Operation management method and system for big data intelligent warehouse

Abstract

The invention discloses an operation management method and system for a big data intelligent warehouse, and relates to the technical field of data processing and intelligent warehouse management, and the method comprises the steps: collecting multi-source heterogeneous data, dividing the multi-source heterogeneous data into grid units, fusing time, space and environment features, generating a spatio-temporal data matrix, inputting the spatio-temporal data matrix into a spatio-temporal diagram convolution network model, and obtaining a spatial-temporal diagram convolution network model; according to the method, the relevance among the grid units is analyzed through the multi-layer space-time convolution, the safety inventory prediction value of each area is obtained, the multi-source heterogeneous data is integrated, the relevance among the data is analyzed through the space-time diagram convolution network model, the safety inventory prediction value of each area can be accurately obtained, and the safety inventory prediction accuracy is improved. The accuracy of inventory management decisions is improved, the generative adversarial network model is utilized to deeply analyze the distribution characteristics of the abnormal candidate regions, and the abnormal regions and the deviation levels thereof are accurately positioned by calculating the deviation degree, so that the false alarm rate is greatly reduced, and the dependence on manual review is reduced.

Description

Technical Field

[0001] The present invention relates to the technical fields of data processing and intelligent warehouse management, and in particular to an operation management method and system for a big data intelligent warehouse. Background Art

[0002] With the rapid development of algorithm technology and the continuous progress of big data analysis and artificial intelligence technology, traditional warehouse management methods are gradually transforming towards intelligence and automation. Modern intelligent warehouses not only need to process massive amounts of data but also be able to respond quickly and accurately to market changes. Early warehouse management mainly relied on manual records, which was inefficient. Over time, the application of barcode scanning technology has greatly improved the speed and accuracy of data collection.

[0003] However, there are still some deficiencies in the existing technology. On the one hand, the ability to fuse multi-source heterogeneous data is limited, and it is impossible to effectively fuse multi-dimensional information such as geographical location, environmental monitoring, and historical inventory, resulting in inaccurate safety inventory prediction values. On the other hand, traditional warehouse management mechanisms usually lack the ability of self-learning and optimization, and cannot automatically adjust parameters according to deviation data in actual operation to improve performance. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an operation management method for a big data intelligent warehouse to solve the problems of insufficient fusion of multi-source heterogeneous data and inaccurate anomaly detection.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides an operation management method for a big data intelligent warehouse, which includes collecting multi-source heterogeneous data and dividing it into grid cells, and at the same time fusing time, space, and environmental features to generate a spatio-temporal data matrix;

[0008] Inputting the spatio-temporal data matrix into a spatio-temporal graph convolutional network model, and analyzing the correlation between grid cells through multi-layer spatio-temporal convolution to obtain the safety inventory prediction value of each region;

[0009] Using the IoT sensor network to obtain the current inventory of each grid cell, and comparing it with the safety inventory prediction value to obtain candidate abnormal regions;

[0010] Constructing a generative adversarial network model, analyzing the distribution characteristics of candidate abnormal regions in the space of the generative adversarial network model, and calculating the degree of deviation. When the deviation value exceeds the statistical threshold, outputting the coordinates of the abnormal region and the deviation level;

[0011] Start different response processes according to the deviation level, and input the deviation data in the deviation level into the feedback spatio-temporal graph convolutional network model to dynamically adjust the parameters and obtain an optimized spatio-temporal graph convolutional network model;

[0012] Based on the optimized spatio-temporal graph convolutional network model, calculate the optimal inventory level for each area, generate replenishment instructions, and generate an operation management report based on the replenishment instructions.

[0013] As a preferred solution of the operation management method of the big data intelligent warehouse described in the present invention, wherein: the multi-source heterogeneous data includes geographical location data, environmental monitoring data, and historical inventory data.

[0014] As a preferred solution of the operation management method of the big data intelligent warehouse described in the present invention, wherein: the steps of obtaining the safety inventory prediction value for each area are as follows,

[0015] Input the spatio-temporal data matrix into the spatio-temporal graph convolutional network model, extract the spatial correlation features between adjacent grids through the first layer of spatio-temporal convolution, and analyze the co-variation law of geographically adjacent areas;

[0016] Capture the dynamic trend features of each grid unit in the time series through the second layer of time convolution, and identify long-term and short-term fluctuation patterns;

[0017] Adopt an attention mechanism to dynamically adjust the weights of the spatial correlation features and dynamic trend features, and output the future safety inventory prediction value of each grid unit.

[0018] As a preferred solution of the operation management method of the big data intelligent warehouse described in the present invention, wherein: the steps of obtaining the abnormal candidate area are as follows,

[0019] Use the IoT sensor network to collect warehouse data in real time, generate the original inventory log, and eliminate interference through multi-read filtering and signal compensation methods to obtain inventory data;

[0020] Dynamically match the inventory data with the safety inventory prediction value obtained by the spatio-temporal graph convolutional network model through real-time streaming window calculation and dynamic time warping algorithm, and use the stream processing engine to mark the grid units exceeding the safety inventory threshold as initially screened anomalies;

[0021] Combine the initially screened anomalies with the rule engine, exclude false alarms through the context-aware rule chain, and use the density clustering algorithm combined with the dynamic neighborhood radius to merge the initially screened anomalies into abnormal candidate areas, and output the coordinates and abnormal types after cross-verification by multi-source sensors.

