Unmanned warehouse ex-warehouse quantity prediction method and system and storage medium thereof

Abstract

The invention discloses an unmanned warehouse delivery quantity prediction method and system and a storage medium thereof. The method comprises the following steps: collecting multi-source data with aligned time sequences; forming a time sequence data set and a multi-dimensional feature data set based on the multi-source data; constructing a dynamic time sequence model, and inputting the time sequence data set into the dynamic time sequence model to output a reference predicted value; constructing a machine learning model, and inputting the multi-dimensional feature data set into the machine learning model to output a residual prediction value; and fusing the reference prediction value and the residual prediction value to obtain a warehouse-out quantity prediction value, and outputting the warehouse-out quantity prediction value to a warehouse downstream system. According to the method, the dynamic ARIMA model is taken as a reference time sequence predictor, the XGBoost residual error correction model is taken as an external influence catcher, a self-optimized hybrid prediction framework is realized through an online learning feedback loop, the prediction precision is high, the adaptability is strong, and the multi-source data of the unmanned warehouse can be fully utilized to perform warehouse-out quantity prediction.

Description

Technical Field

[0001] This invention belongs to the intersection of industrial Internet of Things and intelligent supply chain forecasting technology. Specifically, it is a method, system and storage medium for predicting outbound volume of unmanned warehouses. Background Technology

[0002] In the operation of modern unmanned warehouses, accurate forecasting of outbound volume (i.e., demand) is a core prerequisite for achieving inventory optimization, efficient scheduling of automated equipment, and rational planning of transportation resources. The accuracy of the forecast directly determines the warehouse's operating costs and customer satisfaction.

[0003] Currently, the main forecasting technologies in the industry have the following limitations: 1. The static nature and univariate limitations of classic time series models (such as ARIMA): While ARIMA models can effectively capture the inherent trends and patterns of time series, their model parameters remain fixed once determined, making them unable to adapt to rapid business growth or changes in the market environment. More importantly, as univariate models, they cannot utilize rich internal and external features (such as promotional activities, social media buzz, and warehouse equipment operating efficiency), resulting in extremely poor predictive ability for sudden, event-driven demand fluctuations.

[0004] 2. Pure machine learning models are insufficient in capturing time series dependencies: Although models such as XGBoost can handle multi-dimensional features, they are usually difficult to capture the long-term dependencies and periodic patterns of time series in a natural and deep way, like ARIMA. They require complex feature engineering to construct time series features such as lag terms.

[0005] 3. Coarseness of existing hybrid models: Some simple hybrid methods merely weight the prediction results of different models, failing to achieve deep complementarity between models. In particular, the prediction residuals of traditional ARIMA models are often discarded as noise, yet these residuals contain systematic information driven by external factors that the model failed to capture due to insufficient information.

[0006] Therefore, there is an urgent need in this field for an innovative prediction solution that can adaptively evolve and deeply integrate temporal patterns and multi-source information. Summary of the Invention

[0007] Therefore, in order to overcome the above shortcomings, this invention provides a method, system and storage medium for predicting outbound volume in unmanned warehouses. This invention uses a dynamic ARIMA model as the baseline time series predictor, an XGBoost residual correction model as the external influence capturer, and a self-optimizing hybrid prediction framework through an online learning feedback loop. It has high prediction accuracy, strong adaptability and can make full use of multi-source data from unmanned warehouses to predict outbound volume.

[0008] In a first aspect, the present invention provides a method for predicting the outbound volume of an unmanned warehouse, comprising: Collect time-series aligned multi-source data, including historical outbound data, warehouse operation data, business activity data, and external environment data; Time series datasets and multidimensional feature datasets are formed based on multi-source data; Construct a dynamic time series model, input the time series dataset into the dynamic time series model, and output the baseline prediction value; Build a machine learning model, input a multidimensional feature dataset into the machine learning model, and output residual prediction values; The baseline forecast and the residual forecast are combined to obtain the outbound volume forecast, which is then output to the downstream system of the warehouse.

