Method for optimizing storage and taking efficiency of stacker based on time sequence and related assembly

Abstract

The invention discloses a method for optimizing the access efficiency of a stacker based on a time sequence and related components. The method comprises the steps that according to the position of a warehouse-in and warehouse-out platform connected with a stacking machine in a goods allocation roadway, a warehouse-in and warehouse-out buffer area is arranged in the goods allocation roadway; acquiring historical warehouse-in and warehouse-out data corresponding to each product specification in a preset time period, inputting the historical warehouse-in and warehouse-out data into the warehouse-in and warehouse-out prediction model corresponding to the preset time period for prediction, and outputting future warehouse-in and warehouse-out data corresponding to each product specification in the preset time period; and according to the future warehouse-in and warehouse-out data corresponding to each product specification in the preset time period, generating a warehouse-moving task, and controlling the stacker to perform warehouse-in and warehouse-out operation among the corresponding goods allocation roadway, the warehouse-in buffer area and the warehouse-out buffer area according to the warehouse-moving task. According to the method, the warehouse-in and warehouse-out task in the next time period is predicted, the warehouse-moving task is generated, the warehouse-in and warehouse-out buffer area is used for carrying out the warehouse-moving operation in advance so as to reduce the operation distance of the stacking machine when the stacking machine executes the warehouse-in and warehouse-out task in the next time period, and the warehouse-in and warehouse-out efficiency is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of warehouse management, and in particular to a method for optimizing the storage and retrieval efficiency of a stacker based on time series and related components. Background Art

[0002] In current warehouse management, the efficiency of warehouse entry and exit is often considered. From the perspective of logistics equipment, the time required for entry and exit is the time from when the order is issued to when the goods arrive at the designated location. Most of this time is determined by the actual operation time of the equipment. Therefore, optimizing the route and shortening the actual transportation distance of the goods can significantly improve the efficiency of entry and exit.

[0003] In the warehousing and retrieval work of the high-bay warehouse, the access time of each cargo location in the high-bay warehouse varies due to the distance between it and the connected warehousing and retrieval platform. Giving priority to storing the goods to be warehousing and retrieval in cargo locations with shorter access times can significantly improve work efficiency. However, except for a few specific usage scenarios, in most cases it is impossible to predict the specifications and quantity of goods that need to be shipped out or in the warehouse in the next time period (such as the next day). And most of the current automated warehouses usually do not deploy such optimization solutions due to the lack of prediction capabilities for the tasks in the next time period. Summary of the invention

[0004] The purpose of the present invention is to provide a method and related components for optimizing the storage and retrieval efficiency of a stacker based on time series, aiming to solve the problem of lack of forward-looking prediction and effective deployment capabilities in the storage and retrieval operations of existing warehouses, which leads to low storage and retrieval efficiency.

[0005] In a first aspect, an embodiment of the present invention provides a method for optimizing the storage and access efficiency of a stacker based on a time series, comprising: According to the position of the inbound and outbound platform connected to the stacker in each cargo lane, a corresponding inbound and outbound buffer zone is set in each cargo lane; Obtain the historical in-and-out data of each product specification corresponding to the preset time period, and input it into the in-and-out prediction model corresponding to the preset time period for prediction, and output the future in-and-out data of each product specification corresponding to the preset time period; Generate a warehouse transfer task based on the future in-and-out warehouse data corresponding to each product specification in a preset time period, and control the stacker to perform in-and-out warehouse transfer operations between the corresponding cargo lanes, in-warehouse buffers and out-warehouse buffers according to the warehouse transfer task.

[0006] In a second aspect, an embodiment of the present invention provides a device for optimizing the storage and access efficiency of a stacker based on a time series, comprising: An area setting unit is used to set a corresponding inbound and outbound buffer zone in each cargo lane according to the position of the inbound and outbound platform connected to the stacker in each cargo lane; The in-and-out inventory prediction unit is used to obtain the historical in-and-out inventory data corresponding to the preset time period of each product specification, and input it into the in-and-out inventory prediction model corresponding to the preset time period for prediction, and output the future in-and-out inventory data corresponding to the preset time period of each product specification; The in-and-out transfer unit is used to generate transfer tasks according to the future in-and-out data corresponding to each product specification in a preset time period, and control the stacker to perform in-and-out transfer operations between the corresponding storage location aisles, inbound buffers and outbound buffers according to the transfer tasks.

[0007] In a third aspect, an embodiment of the present invention provides a computer device, which includes an access device, a processor, and a computer program stored on the access device and run on the processor, and when the processor executes the computer program, it implements the method for optimizing the stacker access efficiency based on time series as described in the first aspect above.

[0008] In a fourth aspect, an embodiment of the present invention provides a computer-readable access medium, wherein the computer-readable access medium stores a computer program, and when the computer program is executed by a processor, the processor executes the method for optimizing the stacker access efficiency based on time series as described in the first aspect above.

[0009] The beneficial effects of the embodiments of the present invention are as follows: by predicting the inbound and outbound warehousing tasks in the next time period, generating corresponding warehouse transfer tasks based on the prediction results, and then using the inbound and outbound warehousing buffer to transfer and sort the goods to be inbound and outbound in advance, the running distance of the stacker when performing the inbound and outbound warehousing tasks in the next time period can be effectively reduced, thereby significantly improving the inbound and outbound operation efficiency of the stacker. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0011] Figure 1 A schematic flow chart of a method for optimizing stacker access efficiency based on time series provided in an embodiment of the present invention; Figure 2 A schematic diagram of a sub-process method of step S102 provided in an embodiment of the present invention; Figure 3 A schematic diagram of another sub-process method of step S102 provided in an embodiment of the present invention; Figure 4 A schematic diagram of another sub-process method of step S102 provided in an embodiment of the present invention; Figure 5A schematic diagram of a sub-process method of step S103 provided in an embodiment of the present invention; Figure 6 A further flow chart of a method for optimizing stacker access efficiency based on time series provided in an embodiment of the present invention; Figure 7 A further flow chart of a method for optimizing stacker access efficiency based on time series provided in an embodiment of the present invention; Figure 8 A further flow chart of a method for optimizing stacker access efficiency based on time series provided in an embodiment of the present invention; Fig. 9 A further flow chart of a method for optimizing stacker access efficiency based on time series provided in an embodiment of the present invention; Fig.10 A schematic block diagram of a device for optimizing stacker access efficiency based on time series provided by an embodiment of the present invention; Fig.11 A schematic block diagram of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0012] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0013] It should be understood that when used in this specification and the appended claims, the terms "include" and "comprises" indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0014] It should also be understood that the terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, the singular forms "a", "an" and "the" are intended to include plural forms unless the context clearly indicates otherwise.

[0015] It should be further understood that the term "and / or" used in the present description and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0016] To facilitate the understanding of the present application, the working links used in the present application are first introduced. Specifically, in the warehousing and out-of-warehouse work, when the stacker visits the target cargo position of the elevated warehouse (or the warehousing and out-of-warehouse platform), it is generally divided into the following steps and executed separately: Step a, move to the vicinity of the designated position: the time required for this process is related to the distance between the target cargo position and the current position. Since the main motor of the stacker needs to work for a long time, the energy consumption is extremely high. Step b, precise positioning: after arriving near the designated position, it is necessary to fine-tune the position of the stacker to align (it can be understood by analogy to the process of aligning the subway station with the platform door) the target cargo position. The time required for this process is normally distributed, generally takes no more than 5 seconds, and the energy consumption is low. Step C, access the target cargo position through the stacker's fork: this process has a fixed time; the fork movement will consume a certain amount of electrical energy. The optimization involved in the present invention is only performed on step a.

[0017] To this end, the present application aims to propose an optimization scheme, which predicts the warehousing and out-of-warehouse tasks in the next time period and uses the warehousing and out-of-warehouse buffer to perform warehouse transfer operations in advance, so as to reduce the running distance of the stacker when performing the warehousing and out-of-warehouse tasks in the next time period, thereby improving the warehousing and out-of-warehouse operation efficiency of the stacker. The specific optimization process is detailed in the contents of the following embodiments.

[0018] See also Figure 1 , Figure 1 A schematic flow chart of a method for optimizing stacker access efficiency based on time series provided in an embodiment of the present invention; like Figure 1 As shown, the method includes steps S101-S103.

[0019] S101, according to the position of the in-and-out storage platform connected to the stacker in each storage lane, set a corresponding in-and-out storage buffer in each storage lane; In this step, the cargo lane is the passage for the stacker to transport and store goods in the warehouse. In the cargo lane, there are high-bay warehouses for storing and retrieving goods arranged along the lane. The high-bay warehouse has multiple cargo spaces in the height direction. The inbound platform and outbound platform are connected to the corresponding cargo lane. The stacker performs the storage and retrieval of goods between the corresponding cargo lane, inbound platform and outbound platform to complete the inbound and outbound tasks.

