Charging scheduling method and system based on multi-model fusion, electronic device, and storage medium

Abstract

This application relates to the field of warehousing equipment technology and discloses a charging scheduling method, system, electronic device and storage medium based on multi-model fusion. By fusing the prediction results of LSTM base model, Transformer base model and TCN base model, it can comprehensively capture long-range dependencies, short-term memory and local patterns in the task sequence. Thus, based on the fused prediction results and combined with the current online equipment, it can dynamically optimize the charging strategy and improve the robustness of charging scheduling.

Description

Technical Field

[0001] This application relates to the field of warehousing equipment technology, and in particular to a charging scheduling method, system, electronic device and storage medium based on multi-model fusion. Background Technology

[0002] In modern warehousing and logistics centers (such as robotic warehouses and automated storage and retrieval systems), operational efficiency heavily relies on the stable operation of automated equipment, such as Automated Guided Vehicles (AGVs), Autonomous Mobile Robots (AMRs), and stacker cranes. These devices are typically battery-powered and require periodic charging.

[0003] Traditional scheduling systems typically employ static, passive charging strategies, such as charging devices when their battery level falls below a fixed threshold; or, tiered charging, such as designing different battery thresholds based on the average battery level of all operating devices, allowing devices to be charged during off-peak hours; or, prediction and scheduling based on a single learning model, such as collecting historical task volume data, inputting holiday / weekday features, outputting predicted task volume for future hours, and then scheduling charging accordingly.

[0004] However, the scheduling decisions of these schemes are usually based on simple judgments of the prediction results of a single model. The decision risk is concentrated and the scheduling error is easily caused by the accidental failure of a single model, resulting in insufficient robustness of charging scheduling. Summary of the Invention

[0005] This application provides a charging scheduling method, system, electronic device, and storage medium based on multi-model fusion, which improves the robustness of charging scheduling.

[0006] The embodiments of this application provide the following technical solutions: On the one hand, embodiments of this application provide a charging scheduling method based on multi-model fusion, the method comprising: Construct a time series dataset, which includes the historical task volume corresponding to each historical time period; The time series dataset is simultaneously input into the LSTM base model, the Transformer base model, and the TCN base model to obtain the first prediction result output by the LSTM base model, the second prediction result output by the Transformer base model, and the third prediction result output by the TCN base model. The first, second, and third prediction results are fused to obtain the fused prediction result. Based on the fusion prediction results, calculate the number of devices required for each future time period; Charging schedules are implemented based on the number of devices required in different time periods and the current number of online devices.

[0007] In some embodiments, The first, second, and third prediction results are fused to obtain the fused prediction result, which includes: Based on the first prediction result, the second prediction result, and the third prediction result, a first feature vector is generated. Obtain external features for various future time periods; The external features are concatenated with the first feature vector to obtain the concatenated feature vector. The concatenated feature vectors are input into the XGBoost model to obtain the fused prediction results.

[0008] In some embodiments, The concatenated feature vectors are input into the XGBoost model to obtain the fused prediction results, including: The first prediction result, the second prediction result and the third prediction result are weighted to determine the first weight corresponding to the first prediction result, the second weight corresponding to the second prediction result and the third weight corresponding to the third prediction result. Based on the prediction features of the first, second, and third prediction results, combined with external features, the XGBoost model is used to adjust the first weight, and / or the second weight, and / or the third weight to obtain the fused prediction result.

[0009] In some embodiments, The first prediction result includes the first task volume for each future time period, the second prediction result includes the second task volume for each future time period, and the third prediction result includes the third task volume for each future time period. The predictive features include the quantitative relationships between the first task quantity, the second task quantity, and the third task quantity; External characteristics include one or more of the following: time period characteristics, weekday characteristics, holiday characteristics, and promotional characteristics.

[0010] In some embodiments, Based on the fusion prediction results, calculate the number of devices required for each future time period, including: Based on the fusion prediction results and combined with the efficiency coefficient, the number of devices required in each future time period is calculated. The number of devices required in each future time period = fusion prediction result / efficiency coefficient, where the efficiency coefficient represents the amount of tasks that one device can handle per unit time.

[0011] In some embodiments, The method also includes: Obtain the total number of tasks completed within multiple preset historical time periods; The efficiency coefficient is calculated based on the total amount of tasks completed within multiple preset historical time periods. The efficiency coefficient is calculated as the total number of tasks completed within a set number of historical time periods divided by the total number of equipment operating hours. The total number of equipment operating hours is the sum of the operating hours of all equipment involved in the operation.