[0009] Optionally, constructing the dynamic time series model includes: Using the ARIMA model as a dynamic time series model; An update strategy is adopted to enable the dynamic evolution of parameters in the dynamic time series model; The updated dynamic time series model is used to predict future target time periods.

[0010] Optionally, the update strategy is a sliding window update or / and a performance-triggered update.

[0011] Optionally, building a machine learning model includes: Using the XGBoost model as a machine learning model; Calculate the in-sample prediction residuals of the dynamic time series model on historical data; Construct a multidimensional feature dataset aligned with the time series of the residual sequence. The features include time series features, warehouse operation features, business activity features, and external environment features. Using historical residuals as the target variable and the corresponding multidimensional features as input, an XGBoost regression model G is trained. The objective function of the XGBoost model is the mean squared error loss function with L2 regularization, and the hyperparameters are optimized through time-series cross-validation.

[0012] Optionally, the unmanned warehouse outbound volume prediction method further includes online learning and feedback, which includes: Collect actual outbound volume data and calculate the final prediction error data based on the outbound volume forecast; Update the sliding window data based on the final prediction error data, and input it into the dynamic time series model for dynamic updates; Regularly add new data pairs to the training set of the machine learning model to perform incremental learning or retraining.

[0013] Secondly, the present invention provides an unmanned warehouse outbound volume prediction system, including... The data acquisition module is used to collect time-series aligned multi-source data, which includes historical outbound data, warehouse operation data, business activity data, and external environment data. The data processing module is used to generate time series datasets and multidimensional feature datasets based on multi-source data; The dynamic time series model module is used to build dynamic time series models. Inputting time series datasets into the dynamic time series model will output baseline predictions. The machine learning model module is used to build machine learning models. It takes a multidimensional feature dataset as input and outputs residual prediction values. The fusion module is used to fuse the baseline forecast value and the residual forecast value to obtain the outbound quantity forecast value, and output the outbound quantity forecast value to the downstream system of the warehouse.

[0014] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement the aforementioned method for predicting outbound volume from an unmanned warehouse.

[0015] The present invention has the following advantages: This invention discloses a method, system, and storage medium for predicting outbound volume in unmanned warehouses. It features high prediction accuracy, strong adaptability, and the ability to fully utilize multi-source data from unmanned warehouses for outbound volume prediction. The invention constructs a hybrid framework using a dynamic ARIMA model as the baseline predictor and an XGBoost model as the residual corrector. First, multi-source heterogeneous data is collected and preprocessed. Second, the ARIMA model parameters are dynamically updated using a sliding window or performance-triggered mechanism to output the baseline prediction. Then, the prediction residuals and multi-dimensional features of the ARIMA model are used as a training set to construct an XGBoost residual correction model to predict future residual values. Finally, the baseline prediction value and the residual prediction value are added to obtain the final prediction result, and the system achieves self-optimization through an online learning feedback loop. This invention achieves high-precision, adaptive prediction of outbound volume in unmanned warehouses, effectively improving warehouse operational efficiency and resource planning capabilities. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the outbound volume prediction method for an unmanned warehouse as described in this invention. Figure 2 This is a logical diagram of the unmanned warehouse outbound volume prediction method described in this invention; Figure 3 This is a modular block diagram of an unmanned warehouse outbound volume prediction system according to the present invention; In the diagram: 100, Data Acquisition Module; 200, Data Processing Module; 300, Dynamic Time Series Model Module; 400, Machine Learning Model Module; 500, Fusion Module. Detailed Implementation

[0017] The following will be combined with the appendix Figures 1-3 This invention will be described in detail, and the technical solutions in the embodiments of this invention will be clearly and completely described. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0018] As described in the background section, current major forecasting technologies in the industry have the following limitations: 1. The static nature and univariate limitations of classic time series models (such as ARIMA): While ARIMA models can effectively capture the inherent trends and patterns of time series, their model parameters remain fixed once determined, making them unable to adapt to rapid business growth or changes in the market environment. More importantly, as univariate models, they cannot utilize rich internal and external features (such as promotional activities, social media buzz, and warehouse equipment operating efficiency), resulting in extremely poor predictive ability for sudden, event-driven demand fluctuations.