[0022] As a preferred solution of the operation management method of the big data intelligent warehouse described in the present invention, wherein: the steps of outputting the abnormal area coordinates and deviation level include the following,

[0023] Based on a generator and a discriminator, a conditional generative adversarial network model is constructed through contrastive learning supervision and dual-path adversarial training;

[0024] Based on the generative adversarial network model, through the spatio-temporal graph attention mechanism, the spatio-temporal features of the abnormal candidate regions are mapped and matched with the features of the normal distribution, and the spatial deviation degree is calculated based on the triplet network;

[0025] A deviation value is generated based on the spatial deviation degree. When the deviation value exceeds the statistical threshold, the grid coordinates are located through inverse mapping, and the coordinates of the abnormal region and the deviation level are output according to the overrun ratio.

[0026] As a preferred solution of the operation management method of the big data intelligent warehouse described in the present invention, wherein: the deviation data is fed back to the spatio-temporal graph convolutional network model, the parameters are dynamically adjusted, and an optimized spatio-temporal graph convolutional network model is obtained, which specifically includes the following steps,

[0027] After the deviation data is aligned based on the Flink time window, the periodic and trend features are extracted through Prophet residual decomposition, and the gated attention mechanism is used to feed back to the spatio-temporal graph convolutional network model;

[0028] Based on the deviation data, the spatio-temporal graph convolutional network model updates the parameter update direction through EWC, combines mini-batch incremental training to dynamically adjust the spatio-temporal attention weights, separates the EWC importance mask and the momentum term in gradient clipping, and uses the AdamW optimizer to update the convolutional kernel parameters to generate an optimized spatio-temporal graph convolutional network model.

[0029] As a preferred solution of the operation management method of the big data intelligent warehouse described in the present invention, wherein: based on the optimized spatio-temporal graph convolutional network model, the optimal inventory quantity of each region is calculated, a replenishment instruction is generated, and an operation management report is generated based on the replenishment instruction, which specifically includes the following steps,

[0030] Based on the optimized spatio-temporal graph convolutional network model, the data of inventory, orders and environment are fused through the time window sliding alignment mechanism, and combined with the safety inventory prediction value, the dynamic programming algorithm is input to calculate the optimal inventory value under spatio-temporal constraints;

[0031] The replenishment path is generated by integrating the A algorithm and the ant colony algorithm, and the AGV is dynamically scheduled through the priority queue to execute tasks. At the same time, the replenishment time limit requirement and the goods category are injected into the instruction template to generate a replenishment instruction;

[0032] Based on the replenishment instruction, the historical replenishment records and inventory fluctuation curves are aggregated through the time series database, and the multi-dimensional analysis dashboard is automatically generated by using the Grafana tool to output the operation management report.

[0033] In a second aspect, the present invention provides an operation management system for a big data intelligent warehouse, including a collection module, a prediction module, an anomaly location module, a deviation calculation module, a model optimization module, and a report generation module;

[0034] The collection module is used to collect multi-source heterogeneous data, divide it into grid cells, and at the same time fuse time, space, and environmental features to generate a spatio-temporal data matrix;

[0035] The prediction module is used to input the spatio-temporal data matrix into a spatio-temporal graph convolutional network model, analyze the correlation between grid cells through multi-layer spatio-temporal convolution, and obtain the safety inventory prediction value for each region;

[0036] The anomaly location module is used to obtain the current inventory of each grid cell using an IoT sensor network, compare it with the safety inventory prediction value, and obtain anomaly candidate regions;

[0037] The deviation calculation module is used to construct a generative adversarial network model, analyze the distribution characteristics of anomaly candidate regions in the generative adversarial network model space, and calculate the deviation degree. When the deviation value exceeds the statistical threshold, it outputs the coordinates of the anomaly region and the deviation level;

[0038] The model optimization module is used to start different response processes according to the deviation level, and feedback the deviation data in the deviation level to the spatio-temporal graph convolutional network model to dynamically adjust the parameters and obtain an optimized spatio-temporal graph convolutional network model;

[0039] The report generation module is used to calculate the optimal inventory of each region based on the optimized spatio-temporal graph convolutional network model, generate a replenishment instruction, and generate an operation management report based on the replenishment instruction.

[0040] In a third aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and: when the computer program is executed by the processor, it implements any step of the operation management method for the big data intelligent warehouse as described in the first aspect of the present invention.

[0041] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and: when the computer program is executed by the processor, it implements any step of the operation management method for the big data intelligent warehouse as described in the first aspect of the present invention.

[0042] The beneficial effects of the present invention are as follows: By integrating multi-source heterogeneous data and using a spatio-temporal graph convolutional network model to analyze the correlation between these data, the present invention can more accurately obtain the safety inventory prediction value for each region. This not only improves the accuracy of inventory management decisions but also enables the warehouse to reduce the occurrence of overstocking or out-of-stock phenomena while meeting customer demands. By using a generative adversarial network model to deeply analyze the distribution characteristics of abnormal candidate regions and calculating the deviation degree to accurately locate the abnormal regions and their deviation levels, the false alarm rate is greatly reduced, the dependence on manual review is reduced, which not only saves human resources but also speeds up the problem response speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0044] Figure 1 It is a flowchart of the operation management method of the big data intelligent warehouse in Embodiment 1;

[0045] Figure 2 It is a schematic diagram of the operation management method of the big data intelligent warehouse in Embodiment 1;

[0046] Figure 3 It is a general flowchart of the operation management of the big data intelligent warehouse in Embodiment 1;

[0047] Figure 4 It is a flowchart of processing abnormal candidate regions in Embodiment 1. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the following will make a detailed description of the specific embodiments of the present invention with reference to the accompanying drawings of the specification.

[0049] Many specific details are set forth in the following description in order to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0050] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that can be included in at least one implementation manner of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that excludes other embodiments.