[0020] In an embodiment of the division of an inbound and outbound buffer zone, the inbound buffer zone can be set close to the inbound platform, and the outbound buffer zone can be set close to the outbound platform. According to the predicted inbound and outbound tasks in the next time period, the corresponding goods can be moved to the designated inbound and outbound buffer zones in advance according to the inbound and outbound tasks. In this way, when the inbound and outbound tasks are executed in the next time period, the moving distance of the stacker can be effectively shortened, thereby significantly improving the operating efficiency of the stacker.

[0021] In another embodiment of the division of the inbound and outbound buffer zones, the areas of the inbound buffer zone and the outbound buffer zone can also be set to overlap. In this case, the number of cargo spaces dedicated to inbound and outbound storage needs to be specified. The remaining number of cargo spaces will be filled as the outbound storage buffer zone when the cargo spaces are arranged, and will be converted to the inbound storage buffer zone after outbound storage until the next cargo space arrangement.

[0022] S102, obtaining the historical in-and-out data of each product specification corresponding to a preset time period, and inputting it into the in-and-out prediction model corresponding to the preset time period for prediction, and outputting the future in-and-out data of each product specification corresponding to the preset time period; In this step, each product specification, that is, the category and specification of each commodity, can have a preset time period that can be long-term, medium-term or short-term, and can be set according to actual needs. Each preset time period can correspond to an inbound and outbound inventory prediction model. These inbound and outbound inventory prediction models can predict the future inbound and outbound inventory data of the same preset time period in the next time period based on the historical inbound and outbound inventory data corresponding to each product specification in the preset time period, thereby providing a reliable basis for carrying out warehouse transfer tasks in advance.

[0023] S103, generating a warehouse transfer task according to the future in-and-out warehouse data corresponding to each product specification in a preset time period, and controlling the stacker to perform in-and-out warehouse transfer operations between the corresponding cargo location lanes, in-warehouse buffers and out-warehouse buffers according to the warehouse transfer task.

[0024] In this embodiment, based on the warehouse transfer task, the corresponding warehouse transfer operation is performed in advance. For example, the goods that need to be shipped out in the next time period are moved to the outbound buffer in advance, and at the same time, the cargo space in the inbound buffer close to the inbound platform is freed up, so that the goods to be shipped in can be moved from the inbound platform into the warehouse more efficiently in the next time period.

[0025] Based on the process of steps S101-S103, the present application aims to propose an optimization scheme, which predicts the inbound and outbound tasks of the next time period, generates corresponding warehouse transfer tasks according to the prediction results, and then uses the inbound and outbound buffer to transfer and sort the goods to be inbound and outbound in advance. This can effectively reduce the running distance of the stacker when performing inbound and outbound tasks in the next time period, thereby reducing the time required for operation and significantly reducing the energy consumption of the stacker, thereby significantly improving the efficiency of the stacker's inbound and outbound access operations.

[0026] Furthermore, although the warehouse transfer task will consume energy, it can be performed during the off-peak period of the radio wave. The off-peak period of the radio wave is an area where electricity consumption is encouraged, and the total energy consumption cost will still be significantly reduced.

[0027] In one embodiment, if Figure 2 As shown, step S102 includes: S201, obtain the historical in-and-out data corresponding to each product specification in the long-term cycle; S202. Confirm the business characteristics associated with each product specification; S203. According to the business characteristics associated with each product specification, the corresponding historical inbound and outbound data are input into the Prophet model or the SARIMA model for prediction, and the future inbound and outbound data corresponding to each product specification in the long-term period are output.

[0028] In this embodiment, the long-term cycle can be a monthly cycle (generally 30 days, or the number of days in a natural month), and the historical inventory data of each product specification in the monthly cycle is obtained, that is, the historical inventory data of one month as a cycle. The number of cycles is not limited, for example, 8 months, 12 months or 16 months can be set according to actual needs.

[0029] In this embodiment, in the inventory inflow and outflow forecast for the next time period with a monthly cycle, two inventory inflow and outflow forecast models with a monthly cycle are provided, and the corresponding inventory inflow and outflow forecast model needs to be selected according to the business characteristics associated with the product specifications.

[0030] The following are the specific selection criteria for two monthly inventory forecasting models: Model 1: Prophet model, applicable to business features including: Feature 1: The warehouse has multiple obvious periodic patterns (such as annual, quarterly, monthly, and weekly); Feature 2: There are many goods affected by fixed or floating holidays; Feature 3: There are sudden changes in business trends (such as new product lines and customer changes); Feature 4: There are missing historical data (such as equipment maintenance period) or abnormal values ​​(such as sudden large orders).

[0031] Specifically, the basic formula of the Prophet model is: ; Among them, g(t) represents the trend function, which captures the business growth or decline trend; s(t) represents the seasonal function, which describes the periodic changes; h(t) represents the holiday effect, which describes the impact of special events; represents the error term.

[0032] Specifically, the input data of the Prophet model corresponds to the model parameters as follows: 1) Basic time series data (in and out history table); 1.1) Date-quantity pair: used to fit trend function; 1.2) Daily aggregated data: used to identify seasonal patterns; 1.3) Minimum data volume 12 months: ensure capture of the full annual cycle; 1.4) Data frequency: daily, used to determine the basic time unit; 2) Holiday data (special event table); 2.1) Event date: defined impact period; 2.2) Event type: distinguish different holiday effects; 2.3) v impact coefficient: quantify the intensity of holiday effect; The important parameter settings can be as follows (the following parameter values ​​are for reference only and should be adjusted according to the actual business scenario. This implementation does not make specific restrictions): The forecast period can be set to 30 days, corresponding to the monthly business planning period; The seasonal cycle can be set based on the business cycle, such as one year (365 days), one month (30 days), or one week (7 days). Change point detection can use the significant change time points in the special event table; The holiday window can be ±3 days, determined based on the impact of historical holidays.

[0033] Model 2: SARIMA model, applicable to business characteristics including: Feature 1: The warehouse in and out pattern is relatively stable; Feature 2: Mainly manifested as a single periodicity (such as only a weekday period); Feature 3: The types of goods are fixed and the quantity fluctuates regularly; Feature 4: Short-term data has good continuity and obvious fluctuation patterns.

[0034] Specifically, the basic formula of the SARIMA model is: ; Where: B represents the lag operator; d represents the ordinary difference order; D represents the seasonal difference order; s represents the seasonal cycle; represents the general autoregressive polynomial; represents the seasonal autoregressive polynomial; represents the ordinary moving average polynomial; represents the seasonal moving average polynomial.

[0035] Specifically, the input data of the SARIMA model corresponds to the model parameters as follows: 1) Basic time series data (in and out history table); 1.1) Continuous daily data: used for differential calculation and model fitting; 1.2) Historical fluctuation pattern: indicates the continuous change characteristics of the inventory quantity; 1.21) Data format: sequence of differences in the number of inbound and outbound quantities at adjacent time points; 1.22) Calculation method: Δy t = y t -y t -1; 1.23) Purpose: Helps determine the autoregressive order p (the current value is affected by the previous p time points) and the moving average order q (the current error is affected by the previous q errors); 1.3) Periodic pattern: refers to the regular repetitive change characteristics; 1.31) Data form: the number of inbound and outbound items counted at different time scales (day, week, month); 1.32) Example: Large inbound volume every Monday, large outbound volume at the end of each month, etc.; 1.33) Purpose: Determine seasonal parameters P, D, Q (similar to p, d, q but in periodic units); Where d represents the number of differences required to eliminate the overall trend; Where D represents the number of differences required to eliminate the periodic trend; 1.4) Data stationarity: refers to whether the statistical characteristics of the time series change over time; 1.41) Data format: original in-and-out quantity sequence; 1.42) Judgment method: observe whether the mean and variance change significantly over time; 1.43) Purpose: Determine whether difference is needed and the difference order; 2) Seasonal reference data (special event table) 2.1) Weekday marking: helps identify basic cycles; 2.2) Seasonal marking: assists in determining seasonal parameters; The important parameter settings can be as follows (the following parameter values ​​are for reference only and should be adjusted according to the actual business scenario. This implementation does not make specific restrictions): 2.3) Difference order (d, D): determined by data stationarity analysis; 2.4) Seasonal cycle: usually 7 (working cycle), which can be adjusted according to the actual business cycle; 2.5) Forecast step: 30 days, synchronized with business planning.