[0019] 2. Pure machine learning models are insufficient in capturing time series dependencies: Although models such as XGBoost can handle multi-dimensional features, they are usually difficult to capture the long-term dependencies and periodic patterns of time series in a natural and deep way, like ARIMA. They require complex feature engineering to construct time series features such as lag terms.

[0020] 3. Coarseness of existing hybrid models: Some simple hybrid methods merely weight the prediction results of different models, failing to achieve deep complementarity between models. In particular, the prediction residuals of traditional ARIMA models are often discarded as noise, yet these residuals contain systematic information driven by external factors that the model failed to capture due to insufficient information.

[0021] For the reasons mentioned above, such as Figure 1 and Figure 2 As shown, this embodiment provides a method for predicting the outbound volume of an unmanned warehouse, including: Step S100: Collect time-aligned multi-source data.

[0022] For example, time-series aligned multi-source data is collected from various information systems and sensors in the unmanned warehouse. This multi-source data includes historical outbound data, warehouse operation data, business activity data, and external environment data. The above data is then cleaned, missing values ​​are filled in, and standardized to form a unified time-series dataset.

[0023] The historical outbound data refers to historical outbound volume information, accurate to the day. The warehouse operation data includes real-time inventory data, AGV task time, and picking station queue length from the warehouse management system and automated equipment.

[0024] The business activity data includes promotional plans and discount levels.

[0025] The external environment data includes social media indices, holiday markers, and weather data.

[0026] Step S200: Form a time series dataset and a multidimensional feature dataset based on multi-source data; Step S300: Build the component model and output the baseline prediction value and residual prediction value; Step S400: Combine the baseline forecast value and the residual forecast value to obtain the outbound quantity forecast value, and output the outbound quantity forecast value to the downstream system of the warehouse.

[0027] Step S300 includes: Construct a dynamic time series model, input the time series dataset into the dynamic time series model, and output the baseline prediction value; Build a machine learning model, input a multidimensional feature dataset into the machine learning model, and output residual prediction values; For example, the specific method for constructing the dynamic time series model is as follows: Using the ARIMA model as a dynamic time series model: The historical outflow sequence is differencing by order d to obtain a stationary sequence. The general form of the ARIMA(p,d,q) model is: ; in: p It is the order of the autoregressive term; d It is the difference order; q It is the order of the moving average term; ∇ d yes d Order difference operator; B It is a shift operator ( B yt = y t -1); ϕ ( B )=1 - ϕ 1 B - ... - ϕ p B p It is an autoregressive polynomial; θ ( B )=1 + θ 1 B + ... + θ q B q It is a moving average polynomial; ϵ t It is a white noise sequence.

[0028] t represents what time point.

[0029] The optimal (p,d,q) parameters are determined using the grid search method based on the AIC criterion, and an initial ARIMA model is established: AIC=2 k -2ln( L ); in k It is the number of model parameters. L This is the model likelihood function value. The model with the smallest AIC is optimal.

[0030] Use at least one of the following strategies to dynamically evolve the parameters of the ARIMA model: a) Sliding window update: Before each prediction, the ARIMA model parameters are automatically re-identified and fitted using data from the most recent N days (e.g., 90 days).

[0031] b) Performance-triggered updates: Continuously monitor model prediction error. When the error continuously exceeds a preset threshold, automatically trigger model retraining. Define the mean absolute percentage error (MAPE) as the monitoring metric. ; n For the specified number of days, t Current time, t-1 yes t The day before; i Indicates the number of days; y t-i It is the actual value at a specific point in the past. It is a predicted value at a specific point in the past.

[0032] when M t continuous Wsky( W It is the defined consecutive number of days, such as W =3) Exceeds the threshold T (like T If the percentage is 10%, then retraining of the ARIMA model will be triggered immediately.

[0033] The updated dynamic ARIMA model is used to predict the future target period, resulting in a baseline forecast value. F arima .

[0034] For example, a specific method for building a machine learning model is: Residual sequence calculation: Calculate the in-sample prediction residuals of the dynamic ARIMA model on historical data. , in y t This is the actual value. These are ARIMA predicted values.