[0051] Example 1, referring to Figures 1 to 4 , which is the first embodiment of the present invention. This embodiment provides an operation management method for a big data intelligent warehouse, including the following steps:

[0052] S1. Collect multi-source heterogeneous data and divide it into grid cells, and at the same time fuse time, space, and environmental characteristics to generate a spatio-temporal data matrix.

[0053] Specifically, it includes the following steps.

[0054] In the process of realizing the transformation from collecting multi-source heterogeneous data to generating a spatio-temporal data matrix, first, a comprehensive data collection architecture is constructed. The core of this architecture lies in integrating a variety of sensors to ensure that key information such as geographical location data, environmental monitoring data, and historical inventory data can be obtained. For example, GPS devices are deployed to collect location information, environmental monitoring devices such as temperature and humidity sensors and air quality detectors are used to record environmental conditions in real time, and at the same time, the ERP is accessed to extract historical inventory records, providing raw materials for subsequent data processing and analysis.

[0055] After the multi-source heterogeneous data is collected, preprocessing is carried out, including steps such as data cleaning, format conversion, and time synchronization. Since multi-source heterogeneous data may have noise, missing values, or inconsistencies, appropriate data cleaning techniques must be adopted, such as using interpolation methods to fill in missing values and applying statistical methods to identify and correct outliers. At the same time, considering that data from different sources may have different formats, they need to be uniformly converted into a standard format suitable for further analysis. In addition, to ensure consistency in the time dimension, the timestamps of all multi-source heterogeneous data need to be calibrated. This step is particularly important for subsequent fusion of time and space characteristics.

[0056] After the preprocessing of multi-source heterogeneous data is completed, the next step is to divide the grid cells and generate a spatio-temporal data matrix on this basis. First, it is required to determine the size and distribution of the grid cells according to the actual layout. Each grid represents a specific area in the warehouse, and then the preprocessed geographical location, environmental monitoring, and historical inventory data are mapped to the corresponding grid cells. In this process, not only the physical space division needs to be considered, but also the time series needs to be combined, that is, the state changes of each grid cell in different time periods. To capture these dynamic changes more accurately, the concept of a time window can be introduced, and the state evolution of the grid cells can be observed by setting different time intervals.

[0057] Fuse grid cells with time, space, and environmental characteristics to generate a spatio-temporal data matrix. First, convert location information, environmental conditions, and historical inventory levels into numerical features. These numerical features not only reflect the current state of each grid cell but also contain their historical records of changes over time. For each grid cell, collect and record information such as its geographical location, environmental conditions, and inventory level at set time intervals, such as every hour or every day. Then, for each time point, construct a vector to represent the state of all grid cells at that time point. This vector contains numerical features such as the geographical location, environmental parameters, and inventory level corresponding to each grid cell. Subsequently, arrange the vectors in chronological order to form a multi-dimensional array, that is, the spatio-temporal data matrix, which can not only clearly show the state changes of each grid cell at different time periods but also capture the dynamic evolution process of the environmental conditions and inventory levels within the entire warehouse.

[0058] S2. Input the spatio-temporal data matrix into the spatio-temporal graph convolutional network model, and analyze the correlation between grid cells through multi-layer spatio-temporal convolution to obtain the safety inventory prediction value for each region.

[0059] Specifically, it includes the following steps.

[0060] The spatio-temporal graph convolutional network model consists of three layers in total. The first layer is the spatial correlation feature extraction layer, which captures the spatial correlation between adjacent grid cells through graph convolution operations. The second layer is the time series dynamic trend feature capture layer, which identifies potential trends and periodic patterns. The third layer is the attention mechanism layer, which dynamically adjusts the weights of the spatial correlation features and time series dynamic trend features. At the same time, to improve the performance of the spatio-temporal graph convolutional network model, a multi-scale time convolution strategy and residual connections are used to prevent the problem of gradient disappearance, and the attention mechanism is adopted during the training process to ensure that the spatio-temporal graph convolutional network model can flexibly adapt to various situations. Finally, the backpropagation algorithm combined with the Adam optimizer is used to minimize the loss function, and the parameters of the spatio-temporal graph convolutional network model are continuously adjusted to improve the prediction accuracy.

[0061] After the spatio-temporal graph convolutional network model is constructed and trained, in the first stage, the spatial correlation features are extracted. The spatio-temporal data matrix is input into the spatio-temporal graph convolutional network model, and the spatial correlation between adjacent grid cells is captured through spatio-temporal convolution operations. The key in this stage lies in how to effectively utilize the graph convolutional network to process the relationships between grid cells. In specific operations, each grid cell is regarded as a node in the graph, and their physical proximity relationships are represented as edges. These edges are usually defined by an adjacency matrix, which encodes the connection relationships between all grid cells. When performing graph convolution, the spatio-temporal graph convolutional network model not only considers the information of a single grid cell but also synthesizes the data of its neighbor nodes, thereby identifying the co-variation patterns within geographically adjacent regions. For example, if the storage volume within a certain grid cell suddenly increases, this may affect the demand patterns in its neighboring regions. In this way, the spatio-temporal graph convolutional network model can learn and predict the mutual influences between grid cells, laying a foundation for subsequent time series analysis;

[0062] In the second stage, the dynamic trend features in the time series are captured. In this process, temporal convolution is used to analyze the change trends of each grid cell over time. The one-dimensional convolution filter slides along the time axis to extract the dynamic trend features within different time periods from the spatio-temporal data matrix. This method allows the spatio-temporal graph convolutional network model to automatically learn which time intervals are most critical for predicting future inventory levels, and then identify potential trends and periodic patterns. For example, if the demand for certain inventory items surges, this periodic fluctuation can be effectively captured through temporal convolution. To improve the generalization ability and accuracy of the spatio-temporal graph convolutional network model, a multi-scale temporal convolution strategy can be adopted, that is, convolution kernels of different lengths are used simultaneously to capture short-term fluctuations and long-term trends. In addition, technical means such as residual connections can be combined to avoid the problem of gradient disappearance in the training of deep networks;

