Warehouse inventory dynamic prediction and space optimization method and system based on RFID and artificial intelligence

Abstract

The application relates to the field of intelligent warehouse management, in particular to a warehouse inventory dynamic prediction and space optimization method and system based on RFID and artificial intelligence, which comprises the following steps: acquiring sensing data of each inventory unit collected by an RFID sensing array in real time; constructing a space-time correlation model describing the dynamic relationship between the inventory unit and the warehouse space according to the sensing data; performing prediction analysis on the space-time correlation model to obtain inventory prediction information and space congestion prediction information in a future period; analyzing the linkage relationship between inventory distribution changes and space congestion changes, and identifying conflict sources which have causal influence on the space congestion changes; performing a pre-adjustment operation on the inventory unit corresponding to the conflict source before the congestion occurs; collecting actual operation data after the pre-adjustment, which is used for updating the prediction analysis and the conflict source identification process next time. The application can accurately locate the causal source of the congestion, realize prospective space optimization, and improve the warehouse operation efficiency.

Description

Technical Field

[0001] This application relates to the field of intelligent warehouse management technology, and in particular to a method and system for dynamic prediction and space optimization of warehouse inventory based on RFID and artificial intelligence. Background Technology

[0002] In warehouse management, there is a strong coupling relationship between inventory distribution and aisle efficiency. Existing technologies typically utilize sensing devices such as RFID to acquire inventory location information and combine this with predictive models to estimate future congestion, thereby adjusting storage locations or operational routes. However, these methods mostly treat inventory changes and spatial congestion as two relatively independent predictive tasks, making it difficult to reveal the deep causal relationship between them. Specifically, when congestion occurs in a certain area, existing technologies often can only identify the congestion itself but cannot accurately pinpoint which one or more specific storage locations' inventory fluctuations caused the chain reaction of congestion. Due to the lack of ability to locate this causal source, subsequent spatial optimization measures often have a certain degree of blindness and are unable to fundamentally prevent congestion.

[0003] On the other hand, when multiple potential congestion sources exist simultaneously in a warehousing system, and their operational paths need to compete for the same scarce resources, such as sharing elevators, main conveyors, or outbound exits, existing technologies typically employ a first-come, first-served or simple prioritization approach, pre-adjusting each source independently. This independent processing method ignores the mutual influence between multiple sources; adjusting one source may unintentionally consume resources needed by other sources, or even exacerbate overall congestion due to local optimization. Therefore, improvements are needed. Summary of the Invention

[0004] To address one or more problems in the existing technology, the main objective of this application is to provide a method and system for dynamic prediction and space optimization of warehouse inventory based on RFID and artificial intelligence.

[0005] To achieve the aforementioned objectives, this application proposes a method for dynamic prediction and space optimization of warehouse inventory based on RFID and artificial intelligence, the method comprising:

[0006] Real-time acquisition of sensing data from each inventory unit collected by the RFID sensing array;

[0007] Based on the collected sensing data, a spatiotemporal correlation model describing the dynamic relationship between inventory units and storage space is constructed.

[0008] The spatiotemporal correlation model is used for predictive analysis, and based on the results of the predictive analysis, inventory prediction information and spatial congestion prediction information for future periods are obtained.

[0009] Based on the inventory distribution change data and the spatial congestion distribution change data, the linkage between the inventory distribution change data and the spatial congestion change data is analyzed to identify conflict sources. The conflict sources are used to determine the storage locations or storage units that have a causal impact on the spatial congestion changes.

[0010] Based on the conflict source, perform a pre-adjustment operation on the inventory unit corresponding to the conflict source before congestion occurs;

[0011] Based on the results of the pre-adjustment operation, the actual operation data after pre-adjustment is collected through the RFID sensing array and used to update the next predictive analysis and conflict source identification process.

[0012] This application also provides a warehouse inventory dynamic prediction and space optimization system based on RFID and artificial intelligence, including:

[0013] The acquisition module is used to acquire the sensing data of each inventory unit collected by the RFID sensing array in real time;

[0014] The building module is used to construct a spatiotemporal correlation model describing the dynamic relationship between inventory units and storage space based on the collected sensing data;

[0015] The predictive analysis module is used to perform predictive analysis on the spatiotemporal correlation model and obtain inventory prediction information and spatial congestion prediction information for future periods based on the results of the predictive analysis.

[0016] The identification module is used to analyze the linkage between the inventory distribution change data and the spatial congestion distribution change data based on the inventory distribution change data and the spatial congestion distribution change data, and to identify the conflict source. The conflict source is used to determine the storage location or storage unit that has a causal impact on the spatial congestion change.

[0017] The adjustment module is used to perform a pre-adjustment operation on the inventory unit corresponding to the conflict source before the congestion occurs;

[0018] The data acquisition module is used to collect pre-adjusted actual operation data through the RFID sensing array based on the pre-adjustment operation results, and to update the next predictive analysis and conflict source identification process.

[0019] This application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods described above.

[0020] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described above.

[0021] This application presents a method and system for dynamic prediction and spatial optimization of warehouse inventory based on RFID and artificial intelligence. By deeply integrating RFID sensing with artificial intelligence prediction, it achieves the identification of sources of warehouse congestion and proactive spatial optimization. First, it relies on RFID arrays to acquire multi-dimensional sensing data of inventory units in real time, overcoming the blind spots of traditional static layouts in the face of dynamic changes. Second, the constructed spatiotemporal correlation model unifies the expression of inventory distribution and channel status, and combined with a multi-task prediction framework, simultaneously obtains future inventory changes and congestion risks, preventing a disconnect between inventory management and path scheduling decisions. Based on this, by analyzing the temporal linkage between inventory changes and congestion evolution, it accurately locates conflict sources with causal influence, changing the previous approach of passively responding to or misjudging the source after congestion occurs. For the identified conflict sources, pre-adjustment operations are performed before congestion forms, resolving potential problems in their infancy and significantly reducing channel blockage and waste of handling resources. Finally, using closed-loop feedback from actual operational data, the prediction model and identification rules are continuously optimized, enabling the system to adapt to the dynamic changes in the warehouse scenario. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating a method for dynamic prediction and space optimization of warehouse inventory based on RFID and artificial intelligence, according to an embodiment of this application.

[0023] Figure 2 This is a flowchart illustrating another embodiment of the RFID and artificial intelligence-based method for dynamic prediction and space optimization of warehouse inventory.

[0024] Figure 3 This is a schematic block diagram of a warehouse inventory dynamic prediction and space optimization system based on RFID and artificial intelligence according to an embodiment of this application.

[0025] Figure 4 This is a schematic block diagram of the structure of a computer device according to an embodiment of this application;

[0026] Figure 5 This is a schematic diagram showing the correspondence between time slice allocation and cargo location distance in one embodiment of this application.