[0012] In some embodiments, Based on the number of devices required in different time periods in the future, and in conjunction with the current online devices, charging scheduling is carried out, including: Obtain the current status information of online devices, including battery level information; Based on power consumption information, predict the number of devices available in different time periods in the future; If the number of devices required in each future time period is less than or equal to the number of available devices, then the charging strategy decided in the previous time period will be maintained. If the number of devices required in future time periods exceeds the number of available devices, the charging strategy decided in the previous time period will be updated, including: generating active charging scheduling instructions to schedule the device control system to execute the active charging scheduling instructions.

[0013] In some embodiments, Active charging scheduling commands are used for: Mark future periods of high demand; and, Before the peak demand period begins, the first device to enter the charging queue must not be allowed to charge; and, Prioritize charging the second power source during the current idle period; Among them, the power of the first power device is greater than the first power threshold, the power of the second power device is less than the second power threshold, and the first power threshold is greater than the second power threshold.

[0014] On the other hand, embodiments of this application provide a charging scheduling system based on multi-model fusion, including: The data preprocessing module is used to construct a time series dataset, which includes the historical task volume corresponding to each historical time period. The multi-model parallel prediction module is used to simultaneously input time series datasets into the LSTM base model, the Transformer base model, and the TCN base model to obtain the first prediction result output by the LSTM base model, the second prediction result output by the Transformer base model, and the third prediction result output by the TCN base model. The integration and meta-learning module is used to fuse the first, second, and third prediction results to obtain the fused prediction result; based on the fused prediction result, the number of devices required in each future time period is calculated. The charging scheduling decision module is used to schedule charging based on the number of devices required in various future time periods, combined with the current online devices.

[0015] On the other hand, embodiments of this application provide an electronic device, including: At least one processor; At least one memory for storing at least one program; When at least one program is executed by at least one processor, the at least one processor performs the method described above.

[0016] On the other hand, embodiments of this application provide a non-volatile computer-readable storage medium storing a processor-executable program, characterized in that the processor-executable program is used to perform the above-described method when executed by a processor.

[0017] Unlike existing technologies, this application provides a charging scheduling method based on multi-model fusion. The method includes: constructing a time-series dataset, wherein the time-series dataset includes historical task volumes corresponding to each historical time period; simultaneously inputting the time-series dataset into an LSTM base model, a Transformer base model, and a TCN base model to obtain a first prediction result output by the LSTM base model, a second prediction result output by the Transformer base model, and a third prediction result output by the TCN base model; fusing the first prediction result, the second prediction result, and the third prediction result to obtain a fused prediction result; calculating the number of devices required for each future time period based on the fused prediction result; and performing charging scheduling based on the number of devices required for each future time period, combined with the current online devices.

[0018] By fusing the prediction results of LSTM, Transformer, and TCN base models, the system comprehensively captures long-range dependencies, short-term memory, and local patterns in the task sequence. Based on the fused prediction results and combined with the current online equipment, the system can dynamically optimize the charging strategy and improve the robustness of charging scheduling. Attached Figure Description

[0019] One or more embodiments are illustrated by way of example with reference to the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements having the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0020] Figure 1This is a schematic diagram of a warehousing system provided in an embodiment of this application; Figure 2 This is a flowchart illustrating a charging scheduling method based on multi-model fusion provided in an embodiment of this application; Figure 3 yes Figure 2 A detailed flowchart of step S203 in the process; Figure 4 yes Figure 3 A detailed flowchart of step S234 in the process; Figure 5 yes Figure 2 A detailed flowchart of step S205 in the process; Figure 6 This is a schematic diagram of the overall process of a charging scheduling method based on multi-model fusion provided in an embodiment of this application; Figure 7 This is a schematic diagram of the structure of a charging scheduling system based on multi-model fusion provided in an embodiment of this application; Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0021] Explanation of icon numbers: Detailed Implementation

[0022] To facilitate understanding of this application, a more detailed description is provided below with reference to the accompanying drawings and specific embodiments. It should be noted that when an element is described as "fixed to" another element, it can be directly on the other element, or one or more intermediate elements may exist between them. When an element is described as "connected to" another element, it can be directly connected to the other element, or one or more intermediate elements may exist between them. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this specification are for illustrative purposes only.

[0023] Unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The term "and / or" as used in this specification includes any and all combinations of one or more of the associated listed items.