[0063] In the third stage, the attention mechanism is adopted to dynamically adjust the weights of the spatial correlation features and the time series dynamic trend features to obtain a comprehensive feature vector. The core idea of the attention mechanism is to let the spatio-temporal graph convolutional network model assign different weights to different features according to specific situations, so that the final output can better reflect the actual situation. For example, when obtaining the safety inventory prediction value, if the weather conditions are extremely bad during a certain period, then the importance of environmental monitoring data may increase significantly. On the contrary, when the demand is stable, historical inventory data may be more critical. By introducing the attention mechanism, the spatio-temporal graph convolutional network model can flexibly adapt to various situations and highlight the information that is most valuable for the current task. This step greatly improves the prediction accuracy;

[0064] The comprehensive feature vector is standardized or normalized, and the PCA algorithm is used to screen for feature compression. The comprehensively feature vector after feature compression is input into the fully connected layer. The role of the fully connected layer is to map the high-dimensional comprehensive feature vector to a low-dimensional space, thereby realizing the final safety inventory prediction. Each fully connected layer consists of multiple neurons, and each neuron is connected to all neurons in the previous layer. In this process, the spatio-temporal graph convolutional network model performs a linear transformation of weighted summation on the input comprehensive feature vector, and then through the ReLU activation function, the spatio-temporal graph convolutional network model can learn more complex patterns. To ensure the accuracy of the safety inventory prediction value, the parameters need to be optimized. Usually, the backpropagation algorithm is combined with the Adam optimizer to minimize the loss function. The loss function measures the difference between the safety inventory prediction value and the true inventory value. The optimization process aims to find a set of parameters to make this difference as small as possible to obtain the final safety inventory prediction value. Its mathematical formula is as follows:

[0065]

[0066] Among them, represents the safety inventory prediction value, V represents the weight matrix of the output layer, W represents the weight matrix of the fully connected layer, P represents the principal component analysis transformation matrix, k represents the number of principal components retained, P k represents the transformation matrix composed of the first k principal components, T represents the transpose operation, x represents the comprehensive feature vector, μ represents the mean of the comprehensive feature vector, σ represents the standard deviation of the comprehensive feature vector, b represents the bias term value of the fully connected layer, and c represents the bias term value of the output layer;

[0067] Through the above operations, the spatio-temporal graph convolutional network model can efficiently and accurately complete the task of safety inventory prediction, providing strong support for intelligent warehouse management.

[0068] S3. Use the IoT sensor network to obtain the current inventory of each grid cell, and compare it with the safety inventory prediction value to obtain abnormal candidate areas.

[0069] Specifically, it includes the following steps:

[0070] Use the IoT sensor network to collect warehouse data and integrate it to generate the original inventory log. Since there may be signal interference or noise in the actual environment, these original warehouse data need to be preprocessed to improve data quality. The specific operations include adopting the multi-read filtering technology, that is, reading the data at the same location multiple times and taking the average value to reduce random errors. At the same time, the signal compensation method is used to correct the signal attenuation problem caused by physical obstacles to ensure that the final obtained inventory data is as accurate as possible;

[0071] Match and analyze the preprocessed inventory data with the safety inventory prediction values obtained from the spatio-temporal graph convolutional network model. Adopt a real-time streaming window calculation method, that is, set a fixed time window, for example, every 10 minutes or every hour. Collect the inventory data of all grid cells within this time window and compare it with the safety inventory prediction values obtained from the spatio-temporal graph convolutional network model. To further improve the matching accuracy, apply the Dynamic Time Warping (DTW) algorithm to handle the non-linear deformation problem in time series data. The DTW algorithm can find the optimal alignment path between time series of different lengths, minimizing the distance between the series, so as to more accurately determine whether the current inventory level exceeds the safety inventory prediction value. Once the inventory quantity of a certain grid cell exceeds the preset safety inventory prediction value, the stream processing engine will mark this grid cell as a preliminary screening anomaly;

[0072] To reduce the false alarm rate, the preliminary screening anomalies need to be secondarily screened in combination with the rule engine. The rule engine evaluates the authenticity and severity of each preliminary screening anomaly through a context-aware rule chain. The context-aware rule chain considers various factors, such as historical inventory data, external environment changes, special situations, etc., to determine which anomalies need attention. In addition, the method of combining the density clustering algorithm with a dynamic neighborhood radius can be used to merge the preliminary screening anomalies that are geographically close and have similar anomaly characteristics into anomaly candidate regions. This method can not only effectively reduce the influence of isolated points but also better discover potential large-scale anomaly phenomena;

[0073] Perform multi-source sensor cross-verification on the identified anomaly candidate regions to ensure the accuracy of the detection results. This means that it is necessary to combine various sensor data, such as camera monitoring data, temperature and humidity sensor data, for comprehensive analysis. For example, if a certain region is marked as an anomaly candidate area, but the camera monitoring shows that there are no obvious abnormal logistics activities in this region, it may indicate that this anomaly is caused by data transmission errors or other non-critical factors rather than a real anomaly area. After cross-verification, output the exact coordinates of each anomaly candidate region and its corresponding anomaly type, such as overstock, out-of-stock, non-compliant environmental conditions, etc., to provide a basis for subsequent targeted measures;

[0074] Through the above operations, potential problems in warehouse operations can be effectively identified, improving the overall management level and efficiency. It not only improves the accuracy of anomaly detection but also reduces the possibility of false alarms, providing strong support for intelligent warehouses.