[0027] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0029] Reference Figure 1 This application provides a method for dynamic prediction and space optimization of warehouse inventory based on RFID and artificial intelligence. The method includes:

[0030] S1. Real-time acquisition of sensing data from each inventory unit collected by the RFID sensing array;

[0031] S2. Based on the collected sensing data, construct a spatiotemporal correlation model describing the dynamic relationship between inventory units and storage space;

[0032] S3. Perform predictive analysis on the spatiotemporal correlation model, and obtain inventory prediction information and spatial congestion prediction information for future periods based on the results of the predictive analysis.

[0033] S4. Based on the inventory distribution change data and the spatial congestion distribution change data, analyze the linkage between the inventory distribution change data and the spatial congestion distribution change data, and identify the conflict source. The conflict source is used to determine the storage location or storage unit that has a causal impact on the spatial congestion change.

[0034] S5. Based on the conflict source, perform a pre-adjustment operation on the inventory unit corresponding to the conflict source before congestion occurs;

[0035] S6. Based on the pre-adjustment operation results, collect the pre-adjusted actual operation data through the RFID sensing array to update the next predictive analysis and conflict source identification process.

[0036] As described in steps S1-S3 above, step one involves deploying RFID reader arrays at the entrances and exits, aisle walkways, and key nodes of the storage area. When forklifts or AGVs transport tagged inventory units, the readers automatically collect the tag information. In addition to reading the identification code, this step also records multi-dimensional data such as the tag's signal strength, phase difference, and angle of arrival. This data reflects the current three-dimensional spatial position of the inventory unit and whether it is being stacked, squeezed, or tilted. Through continuous collection, the movement trajectory and physical state changes of each inventory unit on the time axis can be obtained. This provides a richer and more real-time information foundation than traditional barcode scanning, enabling the system to dynamically perceive the actual occupancy of the storage space, rather than relying solely on pre-entered static layouts. Step two, based on the perceived data obtained in the previous step, constructs a spatiotemporal correlation model that describes the dynamic relationship between inventory units and storage space. This model can be viewed as a network graph that changes over time. The nodes in the graph include two types of entities: one is the location node, representing various storage locations in the warehouse; the other is the inventory unit node, representing each pallet or bin. The edges in the graph are used to depict the relationships between different nodes, such as which storage location a certain inventory unit currently occupies, whether there is a direct passage connection between two storage locations, or whether two inventory units are frequently dispatched simultaneously in historical orders. This model evolves continuously as RFID data is updated. For example, when an inventory unit is moved, its occupancy edge with the old storage location disappears, and an occupancy edge with the new storage location is established. Through this structured representation, the originally scattered sensing data is organized into an organic whole, providing a clear data entry point for intelligent prediction. Step three involves inputting the historical snapshot sequence of the above spatiotemporal correlation model into a pre-trained artificial intelligence model. This model typically employs a multi-task learning architecture, with a shared underlying network for extracting common spatiotemporal features, and then branching into two independent output branches. One branch outputs the inventory quantity change curves for each storage location over a future period, i.e., inventory prediction information; the other branch outputs the operational access probability of each passage or node within the same future period, i.e., spatial congestion prediction information. The two branches run simultaneously, sharing the same input, thus maintaining inherent consistency in their output information. For example, if the model predicts a sharp drop in inventory at a certain storage location, the probability of congestion in the corresponding outbound channel will often increase simultaneously. The core value of this step lies in unifying the previously separate decision-making dimensions of inventory management and channel scheduling within a single predictive framework, thus avoiding the contradictions or lags that might result from separate predictions.

[0037] As described in steps S4-S6 above, after obtaining inventory forecasts and spatial congestion forecasts in step four, the next step is to analyze the linkage between the two to identify conflict sources. Conflict sources refer to storage locations or storage units whose inventory changes have a causal impact on spatial congestion changes. In simpler terms, it's about identifying which storage locations' outbound activities will trigger a domino effect, causing a series of aisle congestions. To distinguish between genuine causal relationships and accidental correlations, this step calculates the correlation coefficient between the inventory change timeline of each storage location and the congestion change timeline of each aisle at different time offsets. If it is found that an inventory reduction event in a certain storage location always precedes an aisle congestion worsening event, and the lead time is stable within a specific range, then that storage location is likely a conflict source. This identification method is much more accurate than simply looking at which area is most congested because it directly points to the source of the problem, rather than the surface manifestation of congestion. Step five involves pre-adjusting the corresponding storage units based on the identified conflict sources before congestion actually occurs. The specific form of the pre-adjustment operation can be flexibly selected according to the actual situation of the warehouse. For example, if there is a vacant target location at the location of the conflict source, the inventory unit can be directly moved to a lower-risk area away from the main aisle. If there is no vacant location available for transfer, or if the inventory unit is too bulky and inconvenient to move, other items with strong access coupling to the inventory unit can be moved to guide the workflow around the congested area by dispersing related inventory. The essence of pre-adjustment is to resolve future congestion in its early stages, rather than waiting until forklifts are stuck in the aisle before temporarily changing routes. After the pre-adjustment operation in step six is ​​completed, the warehouse continues normal operations. At this time, the RFID sensing array continuously collects real data such as actual outbound records, passage time, and congestion status. This data is compared with the prediction results in step three, and the deviation between the two is calculated. A large deviation indicates that the current prediction model or conflict source identification rules are inaccurate. Therefore, these deviations are used as feedback signals to update the parameters of the artificial intelligence model or adjust the judgment threshold of the linkage analysis. After multiple such closed-loop iterations, the entire method will gradually adapt to the warehouse's unique operating rhythm and order patterns, the prediction will become more and more accurate, and the effect of pre-adjustment will become better and better.

[0038] As described above, by deeply integrating RFID sensing with artificial intelligence prediction, the identification and proactive spatial optimization of warehouse congestion sources are achieved. First, relying on RFID arrays to acquire multi-dimensional sensing data of inventory units in real time overcomes the blind spots of traditional static layouts in dealing with dynamic changes. Second, the constructed spatiotemporal correlation model unifies the expression of inventory distribution and channel status, and combined with a multi-task prediction framework, simultaneously obtains future inventory changes and congestion risks, preventing a disconnect between inventory management and path scheduling decisions. Based on this, by analyzing the temporal linkage between inventory changes and congestion evolution, the system accurately locates conflict sources with causal influence, changing the previous approach of passively responding to congestion or misjudging the source after it occurs. For identified conflict sources, pre-adjustment operations are performed before congestion forms, resolving potential problems in their infancy and significantly reducing channel blockage and waste of handling resources. Finally, using closed-loop feedback from actual operational data, the prediction model and identification rules are continuously optimized, enabling the system to adapt to the dynamic changes in the warehousing scenario.