[0036] Specifically, the following are the prediction results output of the Prophet model and the SARIMA model: After selecting the model, the daily inbound and outbound quantities for the next 30 days are predicted, and the output results are: ;

[0037] Where: t represents the forecast date, from 1 to 30, representing the next 30 days; T tIndicates the trend value on the tth day, reflecting the long-term trend; S t represents the seasonal value on the tth day, reflecting the periodic changes; R t Represents the random item on day t, with a probability distribution described by the predicted mean and standard deviation, reflecting short-term fluctuations. For each product ID, a set of such prediction sequences will be generated.

[0038] Specifically, get the prediction results from the Prophet model: T t Indicates directly obtaining the trend function output of the model; S t Indicates directly obtaining the seasonal function output of the model; R t Represents the mean and variance calculation based on the prediction results; μ t represents the point estimate of the model prediction; σ t Represents the residual error based on the model The calculated standard deviation.

[0039] Specifically, obtain the prediction results from the SARIMA model: T t Indicates that the trend decomposition of the forecast sequence is obtained using (1-B) d The reverse calculation result after eliminating the trend; S t Indicates the use of seasonal difference results, based on (1-B S ) D The seasonal component of the term; R t It means that it is calculated based on the model residual. μ t Represents the direct prediction value of the model; σ t Represents the standard deviation calculated based on the residual term.

[0040] Specifically, the final predicted daily inventory quantity is: ; Where R t Used to express the uncertainty range of a prediction.

[0041] In one embodiment, if Figure 3 As shown, step S102 includes: S301, obtaining the historical in-and-out data corresponding to each product specification in the mid-term cycle; S302. Confirm the business characteristics associated with each product specification; S303: According to the business characteristics associated with each product specification, the corresponding historical in-and-out warehouse data is input into the moving average model, LSTM model or XGBoost model for prediction, and the future in-and-out warehouse data corresponding to each product specification in the medium-term cycle is output.

[0042] In this embodiment, the medium-term cycle can be a weekly cycle, and the historical inventory data of each product specification in the weekly cycle is obtained, that is, 7 days of historical inventory data. The number of cycles is not limited and can be set according to actual needs.

[0043] In this embodiment, in the inventory forecast for the next time period based on a weekly cycle, three inventory forecast models based on a weekly cycle are provided, and the corresponding inventory forecast model needs to be selected according to the business characteristics associated with the product specifications.

[0044] Model 1: Moving average model, applicable to business features including: Feature 1: The inventory in and out patterns are relatively stable with little fluctuation; Feature 2: Historical data is strongly correlated with future predictions; Feature 3: The forecast period is short (7-14 days); Feature 4: Focus mainly on recent trends.

[0045] Specifically, the basic formula of the moving average model is: ;

[0046] Among them, F t represents the predicted value on day t; w i represents the weight coefficient of the i-th historical data; X t-i Represents ti days of historical data.

[0047] Specifically, the input data of the moving average model corresponds to the model parameters as follows: 1) Basic time series data (in and out history table); 1.1) Date-quantity pairs: inbound and outbound records in the last 4 weeks; 1.2) Daily summary data: calculate the daily total; 1.3) Data frequency: daily; 2) Weight reference data (special event table); 2.1) Working day / holiday marking: adjust the weight coefficient; 2.2) Seasonal marking: identifying special periods; Important parameter settings (the following parameter values ​​are for reference only and should be adjusted according to actual business scenarios. This implementation does not make specific restrictions): The historical data window can be 28 days (4 weeks), and the experience value can be adjusted according to the stability of the data; Weight allocation: The reference value set based on experience should be dynamically adjusted according to actual data performance. For example, the reference value for the same period last week was 0.4; the reference value for the same period last month was 0.3; and the reference value for special events was 0.3.

[0048] Model 2: The LSTM model is applicable to the following business features: Feature 1: The warehousing and outbound modes are complex, with multiple dependencies; Feature 2: Sufficient historical data (at least 3 months); Feature 3: Need to capture long-term dependencies; Feature 4: High requirements for prediction accuracy.

[0049] Specifically, the LSTM model input data requirements are: 1) Time series features (in and out inventory history table); 1.1) Sliding window data: 14 consecutive days of historical records; 1.2) Data form: daily inventory in and out quantity sequence; 1.3) Purpose: as the main input sequence of LSTM; 2) Time feature vector: 2.1) Weekday encoding: one-hot encoded date type; 2.2) Period position encoding: position within a week (1-7); 2.3) Monthly position code: position within the month (1-31); 3) Trend characteristics: 3.1) Short-term moving average: MA(7); 3.2) Medium-term moving average: MA(14); 3.3) Month-on-month growth rate: (y t -y t-7 ) / y t-7 .

[0050] Specifically, business constraint data (system status table): The storage capacity utilization rate is a real number in the interval [0,1]; The equipment load rate is a real number in the interval [0,1]; Safety stock level is a positive real number; Important parameter settings (the following parameter values ​​are for reference only and should be adjusted according to actual business scenarios. This implementation does not make specific restrictions): The input sequence length is 14 days, based on the effective dependency period verified by the experiment; The hidden layer dimension is 128, which can be adjusted according to the number of features and computing resources; The number of layers is 2, which is an empirical value that balances model complexity and computational efficiency; The dropout rate is 0.2, an empirical parameter to prevent overfitting.

[0051] Model 3: The usage scenarios of the XGBoost model are applicable to the following business features: Feature 1: There are multiple influencing factors; Feature 2: Nonlinear relationships need to be considered; Feature 3: There is a requirement for model interpretability; Feature 4: Computing resources are limited.

[0052] Specifically, the input data requirements of the XGBoost model are: 1) Basic features (warehouse entry and exit history table); 1.1) Historical data characteristics: daily inventory volume for the past 7 days; moving average for the past 7 days; standard deviation for the past 7 days; 1.2) Time features: day of the week: categorical variable (1-7); whether it is a weekday: Boolean value; whether it is the beginning / end of the month: Boolean value; 1.3) Trend characteristics: Weekly year-on-year growth rate; Monthly year-on-year growth rate; Moving average slope; 2) Business constraint characteristics: 2.1) Storage capacity related: current storage capacity utilization rate; storage capacity change rate in the past 7 days; 2.2) Equipment related: equipment utilization rate; task backlog rate; average task completion time; Important parameter settings (the following parameter values ​​are for reference only and should be adjusted according to actual business scenarios. This implementation does not make specific restrictions): The maximum tree depth is 4-6, determined by feature complexity; The feature sampling ratio is 0.8, an empirical value, to prevent overfitting; The learning rate is 0.1, a commonly used empirical value; The minimum split gain is 0.1 and can be adjusted according to the data noise level.

[0053] Specifically, the prediction results are output: The daily inbound and outbound quantities of each product ID in the next 7 days are predicted, and the output results are: ; Where: t represents the forecast date, from 1 to 7, representing the next 7 days; T t Indicates the trend value on the tth day, reflecting the long-term trend; S t represents the seasonal value on the tth day, reflecting the periodic changes; R tRepresents the random item on day t, with a probability distribution described by the predicted mean and standard deviation; for each specification ID, a set of such prediction sequences will be generated.

[0054] Get predictions from each model: T t It indicates that the trend extraction method based on each model is obtained; Moving average means using a sliding window weighted average; LSTM represents long-term trends captured by hidden states; XGBoost represents the trend feature contribution in feature importance ranking; S t It indicates that the seasonal identification method based on each model is obtained; Moving average represents a weighted combination of data from the same period; LSTM represents the hidden representation of periodic patterns; XGBoost represents the periodic contribution of time features; R t It indicates that the error estimation method based on each model is obtained; μ t Represents the direct prediction value of the model; σ t Represents the standard deviation calculated based on historical forecast errors.

[0055] The final predicted daily inventory quantity is: .

[0056] In one embodiment, if Figure 4 As shown, step S102 includes: S401. Obtain the historical in-and-out data of each product specification in the short-term cycle; S402. Confirm the business characteristics associated with each product specification; S403: According to the business characteristics associated with each product specification, the corresponding historical inbound and outbound data are input into the LightGBM model or the Kalman filter model for prediction, and the future inbound and outbound data corresponding to each product specification in the short term cycle are output.

[0057] In this embodiment, the short-term cycle can be a daily cycle, and the historical inventory data of each product specification in the daily cycle is obtained, that is, the historical inventory data of 1 day. The number of cycles is not limited and can be set according to actual needs.

[0058] In this embodiment, in the inventory forecast for the next time period with a daily period, two inventory forecast models with a weekly period are provided, and the corresponding inventory forecast model needs to be selected according to the business characteristics associated with the product specifications.