[0024] Before providing a detailed description of this application, the nouns and terms used in the embodiments of this application are explained, and the nouns and terms used in the embodiments of this application shall be interpreted as follows: (1) LSTM base model refers to a neural network model with Long Short-Term Memory Network (LSTM) as its core and basic architecture. Its structure contains multiple LSTM layers. LSTM can effectively learn the short-term memory and temporal dependency patterns of task sequences through its input gate, forget gate and output gate mechanism, such as "a midday trough is usually followed by a morning peak".

[0025] (2) The Transformer base model refers to a neural network model built on the core and foundation of the Transformer architecture, with the self-attention layer at its core. This model treats each time point in the time series as a vector and calculates the pairwise correlation weights between all time points through the self-attention mechanism. This allows it to directly capture long-range dependencies, such as "the peak in the first week of this month is similar to the peak pattern four weeks ago".

[0026] (3) The TCN base model refers to a neural network model with Temporal Convolutional Network (TCN) as its core and basic architecture. It uses causal dilated convolutional layers to ensure that the prediction does not use future information. By stacking multiple convolutions and gradually increasing the dilation coefficient, TCN can obtain a wide temporal receptive field with low computational cost. It is particularly good at capturing local, multi-scale patterns (such as sudden peaks lasting 2 hours) and is robust to input noise.

[0027] (4) XGBoost model refers to a high-performance, scalable machine learning algorithm based on the gradient boosting framework. It is a gradient boosting decision tree model. Its principle is to train a new decision tree in each round of iteration to correct the residuals (errors) of the overall predictions of all trees in the previous round. Finally, the prediction results of all trees are weighted and summed to obtain a powerful ensemble model. XGBoost model has a strong fitting ability for tabular structured data and can efficiently learn under what circumstances different base models are more reliable. (5) Stacking ensemble is a machine learning technique that improves the performance of a model by combining multiple different base learners. In Stacking ensemble, multiple different base learners are first trained, and their prediction results are then used as input to train a meta-learner to combine the prediction results of these base learners to obtain the final prediction result. Stacking ensemble can usually improve the generalization ability and prediction accuracy of the model.

[0028] Currently, most scheduling systems adopt passive charging strategies based on fixed thresholds (such as charging when the battery level is below 20%); or static scheduling schemes based on simple rules (such as charging before shift change); or tiered charging, for example, designing different battery thresholds based on the average battery level of all operating equipment, and charging equipment during off-peak hours; or prediction and scheduling based on a single learning model, for example, collecting historical task volume data, inputting holiday / weekday features, outputting predicted task volume for future hours, and performing charging scheduling.

[0029] However, the above solution has the following problems: (1) Insufficient utilization of environmental feature information: Existing scheduling decisions focus on historical task volume data itself, and fail to fully explore and utilize the value of other structured features that are strongly related to task volume (such as the number of charging times in the same period in history, the average equipment failure interval, occasional rest periods, and low task volume).

[0030] (2) Single basis for scheduling decisions: Existing scheduling decisions are mostly based on the prediction results of a single model to make simple and rigid rule judgments. They fail to comprehensively consider the prediction evidence provided by multiple models from different perspectives, resulting in concentrated decision risks and easy scheduling errors due to the accidental failure of a single model.

[0031] Based on this, this application provides a charging scheduling method based on multi-model fusion. By fusing the prediction results of LSTM base model, Transformer base model, and TCN base model, it can comprehensively capture long-range dependencies, short-term memory, and local patterns in the task sequence. Thus, based on the fused prediction results and combined with the current online equipment, it can dynamically optimize the charging strategy and improve the robustness of charging scheduling.

[0032] The technical solution of this application is described in detail below with reference to the accompanying drawings: Please see Figure 1 , Figure 1 This is a schematic diagram of a warehousing system provided in an embodiment of this application.

[0033] Understandably, this warehousing system is used for the storage, picking, and inbound / outbound processing of goods, and it is applied to warehouses.

[0034] like Figure 1 As shown, the warehousing system includes shelves, robots, and charging stations. The shelves are used to store toy bins, and the robots are used to move the toy bins. A warehousing system may include one or more charging stations. Figure 1 Taking a charging station as an example, the charging station is equipped with multiple charging piles for the robot to charge.

[0035] Understandably, this warehousing system also includes workstations where robots move bins from the shelves to the workstations. Workers or robotic arms then sort the goods needed for an order from the bins and store them in the corresponding containers. Robots facilitate the movement of goods between the shelves and the workstations.