[0039] Reference Figure 2 In one embodiment, the step of performing predictive analysis on the spatiotemporal correlation model and obtaining future inventory forecast information and spatial congestion forecast information based on the results of the predictive analysis includes:

[0040] S31. Obtain the snapshot sequence of the spatiotemporal correlation model at the current time and multiple consecutive historical times, and organize the snapshot sequence of the current time and multiple consecutive historical times into time-series input data;

[0041] S32. Construct a multi-task prediction model, wherein the multi-task prediction model includes a shared feature extraction network and two task output branches respectively connected to the shared feature extraction network;

[0042] S33. Input the time-series input data into the shared feature extraction network to extract spatiotemporal feature representations;

[0043] S34. The spatiotemporal feature representation is simultaneously sent to two task output branches. The first task output branch is used to generate the predicted inventory quantity of each storage location at each time point in the future period, forming the inventory prediction information. The second task output branch is used to generate the predicted operation pass probability of each passage node in the storage channel at each time point in the future period, forming the spatial congestion prediction information.

[0044] As described above, firstly, snapshots of the model state at the current moment and multiple consecutive past moments are extracted from the pre-constructed spatiotemporal correlation model. These snapshots record the occupancy relationship between inventory units and storage locations, the connectivity between storage locations, and the real-time passage status of each channel at different points in time. These snapshots are then organized into a time-series input sequence in chronological order. The reason for needing snapshots from multiple consecutive moments instead of just the current state is that inventory changes and congestion evolution have significant time dependencies, and static information at a single moment cannot reflect dynamic characteristics such as inventory frequency and handling rhythm. With this time-series sequence, the artificial intelligence model can learn the sequence and trends of events. Next, a multi-task prediction model is pre-constructed. Structurally, this model includes a shared feature extraction network and two independent task output branches. The shared feature extraction network is located at the bottom layer of the model and is responsible for extracting general spatiotemporal feature representations from the input data, such as which storage locations have recently experienced frequent outbound shipments and which channels have seen a decrease in passage speed. The two task output branches are connected to this shared network, with one branch focusing on the inventory prediction task and the other on the congestion prediction task. This shared underlying network design is not arbitrary, but based on an observation: there is an inherent coupling relationship between changes in inventory distribution and changes in channel congestion. For example, concentrated outflows from a particular storage location often increase the traffic pressure in that channel. Therefore, the underlying feature extraction can be shared, eliminating the need to train two independent models for the two tasks. The organized time-series input data is fed into the shared feature extraction network. This network typically employs a structure capable of processing sequential data, such as an architecture based on recurrent neural networks or attention mechanisms. After computation, the network outputs a compact spatiotemporal feature representation. This representation has condensed and encoded the key information in the input sequence, including both the current state of each inventory unit and storage location, and implicitly its patterns of change over a past period. In other words, this feature representation is the common foundation for the subsequent two prediction tasks. The spatiotemporal feature representation obtained in the previous step is simultaneously fed into the output branches of the two tasks. The first branch is responsible for generating the predicted inventory quantity for each storage location at each point in time within the future period, forming inventory prediction information. For example, for a specific storage location, this branch can output the expected inventory quantity curve every five minutes within the next hour. The second branch is responsible for generating the predicted operational passability probability of each passage node in the warehouse aisle at each point in time within a future period, forming spatial congestion prediction information. For example, for a certain intersection, this branch can output the probability value that the AGV can pass smoothly within each five-minute interval in the next hour. The two branches share the same input features and run simultaneously, so the output inventory prediction and congestion prediction naturally maintain consistency in time and space. For example, if the model predicts that the inventory of a certain storage location will drop sharply in ten minutes, then the congestion probability of the corresponding aisle at that storage location will also increase accordingly at the same time.

[0045] It is worth noting that the multi-task prediction model needs to be trained before practical application. The training data comes from historical operational data recorded by the warehouse management system over a period of time, as well as synchronously collected RFID sensing data. By processing this data, snapshots of the spatiotemporal correlation model at each moment can be obtained. For each historical moment, the model input is a sequence of snapshots of the current moment and several consecutive previous moments, and the output labels are two: one is the actual inventory change of each location in future time periods, and the other is the actual operational passage probability or congestion marker for each channel in future time periods. During training, the model continuously adjusts the parameters in the shared feature extraction network and the output branches of the two tasks by comparing the differences between its own prediction results and the actual labels, until the prediction accuracy reaches the preset requirements. After such training, the model can learn the inherent coupling law between changes in inventory distribution and the evolution of channel congestion. It should be noted that the above training process can be completed using the backpropagation algorithm and stochastic gradient descent optimizer commonly used in this field. The specific implementation method can be selected according to the scale and characteristics of the warehouse data, which will not be elaborated here.

[0046] In one embodiment, the step of analyzing the correlation between inventory distribution change data and spatial congestion distribution change data to identify conflict sources includes:

[0047] The inventory distribution change data is decomposed into corresponding inventory change time series according to each storage location, and the spatial congestion distribution change data is decomposed into corresponding congestion change time series according to each channel.

[0048] For the inventory change time series of each storage location and the congestion change time series of each channel, calculate the correlation coefficient between the two at different time offsets, and record the maximum correlation coefficient and the time offset corresponding to the maximum correlation coefficient.

[0049] The location-channel pairs with the maximum correlation coefficient exceeding the preset threshold are selected. Based on the selection results, the locations in these location-channel pairs are marked as candidate conflict sources.

[0050] For each candidate conflict source, obtain the time offset recorded when the candidate conflict source is associated with each channel, and determine whether the candidate conflict source meets the retention condition based on all the time offsets associated with the candidate conflict source.

[0051] If all time offsets associated with the candidate conflict source are positive, then it is retained; otherwise, it is discarded.

[0052] Based on the judgment results, the remaining candidate conflict sources are output as the final identified conflict sources.