[0035] Feature engineering: Construct a multidimensional feature dataset that is time-aligned with the residual sequence, including time-series features, warehouse operation features, business activity features, and external environment features.

[0036] Model training: using historical residuals et Using the target variable and its corresponding multidimensional features as input, train an XGBoost regression model. G The objective function of this model is the mean squared error loss function with L2 regularization, and hyperparameter optimization is performed through temporal cross-validation. The objective function Obj of the XGBoost model is determined by the loss function. L Composed of the regularization term Ω: ; in: R It sets the maximum value of the regression number. r It is the regression number.

[0037] , is the mean square error; e i The ARIMA model is in the 19th century. i The actual predicted residual for the day; These are the residual values ​​predicted by the XGBoost model; (L2 regularization controls model complexity); T It is the total number of leaf nodes in the decision tree; γ It is the L2 regularization weight, which is an adjustable penalty coefficient; w It is the predicted score for each leaf node; λ It is the minimum gain of splitting, and it is another penalty coefficient.

[0038] Residual prediction: For a future target time period, known or predicted feature data are input into a trained XGBoost model to obtain residual prediction values. Fres = G ( Xfuture ).

[0039] For example, in step S400, the baseline forecast value and the residual forecast value are fused to obtain the outbound quantity forecast value, i.e. Ffinal = Farima + Fres .

[0040] In one embodiment, the unmanned warehouse outbound volume prediction method further includes: S500, online learning and feedback; The specific methods for online learning and feedback are as follows: (1) Collect actual outbound data and calculate the final prediction error.

[0041] (2) Add new data points to the sliding window to provide new data for dynamic updates.

[0042] (3) Periodically add new (features, actual residuals) data pairs to the training set of the XGBoost model, and perform incremental learning or retraining of the XGBoost model to achieve closed-loop optimization and continuous improvement of the entire system.

[0043] The technical problem addressed by the aforementioned unmanned warehouse outbound volume prediction method is that existing prediction methods suffer from static and fixed models, inability to utilize multi-source data, and coarse hybrid models. This method deeply couples a time series model (ARIMA) with a machine learning model (XGBoost), rather than simply using a weighted average. ARIMA is responsible for capturing the inherent patterns (trends, seasonality) of the time series, while XGBoost specifically captures systematic biases caused by external factors that ARIMA cannot explain, achieving complementary advantages between the models.

[0044] To address the static and fixed nature of traditional time series models, two adaptive mechanisms, sliding window updates and performance-triggered updates, were introduced into the ARIMA model. This allows the model to automatically adjust its parameters in response to the evolution of business data and the market environment, overcoming the "concept drift" problem and ensuring the continued effectiveness of its predictive capabilities.

[0045] This approach creatively treats the prediction residuals of the ARIMA model as valuable signals rather than useless noise. By leveraging the powerful feature learning capabilities of the XGBoost model, it uncovers the correlation between residuals and multi-dimensional external features (such as promotions, device efficiency, and social media buzz), thereby enabling accurate prediction and correction of systematic biases.

[0046] A complete self-optimizing intelligent system was constructed, enabling closed-loop online learning and feedback optimization. By continuously collecting actual outbound data, a feedback loop is formed, automatically providing new training data for the dynamic ARIMA model and new learning samples for the XGBoost residual model. This allows the entire prediction system to continuously iterate and evolve, reducing long-term maintenance costs.

[0047] Compared to existing technologies, this invention significantly improves prediction accuracy. On the test set, the mean absolute percentage error (MAPE) of the proposed method is significantly lower than that of a single ARIMA model or a single XGBoost model. Furthermore, the system exhibits strong adaptability; its dynamic update mechanism ensures that the model maintains more stable prediction errors even when facing annual increases in business volume, eliminating the need for manual parameter tuning. Simultaneously, operating costs are effectively reduced. Based on predictions, inventory levels can be optimized, and the efficiency of automated equipment scheduling can be improved, thereby reducing the overall operating costs of unmanned warehouses.