[0075] S4. Build a generative adversarial network model, analyze the distribution characteristics of the anomaly candidate regions in the generative adversarial network model space, and calculate the deviation degree. When the deviation value exceeds the statistical threshold, output the coordinates of the anomaly region and the deviation level.

[0076] Specifically, it includes the following steps:

[0077] The conditional generative adversarial network (CGAN) model consists of a generator and a discriminator. The task of the generator is to generate new samples similar to the real data distribution based on the input multi-source heterogeneous data, while the discriminator is responsible for distinguishing the generated new samples from the actual samples. To improve the accuracy and robustness of the conditional generative adversarial network model, contrastive learning supervision and dual-path adversarial training methods are used for training. In specific operations, contrastive learning supervision uses the similarity between positive and negative sample pairs for training to enhance the conditional generative adversarial network's understanding of normal patterns, which are defined based on statistical laws in historical inventory data. Dual-path adversarial training means considering the optimization of both the generator and the discriminator during the training process, enabling them to promote each other and jointly improve performance. For example, in contrastive learning supervision, the conditional generative adversarial network model will learn which feature combinations are closer to the normal pattern, thus helping the generator generate more realistic samples. In dual-path adversarial training, the loss functions of the generator and the discriminator will be optimized simultaneously to ensure that the samples generated by the generator are as difficult as possible to be distinguished by the discriminator, while the discriminator can more accurately distinguish actual samples from the generated new samples. In this way, the conditional generative adversarial network model can learn the data distribution characteristics in the normal pattern and provide a basis for subsequent anomaly detection;

[0078] Based on the trained conditional generative adversarial network model, the spatio-temporal graph attention mechanism is used to map and match the spatio-temporal features of the anomaly candidate regions with the data distribution characteristics in the normal pattern. The spatio-temporal graph attention mechanism is a technique specifically used to process datasets with complex spatio-temporal relationships, which not only considers the proximity relationship in the spatial dimension but also combines the dynamic changes in the temporal dimension. In specific operations, the spatio-temporal features of the anomaly candidate regions are extracted from the original multi-source heterogeneous data, which includes multi-source heterogeneous data such as location information, environmental parameters, and historical inventory records. Then, the spatio-temporal features are input into the spatio-temporal graph attention mechanism and compared with the data distribution characteristics in the normal pattern. The method of the triplet network is used here, which can calculate the spatial deviation degree between the anomaly candidate regions and the normal data distribution. The triplet network constructs three inputs (anchor, positive sample, and negative sample) to learn the relative distance between the actual samples and the generated new samples, thus better capturing the abnormal patterns. For example, if the inventory level of a certain grid cell is significantly higher or lower than the average level of its surrounding grid cells, then this cell may be regarded as an anomaly candidate region. This mechanism can effectively identify those regions that are significantly different from the normal pattern and lay a foundation for further analysis;

[0079] Based on the mapping results, calculate the spatial deviation degree of each abnormal candidate region. This deviation degree is a quantitative index that reflects the difference size between the abnormal region and the data distribution characteristics in the normal mode. For each abnormal candidate region, based on the difference between its spatio-temporal characteristics and the normal distribution, a specific deviation value is calculated. This value can be defined in various ways, such as Euclidean distance or cosine similarity, etc. For example, the average Euclidean distance between the abnormal candidate region and the actual sample set can be calculated as the deviation value. The larger the deviation value, the more obvious the difference between this region and the normal mode. To determine which regions should be regarded as real anomalies, a statistical threshold needs to be set. This threshold can be determined according to the normal fluctuation range in historical data, and usually statistical methods such as standard deviation or percentile are used to set it. For example, it can be set that if the deviation value exceeds twice the standard deviation of historical data, it is regarded as an anomaly. When the deviation value of a certain region exceeds this statistical threshold, it is considered that there is an anomaly in this region. This process not only depends on accurate mathematical calculations but also needs to comprehensively consider various variables in the actual situation to ensure the accuracy and reliability of the final result;

[0080] After determining the abnormal regions that exceed the statistical threshold, locate the specific grid coordinates of these regions through inverse mapping and output their deviation levels according to the over-limit ratio. Through inverse mapping technology, the calculated deviation value can be remapped back to the space of the original multi-source heterogeneous data, so as to accurately locate the specific grid coordinates of the abnormal region. Inverse mapping positioning depends on the spatio-temporal graph attention mechanism constructed before to ensure the accuracy and reliability of the mapping. For example, through the backpropagation algorithm, the deviation value can be converted back to the coordinate position in the space of the original multi-source heterogeneous data, so as to determine the specific position of the abnormal region;

[0081] Finally, evaluate the deviation level of each abnormal region. The deviation level is usually divided according to the size of the deviation value, such as mild, moderate, and severe, etc. In addition, other factors such as the influence range and duration can also be combined for comprehensive evaluation. For example, if the deviation value of a certain abnormal region is large and the duration is long, it may be classified as a severe anomaly. The final output not only includes the accurate coordinates of the abnormal region but also its corresponding deviation level, which is convenient for taking targeted measures subsequently. The whole process aims to provide detailed information support to ensure that potential problems in warehouse operations can be discovered and processed in time;

[0082] Through the above operations, the efficiency and safety of inventory management are ensured, and at the same time, the powerful application potential of modern algorithm analysis technology in intelligent warehouse management is demonstrated.

[0083] S5. Start different response processes according to the deviation level, and send the deviation data to the feedback spatio-temporal graph convolutional network model to dynamically adjust the parameters and obtain an optimized spatio-temporal graph convolutional network model.