[0053] As mentioned above, after obtaining inventory and spatial congestion forecasts, the next key task is to identify the truly causal sources of conflict. This process is not simply comparing which location is most congested, but rather using a series of time-series analyses to quantify and compare the relationship between inventory changes and channel congestion over time. First, the inventory forecast information is broken down by individual storage locations, with each location corresponding to an inventory change curve over time. Similarly, the spatial congestion forecast information is broken down by individual channels, with each channel corresponding to a congestion change curve over time. In this way, the original overall forecast result is transformed into multiple independent time-series signals, facilitating subsequent analysis of the correlation between each storage location and each channel. Next, for each storage location's inventory change curve and each channel's congestion change curve, the correlation coefficient between the two is calculated at different time offsets. Simply put, a time offset involves shifting one curve forward or backward on the time axis by a certain period and then calculating its similarity to another curve. For example, the inventory change curve can be shifted forward by five minutes to see if it better matches the congestion change curve. By trying multiple different offsets, we can find the offset that makes the two curves most similar and the corresponding correlation coefficient. This maximum correlation coefficient reflects the strength of the association between the storage location and the aisle, while its corresponding time offset reveals the order in which they occur. If the maximum correlation coefficient occurs when inventory changes precede congestion changes, such as when aisle congestion worsens five minutes after inventory decreases, then there is reason to believe that inventory changes may be the cause of congestion. We set a correlation coefficient threshold and only retain storage location-aisle pairs whose maximum correlation coefficient exceeds this threshold. These retained pairs indicate a relatively significant time correlation between storage locations and aisles, warranting further investigation. We extract the storage locations from all retained pairs and mark them as candidate conflict sources. At this point, the candidate conflict sources only indicate a statistically strong correlation between inventory changes at these storage locations and congestion in certain aisles, but the direction of causality cannot be determined yet, as congestion changes may precede inventory changes, for example, the aisle may become congested first, preventing forklifts from approaching the storage location, thus causing inventory buildup. For each candidate conflict source, we examine the time offsets corresponding to all the aisles associated with it. These offsets are recorded from the previous calculation, and each offset represents the lead or lag relationship when the maximum correlation coefficient between the storage location and the channel occurs. If all time offsets associated with a candidate conflict source are positive, it means that the inventory change of that storage location always leads the congestion changes of the channels it affects, with no reverse trend. Conversely, if any offset is zero or negative, it indicates that the storage location and at least one channel either change simultaneously or the congestion change precedes it. In this case, the storage location is unlikely to be a true conflict source because the causal relationship is not in the correct direction or there is a possibility of reverse causality.Based on the above judgment, only those candidate conflict sources with all associated time offsets being positive are retained, and these are output as the final identified conflict sources. These retained storage locations or inventory units are the congestion sources that have undergone rigorous causal direction testing. Subsequent pre-adjustment operations can then address them more specifically.

[0054] In one embodiment, the step of performing a pre-adjustment operation on the inventory unit corresponding to the conflict source before congestion occurs includes:

[0055] Determine whether the inventory unit corresponding to the conflict source meets the preset immovable condition;

[0056] If the immovable condition is met, then obtain the set of associated inventory units that have a strong access coupling relationship with the inventory unit corresponding to the conflict source, wherein the strong access coupling relationship is that the frequency of two inventory units being accessed simultaneously by the same outbound task in the historical access record exceeds a preset frequency threshold.

[0057] For each associated inventory unit in the set of associated inventory units, calculate the overlap of the retrieval paths between the associated inventory unit and the inventory unit corresponding to the conflict source.

[0058] Based on the calculation results, the associated inventory unit with the highest path overlap is selected and the associated inventory unit is transferred to the target storage location located in different access directions.

[0059] If the immovable condition is not met, a location transfer operation is directly performed on the inventory unit corresponding to the conflict source.

[0060] As mentioned above, after identifying the source of conflict, targeted pre-adjustment operations need to be performed based on the actual situation. Not all sources of conflict are suitable for direct movement; therefore, this method first determines whether the inventory unit corresponding to the conflict source meets the immovable condition. The immovable condition can include several typical situations: for example, there is no available space around the location where the inventory unit is located for relocation, or the inventory unit is too large to be moved in a single operation, or it is predicted that the inventory unit will continue to be a source of conflict for a long time in the future, and will quickly become a new source of congestion after being moved. The purpose of these judgments is to avoid wasting effort and prevent forced movement when costs are too high or the results are poor. If the judgment finds that the conflict source does indeed meet the above immovable conditions, then the conventional direct movement strategy is not applicable. At this time, a different approach is taken: instead of moving the conflict source itself, move other inventory units that are closely related to it. Specifically, it is necessary to first identify the related inventory units that have a strong access coupling relationship with the conflict source. A strong access coupling relationship refers to two inventory units that are frequently accessed by the same order or the same handling task in historical outbound records. For example, a certain accessory and its dedicated packaging box, or two products that are frequently shipped together. This coupling relationship can be obtained by statistically analyzing historical data. When the frequency of simultaneous retrieval exceeds a certain preset threshold, it is considered that there is a strong business connection between the two. After finding these related inventory units, one cannot arbitrarily choose to move one of them. It is also necessary to further calculate the retrieval path overlap between each related inventory unit and the conflict source. The retrieval path overlap can be understood as the proportion of the routes that the forklift or AGV needs to travel when the two inventory units are retrieved separately overlap. The higher the overlap, the more likely the positional relationship of the two inventory units is to cause congestion accumulation. Because if the conflict source has already caused congestion in a channel, and its strongly coupled inventory happens to be located in the same channel, then the user will repeatedly pass through the same congested area in order to retrieve both goods at the same time, and the problem will become more serious. Based on the calculated overlap, the related inventory unit with the highest overlap is selected and moved to a target storage location located in a different access direction from the conflict source. Different access directions mean that the new location after relocation will not share the same main channel or the same entrance / exit with the conflict source. In this way, when orders require the retrieval of both inventory units simultaneously, the users will have to travel in two different directions, distributing the traffic pressure that was originally concentrated on one aisle to multiple directions. Although the total walking distance may increase slightly, it avoids a single aisle becoming completely blocked due to the combined flow of traffic. If the source of the conflict does not meet the immovability requirement, then following conventional methods, the inventory unit can be directly moved to a more suitable location, such as an area near the exit with wide aisles, or an vacant storage space away from the predicted congestion hotspot. This step is simple and direct, and is the most efficient when conditions permit.

[0061] In one embodiment, after the step of analyzing the correlation between inventory distribution change data and spatial congestion distribution change data to identify conflict sources, the method further includes:

[0062] The RFID sensing array collects real-time data on the phase change of radio frequency signals around each inventory unit.

[0063] When the inventory of the inventory unit corresponding to the conflict source changes, the phase change data is tracked in space to generate a physical disturbance propagation path.

[0064] Obtain the location-channel pair corresponding to the conflict source, wherein the location in the location-channel pair is the location corresponding to the conflict source, and the channel in the location-channel pair is the channel whose correlation coefficient with the inventory change time series of the location exceeds a preset threshold in the correlation analysis;

[0065] Analyze the spatial order of the channels in the location-channel pair, and construct the linkage path of the conflict source based on the spatial order of the channels in the location-channel pair. The linkage path is used to characterize the spatial transmission relationship between the inventory distribution change data and the spatial congestion distribution change data.