[0059] Model 1: LightGBM model, the business characteristics applicable to the use scenario include: Feature 1: Real-time and fast prediction is required; Feature 2: Data features are numerous and sparse; Feature 3: Sensitive to computing resources; Feature 4: Short-term forecast (within 24 hours); The basic formula of the LightGBM model: ; Among them, F t Represents the predicted value at the tth time point; Tree i represents the i-th decision tree; x t The input feature vector representing the time instant.

[0060] Specifically, the input data requirements of the LightGBM model are: 1) Real-time sequence features (in and out inventory history table); 1.1) Recent data: the volume of inflows and outflows in the past 24 hours; hourly moving average; real-time growth rate; 1.2) Time characteristics: hour code: (0-23); whether it is peak time: Boolean value; distance from the start time of the inbound and outbound shift: number of hours; 1.3) Task characteristics: number of current backlog tasks; task priority distribution; average execution time; 2) Real-time status characteristics (system status table) 2.1) Equipment status: used to evaluate the current task execution capability; 2.11) Current position coordinates: [x, y, z] three-dimensional coordinates; 2.2) Feature form: Manhattan distance to the target cargo location; 2.21) Running speed: three-axis real-time speed vector; 2.3) Characteristic form: velocity scalar and its moving average; 2.31) Energy consumption level: current power consumption; 2.4) Feature form: normalized energy consumption ratio [0,1]; 2.5) Storage location status: used to evaluate storage space utilization efficiency; 2.51) Occupancy rate of adjacent cargo spaces: the usage of n adjacent cargo spaces; 2.6) Feature form: occupancy percentage [0,1]; 2.61) Preferred area availability: the idleness of the efficient access area; 2.7) Characteristic form: regional availability vector; 2.71) Distribution of distances between warehouse entrances and exits: statistics on the distances from each cargo location to the warehouse entrances and exits; 2.8) Feature form: statistical characteristics of distance distribution (mean, variance, etc.); Important parameter settings (the following parameter values ​​are for reference only and should be adjusted according to actual business scenarios. This implementation does not make specific restrictions): The number of leaves is 31, which is an empirical reference value and can be adjusted according to the data scale; The learning rate is 0.1, a commonly used empirical value; The feature binning is 255, the default empirical value; The minimum data volume is 100, an empirical value determined based on real-time requirements.

[0061] Model 2: Kalman filter model, applicable to business features including: Feature 1: Real-time state estimation is required; Feature 2: System status can be modeled; Feature 3: There is measurement noise; Feature 4: Very short-term prediction (hourly level); The basic formula of Kalman filter model: ; .

[0062] Among them, X t represents the state vector at the moment; u t represents the control input; z t represents the observed value; W t 、v t represent system noise and observation noise respectively; A, B, H represent state transfer matrix, control matrix and observation matrix respectively.

[0063] Input data requirements for the Kalman filter model: 1) State vector (system state table); 1.1) Basic state quantity: current inbound and outbound rate; task backlog; equipment location; 1.2) Control quantity: number of planned tasks; priority setting; 2) Observation data (in and out storage history table); 2.1) Real-time recording; Important parameter settings (the following parameter values ​​are for reference only and should be adjusted according to actual business scenarios. This implementation does not make specific restrictions): The status dimension can be determined based on monitoring indicators, without fixed reference values; The observation dimension can be determined based on the number of sensors, without a fixed reference value; The update frequency is minute-level, determined according to the real-time requirements of the system; The prediction step is 1-24 hours, determined according to short-term business needs.

[0064] Specifically, the prediction results are output: The hourly inbound and outbound quantities of each product ID in the next 24 hours are predicted, and the output results are: ; Among them, t represents the prediction time point, from 1 to 24, representing the next 24 hours; T t Indicates the trend value of the tth hour, reflecting the short-term trend; S t represents the seasonal value at hour t, reflecting the intraday cycle changes; R t Represents the random item at hour t, with a probability distribution described by the predicted mean μ_t and standard deviation σ_t; for each specification ID, a set of such prediction sequences will be generated.

[0065] Specifically, get the prediction results from each model: T t It indicates that the trend extraction method based on each model is obtained; LightGBM represents the trend feature contribution in feature importance ranking; Kalman filtering represents the trend component of state estimation; S t It indicates that the periodicity identification method based on each model is obtained; LightGBM represents the periodic contribution of temporal features; Kalman filtering represents the estimated value of the periodic state variable; R t It indicates that the error estimation method based on each model is obtained; μ t Represents the direct prediction value of the model; σ t Represents the standard deviation calculated based on the prediction errors.

[0066] The final predicted hourly inbound and outbound quantities are: .

[0067] In one embodiment, if Figure 5 As shown, step S103 includes: S501. Estimate the in-and-out quantity of each product specification in the next period according to the future in-and-out data corresponding to each product specification in a preset time period, and generate a transfer task for the corresponding product specification; S502, determine the outbound product specifications and quantity in the warehouse transfer task, and control the stacker to move the corresponding product specifications in the area far from the outbound platform or the area close to the inbound platform on the cargo lane to the outbound buffer zone during the electric wave valley time period; S503, determine the total quantity of incoming goods in the warehouse transfer task, and control the stacker to move the corresponding quantity and specifications in the incoming goods buffer to the area close to the incoming goods buffer on the cargo lane during the electric wave valley time period.

[0068] In this embodiment, based on the estimated specifications and quantity of the outbound goods in the next time period, the stacker is controlled to establish a warehouse transfer task in whole or in part (when the outbound buffer is insufficient, some specifications and quantities are randomly dispersed) during the radio wave valley time period. When selecting the source cargo location of the outbound goods, the cargo location in the same cargo location lane that is far away from the outbound platform or close to the inbound platform is preferred, and moved to the outbound buffer. In this way, when the outbound buffer is subsequently outbound to the outbound platform, the moving distance of the stacker during the outbound process can be minimized to improve the outbound efficiency. At the same time, performing the warehouse transfer operation during the radio wave valley time period can also effectively utilize electricity resources and reduce energy consumption costs. In addition, when the outbound buffer capacity is insufficient, by randomly dispersing and selecting some specifications and quantities for warehouse transfer, the smooth progress of the outbound task can be ensured, and outbound delays caused by buffer capacity limitations can be avoided.

[0069] In this embodiment, based on the estimated total number of incoming goods in the next time period (no product specifications are required), a transfer task is established for part (the same as the predicted total number of incoming goods) or all (when the number of cargo locations in the incoming buffer is less than the predicted total number of incoming goods) of the goods on the cargo locations in the incoming buffer. When selecting the target cargo location for the goods, priority is given to cargo locations in the same cargo location lane, close to the current location, and not used as a buffer zone; then the goods (product specifications) in the incoming buffer are transferred to the selected target cargo location, thereby freeing up cargo locations in the incoming buffer zone. In this way, when the goods are subsequently transferred from the incoming platform to the incoming buffer zone, the moving distance of the stacker during the incoming storage process can be minimized, thereby improving the efficiency of outbound storage. At the same time, performing the transfer operation during the electric wave valley time period can also effectively utilize power resources and reduce energy consumption costs.

[0070] In this embodiment, the stacker access efficiency optimization method based on time series prediction and real-time status monitoring can significantly improve the operational efficiency and accuracy of the warehouse.

[0071] In one embodiment, if Figure 6 As shown, the method for optimizing the storage and retrieval efficiency of a stacker based on a time series further includes: S601. Confirm the future inbound and outbound data of each product specification corresponding to the preset time period, where the preset time period includes a long-term period, a medium-term period and a short-term period; S602: Perform top-down transfer constraint processing on the prediction results of future inbound and outbound data of each product specification according to the order of long-term cycle, medium-term cycle and short-term cycle, so as to adjust the prediction results of each product specification at different cycle scales.

[0072] In this embodiment, the role of steps S601-S602 is to ensure the consistency of prediction results at different time scales, prevent short-term predictions from deviating from long-term trends, and constrain the prediction range of each level to avoid divergence of prediction results. Steps S601-S602 adopt top-down transfer constraint processing, which is a method of predicting data integration. First, a rough framework is formulated based on the prediction results of the long-term cycle, and then this framework is passed as a constraint to the prediction of the medium-term cycle, and the prediction results of the medium-term cycle are passed as constraints to the prediction of the short-term cycle. In this way, it can be ensured that the prediction results of different time periods are coordinated and consistent with each other to avoid contradictions. In short, this process is to first predict a large trend over a longer period of time, and then use this trend as a basis to guide and restrict the specific predictions in the next shorter period of time, so that the prediction results remain consistent and reasonable in different time periods.