[0048] Secondly, such as Figure 3 As shown, the present invention provides an unmanned warehouse outbound volume prediction system, including... The data acquisition module 100 is used to collect time-series aligned multi-source data, which includes historical outbound data, warehouse operation data, business activity data, and external environment data. Data processing module 200 is used to generate time series datasets and multidimensional feature datasets based on multi-source data; The dynamic time series model module 300 is used to construct a dynamic time series model. The time series dataset is input into the dynamic time series model and the output is a baseline prediction value. Machine learning model module 400 is used to build machine learning models. The multidimensional feature dataset is input into the machine learning model and the output is a residual prediction value. The fusion module 500 is used to fuse the baseline forecast value and the residual forecast value to obtain the outbound quantity forecast value, and output the outbound quantity forecast value to the downstream system of the warehouse.

[0049] This application also provides a computer-readable storage medium storing a computer program thereon, characterized in that the program is executed by a processor to implement the above-described method for predicting the outbound volume of an unmanned warehouse.

[0050] If the modules / units integrated in the unmanned warehouse outbound volume prediction system / terminal device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above.

[0051] The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0052] In the several embodiments provided in this application, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection of devices or units may be electrical, mechanical, or other forms.

[0053] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0054] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0055] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The above embodiments are only used to illustrate the technical solution of this application and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of this application.

[0056] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting outbound volume in an unmanned warehouse, characterized in that, include Collect time-series aligned multi-source data, including historical outbound data, warehouse operation data, business activity data, and external environment data; Time series datasets and multidimensional feature datasets are formed based on multi-source data; Construct a dynamic time series model, input the time series dataset into the dynamic time series model, and output the baseline prediction value; Build a machine learning model, input a multidimensional feature dataset into the machine learning model, and output residual prediction values; The baseline forecast and the residual forecast are combined to obtain the outbound volume forecast, which is then output to the downstream system of the warehouse.

2. The method for predicting outbound volume of an unmanned warehouse according to claim 1, characterized in that, The construction of the dynamic time series model includes: Using the ARIMA model as a dynamic time series model; An update strategy is adopted to enable the dynamic evolution of parameters in the dynamic time series model; The updated dynamic time series model is used to predict future target time periods.

3. The method for predicting outbound volume of an unmanned warehouse according to claim 2, characterized in that, The update strategy is a sliding window update or / and a performance-triggered update.

4. The method for predicting outbound volume of an unmanned warehouse according to claim 1, characterized in that, Building machine learning models includes: Using the XGBoost model as a machine learning model; Calculate the in-sample prediction residuals of the dynamic time series model on historical data; Construct a multidimensional feature dataset aligned with the time series of the residual sequence. The features include time series features, warehouse operation features, business activity features, and external environment features. Using historical residuals as the target variable and the corresponding multidimensional features as input, an XGBoost regression model G is trained. The objective function of the XGBoost model is the mean squared error loss function with L2 regularization, and the hyperparameters are optimized through time-series cross-validation.

5. The method for predicting outbound volume of an unmanned warehouse according to claim 1, characterized in that, It also includes online learning and feedback, which includes: Collect actual outbound volume data and calculate the final prediction error data based on the outbound volume forecast; Update the sliding window data based on the final prediction error data, and input it into the dynamic time series model for dynamic updates; Regularly add new data pairs to the training set of the machine learning model to perform incremental learning or retraining.

6. A system for predicting outbound volume in an unmanned warehouse, characterized in that: include The data acquisition module is used to collect time-series aligned multi-source data, which includes historical outbound data, warehouse operation data, business activity data, and external environment data. The data processing module is used to generate time series datasets and multidimensional feature datasets based on multi-source data; The dynamic time series model module is used to build dynamic time series models. Inputting time series datasets into the dynamic time series model will output baseline predictions. The machine learning model module is used to build machine learning models. It takes a multidimensional feature dataset as input and outputs residual prediction values. The fusion module is used to fuse the baseline forecast value and the residual forecast value to obtain the outbound quantity forecast value, and output the outbound quantity forecast value to the downstream system of the warehouse.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement a method for predicting outbound volume from an unmanned warehouse as described in any one of claims 1-6.

Images