[0066] The physical disturbance propagation path and the linkage path are spatially overlapped and compared.

[0067] If the overlap exceeds a preset threshold, the causality of the conflict source is confirmed to be valid, and the identification result of the conflict source is retained.

[0068] If the overlap is lower than a preset threshold, the conflict source is marked as a false conflict source and removed from the identification results.

[0069] As mentioned above, when an RFID reader reads a tag, in addition to acquiring the tag's identification code, it also generates a signal phase value. This phase value is highly sensitive to changes in the distance between the tag and the antenna, as well as the surrounding environment. When an inventory unit is removed or placed, it causes minute disturbances in the surrounding shelves, adjacent goods, and even electromagnetic waves in the air. This method records these minute physical disturbances by collecting real-time radio frequency signal phase change data around each inventory unit. This data is often discarded or ignored in conventional warehouse management, but here it is given a new purpose. When an inventory unit initially identified as a source of conflict undergoes an actual inventory change, such as when goods are removed, the system traces the continuous spatial propagation trajectory of the phase change data from that location. Because the disturbance spreads outwards like ripples in water, other inventory units along the path are also affected, generating phase fluctuations. By analyzing the order and spatial location of these fluctuations, a physical disturbance propagation path can be generated. This path is actually transmitted through physical media and is unaffected by order allocation or human scheduling, thus possessing high objectivity. Meanwhile, reviewing the previous correlation analysis results, identify all location-channel pairs corresponding to the source of conflict. The channels in these pairs are those statistically highly correlated with the inventory change timeline of the conflict source; simply put, these are the channels the statistical model considers easily affected by the conflict source. Connect the channels in these location-channel pairs in natural spatial order to form a linked path starting from the conflict source location, passing through each affected channel sequentially, and finally reaching the location of congestion. This path expresses the spatial transmission relationship of congestion as perceived by the statistical model. It describes two versions of the same event as the previously generated physical disturbance propagation path: one is a logical deduction based on statistical data, and the other is the actual propagation based on RFID physical sensing. Overlap the physical disturbance propagation path and the linked path on the same warehouse space map and calculate the degree of spatial overlap between the two paths. If the overlap is high, it indicates that the transmission relationship identified by the statistical model is consistent with the actual direction of physical disturbance propagation; the conflict source did indeed physically trigger a chain reaction, therefore its causality is valid, and this identification result can be retained. Conversely, if the overlap between two paths is very low, it means that the correlation seen in statistics may just be a coincidence. For example, the congestion of two different channels happens to coincide with the outbound time of the warehouse, but there is actually no physical propagation relationship. In this case, the source of conflict should be marked as false and removed from the identification results.

[0070] In one embodiment, the method further includes: when multiple conflict sources are identified, before performing a pre-adjustment operation on the inventory units corresponding to the conflict sources, the steps include:

[0071] Obtain the storage and retrieval operation paths of the inventory units corresponding to each conflict source;

[0072] Based on the storage and retrieval operation path, detect whether each storage and retrieval operation path passes through the same shared resource node. The shared resource node includes the warehouse exit, elevator, conveyor belt junction or alleyway intersection.

[0073] If a shared resource node is traversed by access paths of two or more conflict sources, the shared resource node is marked as a competing resource node, and the conflict sources that depend on the competing resource node are grouped into the same conflict source group.

[0074] As mentioned above, multiple conflict sources may exist simultaneously in a real-world warehousing environment. If these conflict sources are pre-adjusted indiscriminately, mutual interference is likely, especially when their operational paths pass through the same bottleneck location. Therefore, a pre-judgment step is added before performing specific pre-adjustment operations to identify which conflict sources have resource competition relationships. First, for each identified conflict source, the path that its corresponding inventory unit needs to traverse during normal storage and retrieval operations is obtained. These paths can be obtained from preset routes in the warehouse management system or historical handling trajectories. For example, when a pallet located deep in the shelving is retrieved, the forklift needs to traverse several main aisles, make turns, and may also take elevators or pass through conveyor belt junctions. Recording these paths yields the operational route for each conflict source. Next, these operational paths are compared to check if a shared resource node is being used simultaneously by two or more conflict sources. Shared resource nodes refer to locations in the warehouse that are likely to become bottlenecks, such as a single outbound exit, a limited number of elevators, the intersection of multiple conveyor belts, or an aisle intersection. These nodes have limited capacity, and they are prone to blocking when multiple tasks request them simultaneously. By comparing paths, it's possible to identify which conflict sources use the same exit or the same elevator. Once a shared resource node is found to be traversed by the work paths of multiple conflict sources, that node is marked as a competing resource node, and the conflict sources that depend on that node are grouped into the same conflict source group. The significance of this grouping operation is that it aggregates multiple conflict sources that originally seemed independent according to their underlying resource competition relationships. For example, conflict sources from three different storage locations, although geographically dispersed, all need to use the same elevator when exiting the warehouse; therefore, they belong to the same conflict source group.

[0075] In one embodiment, after the step of marking the shared resource node as a competing resource node and grouping the competing resource nodes into the same conflict source group if a shared resource node is traversed by access operation paths of two or more conflict sources, the step further includes:

[0076] For each conflict source in the conflict source group, target storage locations that meet the conditions are selected from the candidate target storage location set. The condition is that when the inventory unit corresponding to the conflict source is transferred to the target storage location, the new storage and retrieval operation path of the inventory unit does not pass through the competing resource node.

[0077] If a target storage location that meets the conditions exists, the inventory unit corresponding to the source of conflict will be transferred to the target storage location.

[0078] If no target storage location meets the conditions, the inventory unit corresponding to the conflict source will be transferred to the target storage location furthest from the competing resource node.