[0073] Specifically, input data: 1) Long-term forecast results (monthly, 30 days): 1.1) Basic prediction value: ; 1.2) Decomposition results: 1.21) Trend term T t : reflects the 30-day change trend; 1.22) Seasonal term S t : reflects periodic changes; 1.23) Random term R t : reflects uncertainty, including mean μ t and standard deviation σ t ; 1.3) Confidence Interval: ; 1.4) Forecast accuracy index: ; 2) Medium-term forecast results (weekly, 7 days): 2.1) Basic prediction value: ; 2.2) Decomposition results: 2.21) Trend term T t : reflects the 7-day change trend; 2.22) Seasonal term S t : reflects the distribution characteristics within the week; 2.23) Random term R t: reflects uncertainty, including mean μ_t and standard deviation σ_t; 2.3) Confidence Interval: ; 2.4) Forecast accuracy index: ; 3) Short-term prediction results (hourly, 24 hours): 3.1) Basic prediction value: ; 3.2) Decomposition results: 3.21) Trend term T t : Reflects the 24-hour change trend; 3.22) Seasonal term S t : reflects the intraday distribution characteristics; 3.23) Random term R t : reflects uncertainty, including mean μ t and standard deviation σ t ; 3.3) Confidence Interval: ; 3.4) Forecast accuracy index: .

[0074] Specifically, the output results are: 1) Medium-term forecast constraints: Medium-term forecast adjustment value = medium-term basic forecast value × long-term trend constraint coefficient; in: ; ΔT long =(T t -T t-1 ) / T t-1 Long-term trend rate of change; ΔS long =(S t -S t-1 ) / S t-1 long-term seasonal rate of change; W trend =0.6 / / Trend weight; W season =0.4 / / Seasonal weight; It should be noted that the above weight values ​​are for reference only and should be adjusted according to actual business scenarios, and are not specifically limited in this embodiment.

[0075] Specifically, the constraints are: 1) Weekly total quantity constraints: Σ(mid-term forecast adjustment value) ≤ monthly forecast value / 4 × (1 + tolerance_factor); Among them, tolerance_factor = 0.1, allowing 10% fluctuation, based on the stability characteristics of monthly data; 2) Fluctuation range constraints: |Medium-term forecast adjustment value_t-medium-term forecast adjustment value_{t-1}| ≤ max_daily_change; Where max_daily_change = the 95% quantile of the historical daily change rate; 3) Short-term forecast constraints: Short-term forecast adjustment value = short-term basic forecast value × medium-term constraint coefficient; in: ; ΔS mid = (T t -T t-1 ) / T t-1 Medium-term trend rate of change; ΔS mid = (S t -S t-1 ) / S t-1 mid-term pattern change rate; W mid =0.7 / / Medium-term trend weight; W pattern =0.3 / / Intraday mode weight; It should be noted that the above weight values ​​are for reference only and should be adjusted according to actual business scenarios, and are not specifically limited in this embodiment.

[0076] Specifically, the constraints are: 1) Daily total amount constraints: Σ(short-term forecast adjustment value) ≤ weekly forecast value / 7 × (1 + tolerance_factor); Among them, tolerance_factor = 0.15, allowing 15% fluctuation, considering the large intraday fluctuations; 2) Time period distribution constraints: Short-term forecast adjustment value_t ∈ [min_hourly, max_hourly]; The intervals are based on the [5%, 95%] quantiles of historical data for the same period.

[0077] In one embodiment, the method for optimizing the storage and retrieval efficiency of a stacker based on a time series further includes: Based on the current business constraints of the warehouse, the future inbound and outbound data of each product specification corresponding to the preset time period is horizontally supplemented in the time dimension and the business dimension to optimize the prediction results of the future inbound and outbound data of each product specification corresponding to the preset time period.

[0078] This embodiment considers business constraints to adjust the prediction, which can improve the stability and reliability of the prediction.

[0079] Specifically, input data: 1) Prediction results at each level: 1.1) Long term (monthly): 1.11) Basic prediction sequence: {y_t}, t∈[1,30]; 1.12) Forecast accuracy: MAPE_long; 1.13) Forecast stability: RMSE_long; 1.2) Medium term (weekly): 1.21) Basic prediction sequence: {y_t}, t∈[1,7]; 1.22) Prediction accuracy: MAPE_mid; 1.23) Prediction stability: RMSE_mid; 1.3) Short term (hourly): 1.31) Basic prediction sequence: {y_t}, t∈[1,24]; 1.32) Forecast accuracy: MAPE_short; 1.33) Forecast stability: RMSE_short; 2) Business status indicators: 2.1) Storage capacity utilization rate: capacity_rate ∈ [0,1]; Calculation method: number of used cargo spaces / total cargo spaces; Example: A stacker manages 1,000 cargo locations, 850 of which are currently in use; capacity_rate = 850 / 1000 = 0.85; 2.2) Equipment load rate: load_rate ∈ [0,1]; Calculation method: actual working time per hour / 60 minutes; Example: The stacker actually runs for 45 minutes in a certain hour; load_rate = 45 / 60 = 0.75; 2.3) Task backlog rate: backlog_rate ∈ [0,1]; Calculation method: number of tasks to be executed / maximum number of tasks processed per hour; Example: There are currently 20 tasks to be executed, and the stacker can process a maximum of 30 tasks per hour; backlog_rate = 20 / 30 = 0.67.

[0080] Specifically, the output results are: 1) Time dimension complementarity: Complementary forecast value_t = w_long × long-term decomposition value_t + w_mid × mid-term decomposition value_t + w_short × short-term forecast value_t; The weight calculation is: w_long = exp(-λ_long × MAPE_long) / Z; w_mid = exp(-λ_mid × MAPE_mid) / Z; w_short = exp(-λ_short × MAPE_short) / Z; Z = exp(-λ_long × MAPE_long) + exp(-λ_mid × MAPE_mid) + exp(-λ_short × MAPE_short); λ_long = 0.5 / / Long-term prediction attenuation coefficient; λ_mid = 0.3 / / mid-term forecast attenuation coefficient; λ_short = 0.2 / / Short-term prediction decay coefficient; It should be noted that the selection of the attenuation coefficient is based on the characteristics of the prediction time scale. The impact of long-term prediction deviation decays quickly, while the impact of short-term prediction deviation lasts for a long time. The specific values ​​are for reference only and are not included in the scope of patent protection; Specifically, the prediction interval is: Upper limit = complementary prediction value_t × (1 + margin_factor); Lower limit = complementary prediction value_t × (1 - margin_factor); Among them, margin_factor=0.1+0.2×(1-min(MAPE_long, MAPE_mid, MAPE_short)); It should be noted that the calculation method of margin_factor takes into account the impact of prediction accuracy on the confidence interval; 2) Complementarity in business dimensions: Business adjustment factor = 1 + w_time × (1 - load_rate) + w_energy × (1 -capacity_rate) - w_backlog × backlog_rate; Where: w_time = 0.5 / / time weight; w_energy = 0.3 / / Energy consumption weight; w_backlog = 0.2 / / backlog weight; It should be noted that the weight value is set based on the service priority, and the specific value is for reference only and is not specifically limited in this embodiment; Final forecast value = complementary forecast value_t × business adjustment coefficient; Constraint: 0.7 ≤ Business Adjustment Factor ≤ 1.3 / / Limit the adjustment range to avoid over-correction.

[0081] In one embodiment, if Figure 7 As shown, the method for optimizing the storage and retrieval efficiency of a stacker based on a time series further includes: S701, real-time monitoring of the forecast error of future inbound and outbound data of each product specification corresponding to a preset time period, where the preset time period includes a long-term period, a medium-term period, and a short-term period; S702, real-time monitoring of the warehouse's task completion rate, equipment utilization rate, and business indicator data of equipment utilization rate; S703, real-time monitoring of the prediction status of the inbound and outbound prediction model corresponding to each preset time period; S704: Based on the prediction error, business indicator data and prediction status, each inbound and outbound prediction model is updated with feedback from the bottom up.

[0082] The purpose of this embodiment is to: timely capture the changing trend and prevent the accumulation of erroneous prediction values ​​from causing excessive deviations; dynamically adjust the prediction model parameters; and maintain the adaptability of the prediction system.

[0083] Specifically, input data: 1) Real-time monitoring data: 1.1) Forecast error series: ; 1.2) Error statistical characteristics: mean: mean(e_t); standard deviation: std(e_t); skewness: skewness(e_t); 2) Business indicator data: 2.1) Task completion rate = number of completed tasks / total number of planned tasks; Example: 100 tasks were planned to be completed, and 85 were actually completed; task completion rate = 85 / 100 = 0.85; 2.2) Equipment utilization = actual working time / total available time; Example: In 8 working hours, the actual operation time is 6.5 hours; equipment utilization rate = 6.5 / 8 = 0.81; 2.3) Energy consumption level = actual energy consumption / benchmark energy consumption; The baseline energy consumption indicates the historical average energy consumption under the same task volume; for example: the baseline energy consumption is 100 degrees, and the actual consumption is 90 degrees; the energy consumption level = 90 / 100 = 0.9; 3) Prediction model status: 3.1) Prediction accuracy of models at each level; 3.2) Model parameter update time; 3.3) The time of the most recent retraining.