[0084] Specifically, the following operations are included:

[0085] Based on the deviation value and its corresponding deviation level, initiate the corresponding response process. For mild anomalies, use monitoring devices to continuously track and record deviation data, and trigger the alarm mechanism by setting thresholds to ensure that any subtle changes can be recorded in a timely manner. For moderate anomalies, introduce data analysis tools such as Pandas in Python for in-depth analysis, identify potential patterns and causes, and use simulation techniques to evaluate the effects of different preventive measures, so as to select the optimal solution for implementation. For severe anomalies, immediately enable a real-time decision support mechanism, such as a fast response mechanism based on a rule engine, combine historical deviation data for rapid diagnosis, and automatically execute preset corrective measures, such as adjusting inventory allocation or optimizing the logistics path, to ensure that problems are resolved quickly and effectively, avoiding greater losses;

[0086] During the process of initiating the response process, input the deviation data into the Apache Flink stream processing framework, and use the time window function to align the deviation data. The specific operations include using Flink's sliding window function to group the deviation data at fixed time intervals. For example, a time window of one hour can be set to ensure that all deviation data is compared and analyzed within the same time period, effectively synchronizing data from sensors and eliminating errors caused by differences in collection times. Subsequently, input the deviation data aligned by the time window into the Prophet model for residual decomposition to extract periodic and trend features. Prophet is a tool specifically used for time series prediction, which can effectively capture seasonal and long-term trends in deviation data. Through residual decomposition, the periodic component and trend component can be separated from the original deviation data, which not only helps to identify potential abnormal patterns but also helps to understand the normal fluctuations in deviation data, thereby improving the accuracy of overall analysis;

[0087] To feedback the deviation data to the spatio-temporal graph convolutional network model, a gated attention mechanism is adopted. The gated attention mechanism allows the spatio-temporal graph convolutional network model to dynamically determine the importance of different features according to the current context, and through the gating mechanism, weights the periodic and trend features in the deviation data, highlighting the information that is most crucial for the current task. This not only improves the flexibility of the spatio-temporal graph convolutional network model but also enhances its ability to capture complex patterns. In specific operations, the gated attention mechanism weights the inventory features, anomaly detection features, and periodic and trend features at each time point, enabling the spatio-temporal graph convolutional network model to pay more attention to the regions with significant changes or anomalies while ignoring the relatively stable regions. Then, the features weighted by the gated attention mechanism are fed back to the spatio-temporal graph convolutional network model. These weighted features contain rich spatio-temporal information, which helps the spatio-temporal graph convolutional network model better understand and predict the inventory changes in the warehouse, effectively improving the adaptability of the spatio-temporal graph convolutional network model and enabling it to maintain a high prediction accuracy in a changing environment;

[0088] Based on the feedback deviation data, the parameters of the spatio-temporal graph convolutional network model are dynamically adjusted. First, the Elastic Weight Consolidation (EWC) technique is used to constrain the parameter update direction. EWC is a technique to prevent catastrophic forgetting. By constraining the parameter update direction, it protects the parameters that have been trained with historical multi-source heterogeneous data. EWC can help the spatio-temporal graph convolutional network model maintain its understanding of the existing parameters while adapting to new changes when inputting new multi-source heterogeneous data. Then, the mini-batch incremental training method is adopted to gradually and dynamically adjust the spatio-temporal attention weights, enabling it to more accurately capture the spatio-temporal correlation between grid cells. The mini-batch incremental training method can continuously improve the performance of the spatio-temporal graph convolutional network model without significantly increasing computational resources. Each training only uses a small part of the new multi-source heterogeneous data, allowing the spatio-temporal graph convolutional network model to quickly adapt to environmental changes. In addition, during the gradient clipping process, the EWC importance mask is separated from the momentum term to ensure that the two do not interfere with each other;

[0089] Finally, the AdamW optimizer is used to adjust the convolutional kernel parameters in the spatio-temporal graph convolutional network model. AdamW is a variant of the Adam optimizer. It effectively prevents overfitting problems through improvements in weight decay, generating an optimized spatio-temporal graph convolutional network model;

[0090] Through the above operations, the accuracy and efficiency of anomaly detection are improved, effectively reducing the operating costs and enhancing the overall business efficiency.

[0091] S6. Based on the optimized spatio-temporal graph convolutional network model, calculate the optimal inventory levels for each region, generate replenishment instructions, and generate an operation management report based on the replenishment instructions.

[0092] Specifically, it includes the following steps:

[0093] Based on the optimized spatio-temporal graph convolutional network model, use the time window sliding alignment mechanism to fuse inventory, order, and environmental data. In specific operations, data collected from different sources have different collection frequencies and formats. Through the time window sliding mechanism, it is ensured that multi-source heterogeneous data is processed and compared within the same time period. This process uses a stream processing framework such as Apache Flink to achieve data synchronization, ensuring the consistency and accuracy of multi-source heterogeneous data. At the same time, combine the safety inventory prediction value and input it into the dynamic programming algorithm to calculate the optimal inventory value under spatio-temporal constraints. The dynamic programming algorithm can effectively solve multi-stage decision-making problems and help determine the best inventory level for each grid cell within a specific time period, thus minimizing costs.