[0079] As mentioned above, after grouping multiple conflict sources with resource competition into the same conflict source group, targeted pre-adjustment is needed for these conflict sources. While conventional queuing methods can prevent resources from being requested by multiple tasks simultaneously, they do not fundamentally alleviate dependence on competing resources. This method attempts to adjust the target locations of these conflict sources so that they no longer depend on the same competing resource node after pre-adjustment. For each conflict source in the conflict source group, it first filters through its set of candidate target storage locations. The filtering condition is: when the inventory unit corresponding to the conflict source is moved to a candidate storage location, the new operational path formed when that inventory unit is accessed in the future will no longer pass through the previously marked competing resource node. In other words, it finds a new storage location that bypasses the bottleneck. For example, if the outbound path of a conflict source originally had to pass through an elevator, if a new storage location can be found so that the outbound path from that new storage location can take the stairs or another less congested elevator, then this new storage location meets the condition. This filtering process relies on warehouse layout data and path planning algorithms and can be completed automatically based on existing navigation maps. If at least one target storage location meets the above conditions, it means that the source of conflict can completely avoid dependence on competing resources by changing its storage location. In this case, the inventory unit corresponding to the source of conflict is moved to one of the storage locations that meets the conditions. The direct effect of this is that after pre-adjustment, the source of conflict no longer competes with other sources of conflict within the group for the same bottleneck resource, which is equivalent to eliminating the competition relationship from the source. Multiple sources of conflict that originally needed to queue or coordinate scheduling no longer affect each other after this process, and the complexity of the entire source of conflict group is reduced. If, after screening, it is found that all candidate target storage locations cannot make the new storage and retrieval operation path avoid the competing resource node, it means that the competing resource is irreplaceable in the warehouse layout, such as the warehouse having only one exit, or all paths to a certain area having to pass through the same elevator. In this case, as a second-best option, the inventory unit corresponding to the source of conflict is moved to the target storage location farthest from the competing resource node. Farthest distance means that the path length from the new storage location to the competing resource node is the longest, or the frequency of storage and retrieval tasks starting from the new storage location occupying the competing resource is minimized. For example, moving goods to the farthest corner of the warehouse will still use the same exit gate when goods are shipped out, but because the walking distance is longer, the frequency of shipments per unit time will naturally decrease, and the pressure on the exit gate will be reduced accordingly. The essence of this strategy is to trade spatial distance for a reduction in resource occupancy, which is a suboptimal choice when it cannot be completely avoided.

[0080] For example, in actual warehousing operations, multiple conflict sources often coexist. For instance, a warehouse may have two different storage areas located on different floors, but their outbound operations must pass through the same elevator. Conventionally, if both storage areas are identified as conflict sources, the system might perform pre-adjustment operations on them separately, such as moving goods from one storage area to a nearby location. However, this independent handling approach can introduce new problems: the moved goods may still need to pass through the same elevator, or the elevator may be occupied during the movement, preventing pre-adjustment of the other conflict source, or even unintentionally affecting the coupled inventory of the other conflict source due to the movement of a single item. In short, handling them separately may actually worsen the situation. For scenarios where multiple conflict sources compete for the same shared resource, this method assumes that two conflict sources exist simultaneously in a warehouse, denoted as conflict source A and conflict source B. Conflict source A is located in the second-floor east area, and its storage and retrieval path requires passing through elevator number one; conflict source B is located in the second-floor west area, and its storage and retrieval path also requires passing through elevator number one. In addition, the warehouse has another elevator, number two, but it is relatively far from both conflict sources. First, the system obtains the access paths for conflict sources A and B and compares them. The comparison reveals that both paths pass through Elevator 1, a typical shared resource node with limited capacity, unable to handle the high-frequency use of multiple handling tasks simultaneously. Therefore, the system marks Elevator 1 as a competing resource node and groups conflict sources A and B into the same conflict source group. This step demonstrates that the system recognizes these two conflict sources cannot be handled independently and must be considered as a whole. Next, for each conflict source in this group, the system searches for a new target storage location. The goal is to find a location where, after transferring the corresponding inventory unit, the new access path no longer passes through Elevator 1. For conflict source A, the system searches among candidate storage locations on its floor and finds a vacant area in the southeast corner of the second floor. Goods from this area can be directly accessed to the first floor via stairs, completely eliminating the need for Elevator 1. Therefore, this target storage location meets the criteria, and the system transfers the goods from conflict source A to this new location. After the transfer, conflict source A no longer relies on elevator number one, effectively withdrawing from this conflict source group. For conflict source B, the system's search of its area revealed that all possible target storage locations ultimately converge on elevator number one, as all passageways in the second-floor west section eventually connect to the corridor where elevator number one is located. In other words, as long as the goods remain in the second-floor west section, this elevator must be used. In this situation, the goods from conflict source B are moved to the target storage location furthest from elevator number one, which is the furthest corner of the second-floor west section.The effect of this approach is that although elevator number one is still used for outbound shipments, the increased distance from the furthest point to the elevator naturally reduces the frequency of outbound shipments per unit time, thus decreasing the pressure on elevator number one. In other words, conflict source A has completely eliminated its dependence on competing resources, while conflict source B, although still dependent, has reduced its dependence. The two conflict sources, which previously required complex coordination, can now be handled separately according to their respective circumstances, overall reducing the risk of congestion on elevator number one.

[0081] In one feasible embodiment, after grouping conflict sources that depend on the same competing resource node into the same conflict source group, the method further includes:

[0082] Based on the predicted congestion severity or operational urgency of each conflict source in the conflict source group, the available time resources of the competing resource nodes in the future period are divided into multiple time slices, and each conflict source is allocated one time slice.

[0083] For each conflict source, target storage locations that meet the location transfer conditions are selected from the candidate target storage location set. The location transfer conditions include: after the inventory unit corresponding to the conflict source is transferred to the target storage location, the expected travel time from the target storage location to the competing resource node falls within the time slice allocated to the conflict source.

[0084] If all conflict sources find a target storage location that meets the aforementioned location transfer conditions, then the transfer operation is performed according to the screening results.

[0085] If at least one conflict source cannot find a target location that meets the location transfer conditions, the time slice length or start and end times of each conflict source are adjusted, and the target location is screened again until all conflict sources find a feasible location or the preset number of iterations is reached.

[0086] As mentioned above, in warehousing operations, when multiple conflict sources rely on the same unavoidable shared resource—for example, a warehouse with only one outbound exit, or all handling tasks in a certain area must pass through the same elevator—the conventional pre-adjustment strategy usually involves moving some conflict source inventory units to locations farther away from the resource to reduce their resource occupancy frequency. However, this approach ignores the differences in resource demand among different conflict sources. For instance, if a high-frequency outbound conflict source is moved to a distant location, although the individual occupancy time increases, its total occupancy frequency remains high. This could lead to prolonged occupation of the resource passage due to increased walking distance, severely delaying the operations of other conflict sources. Another common practice is to queue according to a first-come, first-served principle, but this cannot resolve congestion in advance and only allows for passive waiting.