[0084] Specifically, the output results are: 1) Short- to medium-term feedback: ; Where: k 1 = min(0.3, max(0.1, exp(-λ × MAPE_short))); λ = 0.5 / / attenuation coefficient; It should be noted that k 1 The value range and calculation method of take into account the impact of prediction accuracy on the correction strength.

[0085] Specifically, the short-term cumulative deviation rate is calculated as follows: 1) Calculate the standardized error: z_t = (e_t - mean(e_t)) / std(e_t); 2) Calculate the weighted cumulative sum: cum_error = Σ(w_t × z_t), t∈last 24 hours; w_t = exp(-0.1 × (t_now - t)) / / time decay weight; 3) Deviation rate = cum_error / Σ(w_t).

[0086] Specifically, the trigger conditions are: 1) |Deviation rate|>0.2 or the deviations are in the same direction for three consecutive periods; 2) Medium to long-term feedback: ; in: ; λ = 0.3 / / attenuation coefficient; It should be noted that The value range is less than , reflecting the low sensitivity of long-term forecasts to short-term fluctuations.

[0087] Specifically, the medium-term trend change rate is calculated as follows: 1) Calculate weekly forecast error: week_error_t = (actual weekly total - predicted weekly total) / predicted weekly total; 2) Calculate trend changes: trend_change = (week_error_t- week_error_{t-1}) / |week_error_{t-1}|; Trigger conditions: |trend_change|>0.15 or the trend is in the same direction for 2 consecutive weeks; 3) Model update strategy: if any of the following conditions are met: Condition 1: |Short-term cumulative deviation rate|>0.3; Condition 2: |Medium-term trend change rate|>0.2; Condition 3: The evaluation index deteriorates for 5 consecutive forecast cycles: MAPE>historical_MAPE × 1.2 or RMSE>historical_RMSE × 1.2 or MAE>historical_MAE × 1.2; Then, trigger the parameter update of the corresponding level model; if the effect is still not improved after the parameter update, trigger the model retraining; update all evaluation indicators (MAPE, RMSE, MAE).

[0088] In one embodiment, if Figure 8 As shown, the method for optimizing the storage and retrieval efficiency of a stacker based on a time series further includes: S801. Confirm the in-and-out quantity of each product specification in the next period; S802, confirm the current inventory status in the warehouse; S803. During the radio wave valley period, move the products with high outbound quantity in the next period to an area close to the outbound platform and easy to get; S804, during the radio wave valley period, move the product specifications with high storage quantity in the next period to an area close to the storage platform; S805. During the radio wave valley period, move the products in the warehouse that have no short-term demand for shipment to an area that is far away from the shipment platform and not easily accessible.

[0089] In this embodiment, the core of steps S801-S805 is to solve a problem: knowing how much goods of each specification will be shipped in / out in the next time period, and how to arrange the cargo space in advance to maximize the efficiency of warehousing.

[0090] The specific technical ideas are as follows: 1) Input parameters are: the predicted outbound quantity of each product specification in the future; the predicted inbound quantity of each product specification in the future; the current inventory status (which storage locations have goods and what product specifications are placed); 2) Goal setting: The primary goal is to reduce the stacker crane’s operating time (improve efficiency); the secondary goal is to reduce energy consumption (save costs); 3) Strategy implementation: For products that are predicted to be shipped out frequently, they can be placed as close to the shipping port as possible and in a location that is easy to access (not blocked by other goods). 4) For products with a large predicted incoming volume: reserve appropriate cargo space in advance; the reserved cargo space should be close to the incoming warehouse; 5) For products that do not need to be shipped out in the short term: they can be placed in a farther location; they can be placed in a location that is not easy to access; 6) Specific implementation: During the time when electricity prices are low and not busy (i.e., the off-peak period), adjust the cargo positions through warehouse transfer tasks to prepare for the upcoming peak in warehouse entry and exit; 7) The application of collaborative results ultimately outputs two key priorities: storage location priority; task execution priority.

[0091] Based on the above technical ideas 1)-7), the specific implementation process is as follows: Input data: 1) Multi-level prediction results: 1.1) Long-term trend (monthly): Forecast sequence: {y_t}, t∈[1,30]; Confidence interval: [lower_t,upper_t]; Trend characteristics: growth rate, cyclical intensity; 1.2) Medium-term model (weekly): Forecast sequence: {y_t}, t∈[1,7]; Intraday distribution pattern; Volatility characteristics; 1.3) Short-term state (hourly): Prediction sequence: {y_t}, t∈[1,24]; Real-time adjustment suggestions; Confidence level; 2) System operation status: 2.1) Storage status: cargo space usage distribution map; regional utilization statistics; cargo turnover speed; 2.2) Equipment status: real-time location and speed; energy consumption level; maintenance plan; 2.3) Task status: queue to be executed; priority distribution; completion rate statistics.

[0092] Based on the above implementation process, the specific output results are: 1) Storage optimization solution: Optimization goal: min(w_time × task completion time + w_energy × energy consumption cost + w_backlog × backlog cost); Where: w_time = 0.5 / / time weight; w_energy = 0.3 / / Energy consumption weight; w_backlog = 0.2 / / backlog weight; It should be noted that the weight value is set based on the service priority, and the specific value is for reference only and is not specifically limited in this embodiment; Backlog cost = backlog_rate × backlog cost per unit time; Backlog cost per unit time = average task value × opportunity cost rate per unit time.

[0093] Specifically, the constraints are: 1.1) Capacity constraints: Σ(used cargo space + reserved cargo space) ≤ total cargo space × max_utilization; max_utilization = 0.95 / / Maximum utilization, taking into account system operation redundancy; 1.2) Distance constraint: Task completion time ≤ historical average time × (1 + time_tolerance); time_tolerance = 0.2 / / Allow 20% time increase based on actual operation experience; 1.3) Energy Consumption Constraints: Peak energy consumption ≤ Historical peak energy consumption × 0.9 / / Reduce peak energy consumption by 10%, taking energy saving goals into consideration; 1.4) Backlog Constraints: backlog_rate ≤ max_backlog_rate; max_backlog_rate = 0.3 / / Maximum allowed backlog rate, based on system processing capacity.

[0094] 2) Task execution plan: ; in: = 0.4 / / distance weight; = 0.4 / / time weight; = 0.2 / / Energy consumption weight; It should be noted that the weight values ​​are allocated based on the characteristics of the task, and the specific values ​​are for reference only and are not the scope of patent protection; Distance cost = actual distance / maximum distance / / normalized; Time urgency = exp(-λ× remaining time) × (2 - confidence_level) / / Consider the prediction confidence level; Energy consumption cost = current electricity price / maximum electricity price / / normalized; Where: confidence_level is the confidence level of short-term prediction, ranging from [0,1]; λ = 0.1 / / Time decay factor.

[0095] 2.1) Task classification: Urgent tasks: priority > 0.8; Regular tasks: 0.4 ≤ priority ≤ 0.8; Low priority tasks: priority < 0.4.

[0096] 2.2) Task scheduling constraints: Backlog Control: When backlog_rate>0.2, increase the threshold of urgent tasks; Urgent tasks: priority>0.7; Regular tasks: 0.3 ≤ priority ≤ 0.7; Low priority tasks: priority < 0.3.

[0097] 2.3) Prediction confidence constraints: When confidence_level<0.6, expand the task classification range: Urgent tasks: priority>0.75; Regular tasks: 0.35 ≤ priority ≤ 0.75; Low priority tasks: priority < 0.35.

[0098] 2.4) Dynamic adjustment mechanism: Trigger conditions: 2.41) Forecast Bias: |Actual value - predicted value|> predicted value × 0.2 / / Standardized prediction deviation threshold; 2.42) Continuous Deviation: Forecast deviation for 3 consecutive periods > 10% / / Continuous deviation threshold; 2.43) Status changes: 2.431) Equipment utilization rate change > 20% / / Equipment status significant change threshold; 2.432) Storage capacity utilization rate change > 15% / / Storage status significant change threshold; 2.433) Task backlog increases > 30% / / Task status significant change threshold; It should be noted that the above thresholds are set based on system stability requirements, and the specific values ​​are for reference only and are not specifically limited in this embodiment.