[0094] Next, generate a replenishment path and schedule the AGV to execute tasks. To find the optimal replenishment path, integrate the A algorithm and the ant colony algorithm. The A algorithm can quickly find the shortest path, while the ant colony algorithm finds the global optimal solution by simulating the foraging behavior of ants. The combination of the two can balance the requirements of local optimization and global optimization and generate an efficient replenishment path. The generated replenishment path dynamically schedules the AGV to execute tasks through a priority queue. The priority queue sorts tasks according to factors such as replenishment timeliness requirements and cargo categories to ensure that urgent tasks are processed first. For example, for items that need to be replenished immediately, the AGV can be quickly scheduled through a high-priority queue. In addition, inject the replenishment timeliness requirements and cargo categories into the instruction template to generate standardized replenishment instructions, clarifying the specific content, execution time, and priority of replenishment, facilitating the efficient execution of tasks by the AGV.

[0095] Finally, a detailed operation management report is generated based on the replenishment instructions. By aggregating historical replenishment records and inventory fluctuation curves in a time series database, a comprehensive understanding of past replenishment situations and their effects can be obtained. The time series database is specifically used to store and query data with timestamps and is very suitable for processing such time series data. Using the Grafana tool, information can be extracted from the aggregated historical replenishment data to generate a multi-dimensional analysis dashboard, which not only shows basic information such as inventory levels and replenishment frequencies but also enables in-depth analysis of the performance in different regions, time periods, or item categories to assist the decision-making process. For example, by viewing the inventory fluctuation curve of a certain region in the past month, potential inventory shortages or surpluses can be identified. Based on the above analysis results, a detailed operation management report is finally generated, covering aspects such as the inventory status, replenishment efficiency, and AGV operation status in each region, providing comprehensive operation insights. This report can not only serve as an important reference for internal management but also share information with other departments or partners to promote overall business collaboration.

[0096] This embodiment also provides an operation management system for a big data intelligent warehouse, including: a collection module, a prediction module, an anomaly location module, a deviation calculation module, a model optimization module, and a report generation module;

[0097] The collection module is used to collect multi-source heterogeneous data, divide it into grid cells, and at the same time fuse time, space, and environmental features to generate a spatio-temporal data matrix;

[0098] The prediction module is used to input the spatio-temporal data matrix into a spatio-temporal graph convolutional network model, analyze the correlation between grid cells through multi-layer spatio-temporal convolution, and obtain the safety inventory prediction value for each region;

[0099] The anomaly location module is used to obtain the current inventory quantity of each grid cell using the IoT sensor network and compare it with the safety inventory prediction value to obtain candidate anomaly regions;

[0100] The deviation calculation module is used to construct a generative adversarial network model, analyze the distribution characteristics of candidate anomaly regions in the generative adversarial network model space, and calculate the deviation degree. When the deviation value exceeds the statistical threshold, the coordinates of the anomaly region and the deviation level are output;

[0101] The model optimization module is used to start different response processes according to the deviation level and feedback the deviation data in the deviation level to the spatio-temporal graph convolutional network model to dynamically adjust the parameters and obtain an optimized spatio-temporal graph convolutional network model;

[0102] The report generation module is used to calculate the optimal inventory quantity for each region based on the optimized spatio-temporal graph convolutional network model, generate replenishment instructions, and generate an operation management report based on the replenishment instructions.

[0103] This embodiment also provides a computer device, which is applicable to the operation and management method of a big data intelligent warehouse, and includes: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the operation and management method of the big data intelligent warehouse proposed in the above embodiment.

[0104] The computer device can be a terminal. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner. The wireless manner can be implemented through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the housing of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0105] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the operation and management method of the big data intelligent warehouse proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (abbreviated as SRAM), electrically erasable programmable read-only memory (abbreviated as EEPROM), erasable programmable read-only memory (abbreviated as EPROM), programmable read-only memory (abbreviated as PROM), read-only memory (abbreviated as ROM), magnetic memory, flash memory, a magnetic disk, or an optical disc.

[0106] In summary, the present invention integrates multi-source heterogeneous data and uses a spatio-temporal graph convolutional network model to analyze the correlations between these data. As a result, the present invention can more accurately obtain the safety inventory prediction values for each region. This not only improves the accuracy of inventory management decisions but also enables the warehouse to reduce the occurrence of overstocking or out-of-stock situations while meeting customer demands. By using a generative adversarial network model to deeply analyze the distribution characteristics of abnormal candidate regions and calculating the deviation degree to precisely locate the abnormal regions and their deviation levels, the false alarm rate is significantly reduced, and the reliance on manual review is decreased, saving human resources and accelerating the problem response speed.

[0107] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. An operation management method for a big data intelligent warehouse, characterized in that: including, collecting multi-source heterogeneous data and dividing it into grid cells, while fusing time, space, and environmental features to generate a spatio-temporal data matrix; inputting the spatio-temporal data matrix into a spatio-temporal graph convolutional network model, analyzing the correlation between grid cells through multi-layer spatio-temporal convolution, and obtaining the safety stock prediction value for each region; using the IoT sensor network to obtain the current inventory of each grid cell, comparing it with the safety stock prediction value, and obtaining the abnormal candidate regions; constructing a generative adversarial network model, analyzing the distribution characteristics of the abnormal candidate regions in the generative adversarial network model space, and calculating the deviation degree. When the deviation value exceeds the statistical threshold, outputting the coordinates and deviation level of the abnormal region; initiating different response processes according to the deviation level, and feeding the deviation data in the deviation level back to the spatio-temporal graph convolutional network model to dynamically adjust the parameters and obtain an optimized spatio-temporal graph convolutional network model; based on the optimized spatio-temporal graph convolutional network model, calculating the optimal inventory for each region, generating replenishment instructions, and generating an operation management report based on the replenishment instructions.

2. The operation management method of the big data intelligent warehouse according to claim 1, characterized in that: The multi-source heterogeneous data includes geographical location data, environmental monitoring data, and historical inventory data.