[0087] refer to Figure 5To address the aforementioned issues, this embodiment assumes that the only exit of a warehouse is marked as a competing resource node. Three conflict sources, A, B, and C, are grouped into the same conflict source group, with predicted congestion severity levels of high, medium, and low, respectively. The available time resources at the exit of the warehouse for the next hour can be viewed as a continuous time axis. The first step is to divide the next hour's time axis into three time slices of unequal length based on the congestion severity of each conflict source. For example, conflict source A, with the highest severity, receives a 30-minute time slice, conflict source B receives 20 minutes, and conflict source C receives 10 minutes. These time slices can be consecutive or have different start and end times set according to urgency. The second step is to select a suitable location from the candidate target storage locations for each conflict source. The selection criteria are not only that the path does not pass through the exit (which is unavoidable here), but also that a time constraint is met: the walking time required to reach the exit from the target storage location via the normal handling path, plus the estimated dwell time for completing the operation at the exit, should ensure that the time period occupied by the conflict source at the exit falls entirely within its allocated time slice. For example, conflict source A, due to its longer and earlier time slice, can choose a storage location closer to the exit, thus minimizing travel time and fully utilizing the first half of its time slice for multiple outbound operations. Conflict source C, with a shorter and potentially later time slice, needs to choose a storage location farther from the exit, naturally filling its narrow time slice with its longer travel time and avoiding occupying other time slices due to early arrival. Thirdly, if all conflict sources find target storage locations that satisfy the time slice constraints after filtering, the transfer operation is executed according to the filtering results. If a conflict source cannot find a suitable storage location, for example, because its time slice is too short while the travel time to the exit from all candidate storage locations is too long, preventing a complete round-trip operation within its time slice, the system will dynamically adjust the time slice allocation. For example, it may appropriately extend the time slice of that conflict source while correspondingly shortening the time slices of other conflict sources, or adjust the order of the time slices, and then re-filter the storage locations. This process can be repeated multiple times until all conflict sources find feasible storage locations, or the preset iteration limit is reached. If the problem persists, then revert to the furthest distance transfer strategy as a backup plan.

[0088] Reference Figure 3 This application also provides a warehouse inventory dynamic prediction and space optimization system based on RFID and artificial intelligence, including:

[0089] Module 1 is used to acquire the sensing data of each inventory unit collected by the RFID sensing array in real time;

[0090] Module 2 is used to construct a spatiotemporal correlation model describing the dynamic relationship between inventory units and storage space based on the collected sensing data;

[0091] Predictive analysis module 3 is used to perform predictive analysis on the spatiotemporal correlation model, and obtain inventory prediction information and spatial congestion prediction information for future periods based on the results of the predictive analysis.

[0092] The identification module 4 is used to analyze the linkage between the inventory distribution change data and the spatial congestion distribution change data based on the inventory distribution change data and the spatial congestion distribution change data, and to identify the conflict source. The conflict source is used to determine the storage location or storage unit that has a causal impact on the spatial congestion change.

[0093] Adjustment module 5 is used to perform a pre-adjustment operation on the inventory unit corresponding to the conflict source before congestion occurs, based on the conflict source.

[0094] The acquisition module 6 is used to acquire pre-adjusted actual operation data through the RFID sensing array based on the pre-adjustment operation results, and to update the next predictive analysis and conflict source identification process.

[0095] As described above, it is understood that each component of the RFID and artificial intelligence-based warehouse inventory dynamic prediction and spatial optimization system proposed in this application can realize the function of any of the RFID and artificial intelligence-based warehouse inventory dynamic prediction and spatial optimization methods described above, and the specific structure will not be repeated.

[0096] Reference Figure 4 This application also provides a computer device, which may be a server, and its internal structure may be as follows: Figure 4 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores monitoring data and other data. The network interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for dynamic prediction and space optimization of warehouse inventory based on RFID and artificial intelligence.

[0097] The processor described above executes the RFID and AI-based dynamic inventory prediction and space optimization method, including: acquiring real-time sensing data of each inventory unit collected by the RFID sensing array; constructing a spatiotemporal correlation model describing the dynamic relationship between inventory units and warehouse space based on the collected sensing data; performing predictive analysis on the spatiotemporal correlation model, and obtaining inventory prediction information and space congestion prediction information for future periods based on the results of the predictive analysis; analyzing the linkage relationship between the inventory distribution change data and the space congestion distribution change data based on the inventory distribution change data and the space congestion distribution change data, and identifying conflict sources, wherein the conflict sources are used to determine the storage locations or inventory units that have a causal impact on the changes in inventory on the changes in space congestion; performing pre-adjustment operations on the inventory units corresponding to the conflict sources before congestion occurs; and collecting the pre-adjusted actual operation data through the RFID sensing array based on the results of the pre-adjustment operations, which is used to update the next predictive analysis and conflict source identification process.

[0098] One embodiment of this application also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements a method for dynamic prediction and space optimization of warehouse inventory based on RFID and artificial intelligence, including the following steps: acquiring real-time sensing data of each inventory unit collected by an RFID sensing array; constructing a spatiotemporal correlation model describing the dynamic relationship between inventory units and warehouse space based on the collected sensing data; performing predictive analysis on the spatiotemporal correlation model, and obtaining inventory prediction information and space congestion prediction information for future periods based on the results of the predictive analysis; analyzing the linkage relationship between the inventory distribution change data and the space congestion distribution change data based on the inventory distribution change data and the space congestion distribution change data, and identifying conflict sources, wherein the conflict sources are used to determine the storage locations or inventory units that have a causal impact on the space congestion change; performing a pre-adjustment operation on the inventory unit corresponding to the conflict source before congestion occurs; and collecting the pre-adjusted actual operation data through the RFID sensing array based on the pre-adjustment operation results, which is used to update the next predictive analysis and conflict source identification process.

[0099] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in this application and in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0100] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0101] The above description is only a preferred embodiment of this application and does not limit the patent scope of this application. Any equivalent structural or procedural changes made based on the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for dynamic prediction and spatial optimization of warehouse inventory based on RFID and artificial intelligence, characterized in that, The method includes: Real-time acquisition of sensing data from each inventory unit collected by the RFID sensing array; Based on the collected sensing data, a spatiotemporal correlation model describing the dynamic relationship between inventory units and storage space is constructed. The spatiotemporal correlation model is used for predictive analysis, and based on the results of the predictive analysis, inventory prediction information and spatial congestion prediction information for future periods are obtained. Based on the inventory distribution change data and the spatial congestion distribution change data, the linkage between the inventory distribution change data and the spatial congestion change data is analyzed to identify conflict sources. The conflict sources are used to determine the storage locations or storage units that have a causal impact on the spatial congestion changes. Based on the conflict source, perform a pre-adjustment operation on the inventory unit corresponding to the conflict source before congestion occurs; Based on the results of the pre-adjustment operation, the actual operation data after pre-adjustment is collected through the RFID sensing array and used to update the next predictive analysis and conflict source identification process.