[0099] Based on the above output, adjust the objects and policies: 1) Task priority calculation parameter adjustment: 1.1) Time Weight : For every 10% increase in the backlog rate, the weight increases by 5% (upper limit +20%); 1.2) Energy consumption weight : The maximum difference between peak and valley periods is ±15%; 1.3) Task classification threshold: For every 20% increase in system pressure, the emergency task threshold is reduced by 5%; 2) Storage optimization weight adjustment: 2.1) Peak period: w_time increases to 0.65 at the highest and w_energy drops to 0.2 at the lowest; 2.2) Low period: w_energy increases to 0.45 at the highest, and w_time drops to 0.35 at the lowest; 2.3) When the backlog is serious (>50%): w_backlog increases to a maximum of 0.3; 3) Constraint adjustment: 3.1) max_utilization: dynamically adjusted by ±10% based on turnover rate; 3.2) time_tolerance: adjusted by ±15% based on task completion rate; 3.3) max_backlog_rate: Adjust ±20% based on processing capacity.

[0100] Specifically, adjust the constraints: 1) Smoothness requirements: 1.1) Single adjustment step length ≤ 5%; 1.2) Minimum adjustment interval: 30 minutes; 1.3) Maximum cumulative adjustment within 24 hours: ±30%; 2) Stability requirements: 2.1) The sum of weights remains 1; 2.2) Maintain the basic proportional relationship of each parameter; 2.3) Avoid parameter oscillation.

[0101] Based on this, the ultimate goal of steps S801-S805 is to achieve: optimize storage efficiency: guide cargo space allocation through prediction results; save energy: use electricity price troughs to perform warehouse transfer operations; improve response speed: make real-time adjustments to deal with emergencies.

[0102] In one embodiment, if Fig. 9 As shown, the method for optimizing the storage and retrieval efficiency of a stacker based on a time series further includes: S901. Regularly calculate the accuracy score, stability score, and business relevance score of each inbound and outbound forecast model corresponding to each preset time period, and obtain a comprehensive score of each inbound and outbound forecast model after weighted calculation; S902. For multiple inbound and outbound prediction models corresponding to each preset time period, select the inbound and outbound prediction model with the highest comprehensive score and stable performance as the current prediction model corresponding to each preset time period, and select other inbound and outbound prediction models as candidate models corresponding to each preset time period; S903: Count the prediction errors of the current prediction model corresponding to each preset time period, and trigger model retraining or replace the candidate model with the second highest comprehensive score when the prediction error exceeds a preset threshold.

[0103] In this embodiment, the core goal of steps S901-S903 is to build a scientific model evaluation and selection mechanism to ensure that the most appropriate prediction model can be selected in different scenarios. Specifically, it includes four parts: evaluation index system, model selection process, model evaluation cycle and model switching mechanism.

[0104] Specifically, in the evaluation indicator system of the first part: 1) Basic evaluation indicators include: levels, main indicators, secondary indicators, and business indicators. Corresponding to each period, they include the following indicators; 1.1) Long-term, MAPE, RMSE and storage capacity utilization; 1.2) Mid-term, MAE, RMSE and cargo position matching; 1.3) Short-term, RMSEE, MAE and task completion time; The definition and specific implementation of basic indicators are not repeated in the present invention, where MAPE represents mean absolute percentage error; RMSE represents root mean square error; RMSE_normalized represents normalized root mean square error; MAE represents mean absolute error.

[0105] 2) Comprehensive score (Score) calculation: Score = w_acc × AccuracyScore + w_stab × StabilityScore + w_bus × BusinessScore; Where: AccuracyScore = 1 - MAPE / / Accuracy score; StabilityScore = 1 - RMSE_{normalized} Note: Stability score; BusinessScore = \frac{Number of correctly predicted business peak periods}{Total number of peak periods} Note: Business relevance score; Weight distribution: w_acc = 0.5 / / accuracy weight; w_stab = 0.3 / / stability weight; w_bus = 0.2 / / business relevance weight; It should be noted that the above weight values ​​are for reference only and should be adjusted according to actual business scenarios, and this embodiment does not make specific limitations; Score ∈ [0,1], the higher the score, the better the overall performance of the model.

[0106] 3) Example of scoring criteria: Excellent: Score ≥ 0.85; Good: 0.75 ≤ Score<0.85; Pass: 0.65≤ Score<0.75; To be optimized: Score<0.65.

[0107] Specifically, in the model selection process of the second part: 1) Long-term forecast (monthly): Main model: Prophet; Alternative model: SARIMA; Selection criteria: highest score value and MAPE < 20%; 2) Medium-term forecast (weekly): Main model: LSTM; Alternative model: XGBoost; Selection criteria: highest score value and MAPE < 15%; 3) Short-term prediction (hourly): Main model: LightGBM; Alternative model: Kalman filter; Selection criteria: highest score value and MAPE < 10%.

[0108] Specifically, in the model evaluation cycle of the third part: 1) Regular evaluation: Long-term model: monthly evaluation; mid-term model: weekly evaluation; short-term model: daily evaluation; 2) Trigger evaluation: Evaluate immediately if: Case 1: The prediction error exceeds the threshold: MAPE>historical_MAPE×1.2; Case 2: Business indicators change significantly: storage capacity utilization rate changes by >15%; equipment load rate changes by >20%; task backlog rate changes by >30%.

[0109] Specifically, the model switching mechanism in the fourth part is set up with: 1) Switching conditions: The new model score exceeds the current model by more than 5%; the new model performs stably during the validation period (3 days); 2) Smooth switching: Run the new and old models in parallel for 3 days; compare the predicted results with the actual values; and officially switch after confirming stability; 3) Rollback mechanism: save the model status before switching; automatically roll back if the performance of the new model deteriorates; record the switching history for subsequent optimization.

[0110] The embodiment of the present invention further provides a device for optimizing the access efficiency of a stacker based on a time series, and the device for optimizing the access efficiency of a stacker based on a time series is used to execute any embodiment of the method for optimizing the access efficiency of a stacker based on a time series. Fig.10 , Fig.10 It is a schematic block diagram of a device for optimizing the storage and access efficiency of a stacker based on time series provided by an embodiment of the present invention.

[0111] like Fig.10 As shown, the device 1000 for optimizing the storage and access efficiency of a stacker based on time series includes: an area setting unit 1001, an in-and-out storage prediction unit 1002, and an in-and-out storage transfer unit 1003.

[0112] The area setting unit 1001 is used to set a corresponding inbound and outbound buffer zone in each cargo lane according to the position of the inbound and outbound platform connected to the stacker in each cargo lane; The in-and-out forecasting unit 1002 is used to obtain the historical in-and-out data of each product specification corresponding to a preset time period, and input it into the in-and-out forecasting model corresponding to the preset time period for forecasting, and output the future in-and-out data of each product specification corresponding to the preset time period; The in-and-out transfer unit 1003 is used to generate transfer tasks according to the future in-and-out data corresponding to each product specification in a preset time period, and control the stacker to perform in-and-out transfer operations between the corresponding storage location aisles, inbound buffers and outbound buffers according to the transfer tasks.

[0113] The device predicts the inbound and outbound tasks in the next time period, generates corresponding warehouse transfer tasks based on the prediction results, and then uses the inbound and outbound buffer to transfer and sort the goods to be inbound and outbound in advance. This can effectively reduce the running distance of the stacker when performing inbound and outbound tasks in the next time period, thereby significantly improving the stacker's inbound and outbound operation efficiency.

[0114] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0115] The above-mentioned device for optimizing the stacker access efficiency based on time series can be implemented in the form of a computer program, which can be used in Fig.11 Runs on the computer device shown.

[0116] See also Fig.11 , Fig.11 11 is a schematic block diagram of a computer device provided by an embodiment of the present invention. The computer device 1100 is a server, which may be an independent server or a server cluster composed of multiple servers.

[0117] See also Fig.11 The computer device 1100 includes a processor 1102 , an access device, and a network interface 1105 connected via a system bus 1101 , wherein the access device may include a non-volatile access medium 1103 and a memory access device 1104 .

[0118] The non-volatile access medium 1103 can store an operating system 11031 and a computer program 11032. When the computer program 11032 is executed, the processor 1102 can execute a method for optimizing the access efficiency of a stacker based on a time series.

[0119] The processor 1102 is used to provide computing and control capabilities to support the operation of the entire computer device 1100 .

[0120] The memory access device 1104 provides an environment for the operation of the computer program 11032 in the non-volatile access medium 1103. When the computer program 11032 is executed by the processor 1102, the processor 1102 can execute a method for optimizing the access efficiency of the stacker based on time series.