3. The operation management method of the big data intelligent warehouse according to claim 1, characterized in that: The steps for obtaining the safety stock prediction value for each region are as follows: inputting the spatio-temporal data matrix into the spatio-temporal graph convolutional network model, extracting the spatial correlation features between adjacent grids through the first layer of spatio-temporal convolution, and analyzing the co-variation law of geographically adjacent regions; capturing the dynamic trend features of each grid cell in the time series through the second layer of time convolution, and identifying long-term and short-term fluctuation patterns; using the attention mechanism to dynamically adjust the weights of the spatial correlation features and dynamic trend features, and outputting the future safety stock prediction value for each grid cell.

4. The operation management method of the big data intelligent warehouse according to claim 1, characterized in that: The steps for obtaining the abnormal candidate regions are as follows: using the IoT sensor network to collect warehouse data in real time, generating the original inventory log, and removing interference through multi-read filtering and signal compensation methods to obtain the inventory data; dynamically matching the inventory data with the safety stock prediction value obtained from the spatio-temporal graph convolutional network model through real-time streaming window calculation and dynamic time warping algorithm, and using the stream processing engine to mark the grid cells exceeding the safety stock threshold as initially screened abnormalities; combining the initially screened abnormalities with the rule engine, excluding false alarms through the context-aware rule chain, and using the density clustering algorithm combined with the dynamic neighborhood radius to merge the initially screened abnormalities into abnormal candidate regions, and outputting the coordinates and abnormal types after cross-verification by multi-source sensors.

5. The operation management method of the big data intelligent warehouse according to claim 1, characterized in that: The output of the coordinates and deviation level of the abnormal region includes the following steps: constructing a conditional generative adversarial network model based on the generator and discriminator through contrastive learning supervision and dual-path adversarial training; based on the generative adversarial network model, mapping and matching the spatio-temporal features of the abnormal candidate regions with the features of the normal distribution through the spatio-temporal graph attention mechanism, and calculating the spatial deviation degree based on the triplet network; generating a deviation value based on the spatial deviation degree. When the deviation value exceeds the statistical threshold, locating the grid coordinates through inverse mapping, and outputting the coordinates and deviation level of the abnormal region according to the overrun ratio.

6. The operation management method of the big data intelligent warehouse according to claim 5, characterized in that: The feedback of the deviation data to the spatio-temporal graph convolutional network model to dynamically adjust the parameters and obtain an optimized spatio-temporal graph convolutional network model specifically includes the following steps: After aligning the deviation data based on Flink time windows, the periodic and trend features are extracted through Prophet residual decomposition, and the gated attention mechanism is used to feedback to the spatio-temporal graph convolutional network model; Based on the deviation data, the spatio-temporal graph convolutional network model updates the direction through EWC constraints, combines mini-batch incremental training to dynamically adjust the spatio-temporal attention weights, separates the EWC importance mask and the momentum term in gradient clipping, and uses the AdamW optimizer to update the convolutional kernel parameters to generate an optimized spatio-temporal graph convolutional network model.

7. The operation management method of the big data intelligent warehouse according to claim 6, characterized in that: Based on the optimized spatio-temporal graph convolutional network model, calculate the optimal inventory levels for each region, generate replenishment instructions, and generate an operation management report based on the replenishment instructions. The specific steps are as follows: Based on the optimized spatio-temporal graph convolutional network model, fuse the inventory, order, and environmental data through the time window sliding alignment mechanism, and combine the safety inventory prediction value to input into the dynamic programming algorithm to calculate the optimal inventory value under spatio-temporal constraints; Generate a replenishment path by integrating Algorithm A and the ant colony algorithm, and dynamically schedule the AGV to execute tasks through a priority queue. At the same time, inject the replenishment time limit requirement and the goods category into the instruction template to generate replenishment instructions; Based on the replenishment instructions, aggregate the historical replenishment records and inventory fluctuation curves through a time series database, and use the Grafana tool to automatically generate a multi-dimensional analysis dashboard to output an operation management report.

8. An operation management system for a big data intelligent warehouse, based on the operation management method of the big data intelligent warehouse according to any one of claims 1 to 7, characterized in that: Including a data collection module, a prediction module, an anomaly location module, a deviation calculation module, a model optimization module, and a report generation module; The data collection module is used to collect multi-source heterogeneous data and divide it into grid cells, and at the same time fuse time, space, and environmental features to generate a spatio-temporal data matrix; The prediction module is used to input the spatio-temporal data matrix into the spatio-temporal graph convolutional network model, analyze the correlation between grid cells through multi-layer spatio-temporal convolution, and obtain the safety inventory prediction value for each region; The anomaly location module is used to obtain the current inventory level of each grid cell using the IoT sensor network and compare it with the safety inventory prediction value to obtain anomaly candidate regions; The deviation calculation module is used to construct a generative adversarial network model, analyze the distribution characteristics of anomaly candidate regions in the generative adversarial network model space, and calculate the deviation degree. When the deviation value exceeds the statistical threshold, output the anomaly region coordinates and the deviation level; The model optimization module is used to start different response processes according to the deviation level, and feedback the deviation data in the deviation level to the spatio-temporal graph convolutional network model to dynamically adjust the parameters and obtain an optimized spatio-temporal graph convolutional network model; The report generation module is used to calculate the optimal inventory levels for each region based on the optimized spatio-temporal graph convolutional network model, generate replenishment instructions, and generate an operation management report based on the replenishment instructions.

9. A computer device, including a memory and a processor, the memory storing a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the operation management method of the big data intelligent warehouse 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 the processor, it implements the steps of the operation management method of the big data intelligent warehouse according to any one of claims 1 to 7.

Images