2. The method for dynamic prediction and spatial optimization of warehouse inventory based on RFID and artificial intelligence according to claim 1, characterized in that, The steps of performing predictive analysis on the spatiotemporal correlation model and obtaining future inventory forecast information and spatial congestion forecast information based on the results of the predictive analysis include: Obtain the snapshot sequence of the spatiotemporal correlation model at the current time and at multiple consecutive historical times, and organize the snapshot sequence of the current time and multiple consecutive historical times into time-series input data; A multi-task prediction model is constructed, which includes a shared feature extraction network and two task output branches respectively connected to the shared feature extraction network. The time-series input data is input into the shared feature extraction network to extract spatiotemporal feature representations; The spatiotemporal features are simultaneously fed into two task output branches. The first task output branch is used to generate the predicted inventory quantity of each storage location at each time point in the future period, forming the inventory prediction information. The second task output branch is used to generate the predicted operation pass probability of each passage node in the storage channel at each time point in the future period, forming the spatial congestion prediction information.

3. The method for dynamic prediction and spatial optimization of warehouse inventory based on RFID and artificial intelligence according to claim 1, characterized in that, The steps for analyzing the correlation between inventory distribution change data and spatial congestion distribution change data to identify conflict sources include: The inventory distribution change data is decomposed into corresponding inventory change time series according to each storage location, and the spatial congestion distribution change data is decomposed into corresponding congestion change time series according to each channel. For the inventory change time series of each storage location and the congestion change time series of each channel, calculate the correlation coefficient between the two at different time offsets, and record the maximum correlation coefficient and the time offset corresponding to the maximum correlation coefficient. The location-channel pairs with the maximum correlation coefficient exceeding the preset threshold are selected. Based on the selection results, the locations in these location-channel pairs are marked as candidate conflict sources. For each candidate conflict source, obtain the time offset recorded when the candidate conflict source is associated with each channel, and determine whether the candidate conflict source meets the retention condition based on all the time offsets associated with the candidate conflict source. If all time offsets associated with the candidate conflict source are positive, then it is retained; otherwise, it is discarded. Based on the judgment results, the remaining candidate conflict sources are output as the final identified conflict sources.

4. The method for dynamic prediction and spatial optimization of warehouse inventory based on RFID and artificial intelligence according to claim 1, characterized in that, The step of performing a pre-adjustment operation on the inventory unit corresponding to the conflict source before congestion occurs includes: Determine whether the inventory unit corresponding to the conflict source meets the preset immovable condition; If the immovable condition is met, then obtain the set of associated inventory units that have a strong access coupling relationship with the inventory unit corresponding to the conflict source, wherein the strong access coupling relationship is that the frequency of two inventory units being accessed simultaneously by the same outbound task in the historical access record exceeds a preset frequency threshold. For each associated inventory unit in the set of associated inventory units, calculate the overlap of the retrieval paths between the associated inventory unit and the inventory unit corresponding to the conflict source. Based on the calculation results, the associated inventory unit with the highest path overlap is selected and the associated inventory unit is transferred to the target storage location located in different access directions. If the immovable condition is not met, a location transfer operation is directly performed on the inventory unit corresponding to the conflict source.

5. The method for dynamic prediction and spatial optimization of warehouse inventory based on RFID and artificial intelligence according to claim 3, characterized in that, After the step of analyzing the correlation between inventory distribution change data and spatial congestion distribution change data to identify conflict sources, the method further includes: The RFID sensing array collects real-time data on the phase change of radio frequency signals around each inventory unit. When the inventory of the inventory unit corresponding to the conflict source changes, the phase change data is tracked in space to generate a physical disturbance propagation path. Obtain the location-channel pair corresponding to the conflict source, wherein the location in the location-channel pair is the location corresponding to the conflict source, and the channel in the location-channel pair is the channel whose correlation coefficient with the inventory change time series of the location exceeds a preset threshold in the correlation analysis; Analyze the spatial order of the channels in the location-channel pair, and construct the linkage path of the conflict source based on the spatial order of the channels in the location-channel pair. The linkage path is used to characterize the spatial transmission relationship between the inventory distribution change data and the spatial congestion distribution change data. The physical disturbance propagation path and the linkage path are spatially overlapped and compared. If the overlap exceeds a preset threshold, the causality of the conflict source is confirmed to be valid, and the identification result of the conflict source is retained. If the overlap is lower than a preset threshold, the conflict source is marked as a false conflict source and removed from the identification results.

6. The method for dynamic prediction and spatial optimization of warehouse inventory based on RFID and artificial intelligence according to claim 1, characterized in that, The method further includes, when multiple conflict sources are identified, prior to the step of performing a pre-adjustment operation on the inventory units corresponding to the conflict sources, the steps include: Obtain the storage and retrieval operation paths of the inventory units corresponding to each conflict source; Based on the storage and retrieval operation path, detect whether each storage and retrieval operation path passes through the same shared resource node. The shared resource node includes the warehouse exit, elevator, conveyor belt junction or alleyway intersection. If a shared resource node is traversed by access paths of two or more conflict sources, the shared resource node is marked as a competing resource node, and the conflict sources that depend on the competing resource node are grouped into the same conflict source group.

7. The method for dynamic prediction and spatial optimization of warehouse inventory based on RFID and artificial intelligence according to claim 6, characterized in that, Following the step of marking the shared resource node as a competing resource node and grouping the competing resource nodes that depend on it into the same competing resource group if the shared resource node is traversed by access operation paths from two or more conflict sources, the steps further include: For each conflict source in the conflict source group, target storage locations that meet the conditions are selected from the candidate target storage location set. The condition is that when the inventory unit corresponding to the conflict source is transferred to the target storage location, the new storage and retrieval operation path of the inventory unit does not pass through the competing resource node. If a target storage location that meets the conditions exists, the inventory unit corresponding to the source of conflict will be transferred to the target storage location. If no target storage location meets the conditions, the inventory unit corresponding to the conflict source will be transferred to the target storage location furthest from the competing resource node.

8. A warehouse inventory dynamic prediction and space optimization system based on RFID and artificial intelligence, characterized in that, include: The acquisition module is used to acquire the sensing data of each inventory unit collected by the RFID sensing array in real time; The building module is used to construct a spatiotemporal correlation model describing the dynamic relationship between inventory units and storage space based on the collected sensing data; The predictive analysis module is used to perform predictive analysis on the spatiotemporal correlation model and obtain inventory prediction information and spatial congestion prediction information for future periods based on the results of the predictive analysis. The identification module is used to analyze the linkage between the inventory distribution change data and the spatial congestion distribution change data based on the inventory distribution change data and the spatial congestion distribution change data, and to identify the conflict source. The conflict source is used to determine the storage location or storage unit that has a causal impact on the spatial congestion change. The adjustment module is used to perform a pre-adjustment operation on the inventory unit corresponding to the conflict source before the congestion occurs; The data acquisition module is used to collect pre-adjusted actual operation data through the RFID sensing array based on the pre-adjustment operation results, and to update the next predictive analysis and conflict source identification process.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images