[0121] The network interface 1105 is used for network communication, such as providing data information transmission, etc. Those skilled in the art will understand that Fig.11The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device 1100 to which the solution of the present invention is applied. The specific computer device 1100 may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0122] Those skilled in the art will understand that Fig.11 The embodiment of the computer device shown in the figure does not constitute a limitation on the specific structure of the computer device. In other embodiments, the computer device may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently. For example, in some embodiments, the computer device may only include a memory and a processor. In such an embodiment, the structure and function of the memory and processor are the same as those of the embodiment. Fig.11 The embodiments shown are consistent and will not be described again here.

[0123] It should be understood that in the embodiment of the present invention, the processor 1102 may be a central processing unit (Central Processingμnit, CPμ), and the processor 1102 may also be other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application-specific integrated circuits (Application Specific IntegratedCircμit, ASIC), field-programmable gate arrays (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0124] In another embodiment of the present invention, a computer-readable access medium is provided. The computer-readable access medium may be a non-volatile computer-readable access medium. The computer-readable access medium stores a computer program, wherein when the computer program is executed by a processor, the method for optimizing the access efficiency of a stacker based on a time series according to an embodiment of the present invention is implemented.

[0125] The access medium is a physical, non-transient access medium, for example, a μ disk, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disk, etc., which can store and access program codes.

[0126] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described equipment, devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0127] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for optimizing the storage and retrieval efficiency of a stacker based on time series, characterized in that: include: According to the position of the inbound and outbound platform connected to the stacker in each cargo lane, a corresponding inbound and outbound buffer zone is set in each cargo lane; Obtain the historical in-and-out data of each product specification corresponding to the preset time period, and input it into the in-and-out prediction model corresponding to the preset time period for prediction, and output the future in-and-out data of each product specification corresponding to the preset time period; Generate a warehouse transfer task based on the future in-and-out warehouse data corresponding to each product specification in a preset time period, and control the stacker to perform in-and-out warehouse transfer operations between the corresponding cargo lanes, in-warehouse buffers and out-warehouse buffers according to the warehouse transfer task.

2. The method for optimizing stacker access efficiency based on time series according to claim 1, characterized in that: The historical in-and-out data of each product specification corresponding to a preset time period is obtained, and input into the in-and-out prediction model corresponding to the preset time period for prediction, and the future in-and-out data of each product specification corresponding to the preset time period is output, including: Obtain the historical in-and-out data of each product specification in the long-term cycle; Confirm the business characteristics associated with each product specification; According to the business characteristics associated with each product specification, the corresponding historical inbound and outbound data are input into the Prophet model or SARIMA model for prediction, and the future inbound and outbound data corresponding to each product specification in the long-term cycle are output.

3. The method for optimizing stacker access efficiency based on time series according to claim 1, characterized in that: The historical in-and-out data of each product specification corresponding to a preset time period is obtained, and input into the in-and-out prediction model corresponding to the preset time period for prediction, and the future in-and-out data of each product specification corresponding to the preset time period is output, including: Obtain the historical in-and-out data of each product specification in the mid-term cycle; Confirm the business characteristics associated with each product specification; According to the business characteristics associated with each product specification, the corresponding historical inbound and outbound data are input into the moving average model, LSTM model or XGBoost model for prediction, and the future inbound and outbound data corresponding to each product specification in the medium-term cycle are output.

4. The method for optimizing stacker access efficiency based on time series according to claim 1, characterized in that: The historical in-and-out data of each product specification corresponding to a preset time period is obtained, and input into the in-and-out prediction model corresponding to the preset time period for prediction, and the future in-and-out data of each product specification corresponding to the preset time period is output, including: Obtain the historical in-and-out data of each product specification in the short-term cycle; Confirm the business characteristics associated with each product specification; According to the business characteristics associated with each product specification, the corresponding historical inbound and outbound data are input into the LightGBM model or Kalman filter model for prediction, and the future inbound and outbound data corresponding to each product specification in the short term cycle are output.

5. The method for optimizing stacker access efficiency based on time series according to any one of claims 1 to 4, characterized in that: The method generates a warehouse transfer task according to the future warehouse in and out data corresponding to each product specification in a preset time period, and controls the stacker to perform warehouse in and out transfer operations between the corresponding cargo lanes, warehouse inlet buffers and warehouse outlet buffers according to the warehouse transfer task, including: According to the future in-and-out data of each product specification in the preset time period, estimate the in-and-out quantity of each product specification in the next period, and generate the transfer task of the corresponding product specification; Determine the outbound product specifications and quantity in the warehouse transfer task, and control the stacker to move the corresponding product specifications in the area away from the outbound platform or close to the inbound platform on the cargo lane to the outbound buffer zone during the electric wave valley time period; Determine the total quantity of incoming goods in the warehouse transfer task, and control the stacker to move the corresponding quantity and specifications in the incoming goods buffer to the area close to the incoming goods buffer on the cargo lane during the electric wave valley time period.

6. The method for optimizing stacker access efficiency based on time series according to claim 1, characterized in that: Also includes: Confirm the future inbound and outbound data of each product specification corresponding to the preset time period, wherein the preset time period includes a long-term period, a medium-term period and a short-term period; According to the order of the long-term cycle, the medium-term cycle and the short-term cycle, the prediction results of the future in-and-out warehouse data of each product specification are processed with top-down transfer constraints to adjust the prediction results of each product specification at different cycle scales.

7. The method for optimizing stacker access efficiency based on time series according to claim 1, characterized in that: Also includes: Based on the current business constraints of the warehouse, the future inbound and outbound data of each product specification corresponding to the preset time period is horizontally complemented in the time dimension and the business dimension to optimize the prediction results of the future inbound and outbound data of each product specification corresponding to the preset time period.

8. The method for optimizing stacker access efficiency based on time series according to claim 1, characterized in that: Also includes: Real-time monitoring of the forecast error of future inbound and outbound data of each product specification corresponding to a preset time period, wherein the preset time period includes a long-term period, a medium-term period, and a short-term period; Real-time monitoring of warehouse task completion rate, equipment utilization rate and business indicator data of equipment utilization rate; Real-time monitoring of the forecast status of the inbound and outbound forecasting model corresponding to each preset time period; Based on the prediction error, business indicator data and prediction status, each inbound and outbound prediction model is updated with bottom-up feedback.

9. The method for optimizing stacker access efficiency based on time series according to claim 5, characterized in that: Also includes: Confirm the in-and-out quantity of each product specification in the next period; Confirm the current inventory status in the warehouse; During the off-peak period, move the products with high outbound quantity in the next period to an area close to the outbound platform and easy to get; During the radio wave valley period, move the products with high storage quantity in the next period to the area close to the storage platform; During the radio wave valley period, move the products in the warehouse that have no short-term demand for shipment to an area far away from the shipment platform and not easily accessible.

10. The method for optimizing stacker access efficiency based on time series according to claim 1, characterized in that: Also includes: Regularly calculate the accuracy score, stability score and business relevance score of each inbound and outbound forecast model corresponding to each preset time period, and obtain the comprehensive score of each inbound and outbound forecast model after weighted calculation; For multiple inbound and outbound forecasting models corresponding to each preset time period, select the inbound and outbound forecasting model with the highest comprehensive score and stable performance as the current forecasting model corresponding to each preset time period, and select other inbound and outbound forecasting models as candidate models corresponding to each preset time period; The prediction error of the current prediction model corresponding to each preset time period is counted, and when the prediction error exceeds the preset threshold, the model is triggered to retrain or replace the candidate model with the second highest comprehensive score.

11. A device for optimizing the storage and retrieval efficiency of a stacker based on time series, characterized in that: include: An area setting unit is used to set a corresponding inbound and outbound buffer zone in each cargo lane according to the position of the inbound and outbound platform connected to the stacker in each cargo lane; The in-and-out inventory prediction unit is used to obtain the historical in-and-out inventory data corresponding to the preset time period of each product specification, and input it into the in-and-out inventory prediction model corresponding to the preset time period for prediction, and output the future in-and-out inventory data corresponding to the preset time period of each product specification; The in-and-out transfer unit is used to generate transfer tasks according to the future in-and-out data corresponding to each product specification in a preset time period, and control the stacker to perform in-and-out transfer operations between the corresponding storage location aisles, inbound buffers and outbound buffers according to the transfer tasks.

12. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method for optimizing the access efficiency of a stacker based on time series according to any one of claims 1 to 10 is implemented.

13. A computer-readable storage medium, characterized in that: The computer readable medium stores a computer program, which, when executed by a processor, causes the processor to execute the method for optimizing the access efficiency of a stacker based on time series according to any one of claims 1 to 10.

Images