inventory balance change curve to eliminate high-frequency noise interference in the prediction result and generate a smoothed inventory balance change curve; Traversing all time points on the smoothed inventory balance change curve, and extracting a set of time points where the inventory balance value is less than or equal to a preset safety threshold; Performing continuity analysis on the set of time points, and identifying a first time point among the continuous time points as an initial trigger candidate point; Extract the inventory balance value at the time point before the initial trigger candidate point, and determine whether the inventory balance value is greater than a preset safety threshold. If so, retain the initial trigger candidate point as the replenishment trigger time point; otherwise, continue to traverse backwards; Performing end time verification on the last time point in the set of time points, so that the inventory balance value at the subsequent time points of the last time point recovers to above the safety threshold after replenishment, thereby generating a final set of replenishment trigger time points; The supply and demand balance model for material replenishment is constructed, and the replenishment quantity data and the replenishment quantity adjustment range are used as input to generate a replenishment quantity adjustment strategy that meets the inventory safety threshold requirement, including: The input variables of the supply and demand balance model are set as the amount of materials to be replenished, the minimum increment and the maximum increment of the replenishment amount adjustment range, and the output variable is the replenishment amount adjustment data; The constraint condition of the supply and demand balance model is set as follows: the inventory balance after replenishment is maintained at or above a safety threshold for at least a preset time period after the replenishment trigger time point; Calculate the cumulative value of the outbound quantity at each time point after the replenishment trigger time point based on the outbound quantity forecast sequence in the inventory change forecast result; The cumulative value of the outbound quantity and the supply quantity adjustment data are summed to obtain the inventory balance forecast value at each time point after the supply; Through the optimization solution network of the supply and demand balance model, the minimum supply quantity adjustment data that makes the inventory balance forecast value at all time points not lower than the safety threshold is found within the supply quantity adjustment range as the final supply quantity adjustment strategy.

8. The method according to claim 1, characterized in that Feeding back the replenishment triggering condition and the replenishment amount adjustment strategy to the replenishment execution system to perform a dynamic replenishment operation includes: Standardize the data format of the trigger time point, the trigger inventory balance threshold, and the corresponding replenishment quantity adjustment data in the replenishment trigger condition to generate a replenishment instruction data unit that complies with the interface specification of the replenishment execution system; Extracting the trigger time point in the replenishment instruction data unit and performing time synchronization calibration processing with the current system time; Performing matching and verification processing on the supply quantity adjustment data in the supply instruction data unit and the currently available supply resources of the supply execution system to generate a resource matching status identifier; If the resource matching status is marked as a successful match, the replenishment instruction data unit is sent to the execution control node of the replenishment execution system to execute the replenishment operation; If the resource matching status is marked as matching failure, a supply resource scheduling request is generated based on the difference between the supply quantity adjustment data and the available supply resources, and the scheduling request is sent to the resource management system to coordinate the replenishment of the available supply resources before performing the supply operation.

9. A material supply system, characterized in that: include: Memory, used to store program instructions and data; A processor, coupled to a memory, and configured to execute instructions in the memory to implement the method according to any one of claims 1 to 8.

10. A computer storage medium, characterized in that The method comprises instructions which, when executed on a processor, implement the method according to any one of claims 1 to 8.

Images