[0036] Example 1 Please see Figure 2 , Figure 2 This is a flowchart illustrating a charging scheduling method based on multi-model fusion provided in an embodiment of this application. Specifically, this charging scheduling method is applied to an electronic device, and the executing entity of the method is one or at least two processors of the electronic device.

[0037] like Figure 2 As shown, the charging scheduling method based on multi-model fusion includes the following steps S201-S205: Step S201: Construct a time series dataset, which includes the historical task volume corresponding to each historical time period.

[0038] Specifically, historical data is obtained from the charging scheduling system. This historical data includes the historical task volume corresponding to each historical time period, such as the historical task volume per hour / half hour.

[0039] This historical data also includes external features, such as time period features, weekday features, holiday features, and promotional features.

[0040] The historical data also includes engineering features, such as the number of charging requests during the same period in history, average device utilization, and known major event markers.

[0041] After acquiring historical data, it is preprocessed, such as by cleaning, denoising, aligning, and normalizing the historical data, to form a time series dataset with multiple variables for model training. This time series dataset includes historical task volume, external features, engineering features, etc., corresponding to each historical time period.

[0042] For example, historical data can be cleaned and denoised, including handling missing and outlier values. For example, for missing values, if data from a certain point in time is missing due to system data storage issues, it can be filled using data from points before and after that time. For outliers, such as a sudden extreme value in the task volume (querying upstream business volume), data that contradicts the business context will be removed.

[0043] For example, aligning historical data includes aligning data from different data sources in terms of storage dimensions. For instance, if the task volume data is an hourly value, while the "average equipment utilization" in the charging scheduling system data is a point every 5 minutes, then the equipment utilization is aggregated into an hourly average and aligned with the timestamp of the task volume data.

[0044] For example, normalizing historical data includes converting features at different scales to the same scale, such as converting weekdays into weekday codes of [1, 7] and time periods into hour codes of [0, 24].

[0045] In this application embodiment, by preprocessing historical data to generate a time series dataset, this application can use the preprocessed historical data for prediction, thereby improving the accuracy of the prediction.

[0046] Step S202: Input the time series dataset into the LSTM base model, the Transformer base model, and the TCN base model simultaneously to obtain the first prediction result output by the LSTM base model, the second prediction result output by the Transformer base model, and the third prediction result output by the TCN base model.

[0047] Specifically, the LSTM, Transformer, and TCN base models are parallel models. By simultaneously inputting the time series dataset into the parallel-running LSTM, Transformer, and TCN base models, they can perform predictions in parallel, thus obtaining the first prediction result output by the LSTM base model, the second prediction result output by the Transformer base model, and the third prediction result output by the TCN base model. The first, second, and third prediction results are all task-related prediction values.

[0048] It is understandable that the LSTM, Transformer, and TCN base models operate independently. The original input data for the three base models is exactly the same during training and inference to ensure that the three base models learn under the same task and the same set of information. This allows for fair comparison and integration of the prediction results of the three base models. The three base models output prediction sequences for the same time period in the future, that is, prediction values ​​of the task volume for the same time span in the future. For example, the first prediction result is denoted as Pred_l (LSTM base model output), the second prediction result is denoted as Pred_t (Transformer base model output), and the third prediction result is denoted as Pred_c (TCN base model output).

[0049] In this embodiment of the application, by combining the advantages of the three major models, Transformer, LSTM and TCN, it is possible to comprehensively capture the complex patterns (long-term, short-term and local) of task quantity sequences from different scales and time spans, thus avoiding the structural bias of a single model.

[0050] Step S203: The first prediction result, the second prediction result, and the third prediction result are fused to obtain the fused prediction result.

[0051] Please see Figure 3 , Figure 3 yes Figure 2 A detailed flowchart of step S203 is shown.

[0052] like Figure 3 As shown, step S203 involves fusing the first prediction result, the second prediction result, and the third prediction result to obtain a fused prediction result, including the following steps S231-S234: Step S231: Generate the first feature vector based on the first prediction result, the second prediction result, and the third prediction result.

[0053] Specifically, the prediction results Pred_l, Pred_t, and Pred_c from the three base models are combined as the first feature vector, for example, the first feature vector is [Pred_l, Pred_t, Pred_c].

[0054] Step S232: Obtain the external features for each future time period.

[0055] Specifically, external features include time period features, weekday features, holiday features, promotional features, etc., and may include one or more of the above features. For example, the weekday feature is "tomorrow is Wednesday", and the holiday feature is "whether it is a holiday in 2 hours".

[0056] Step S233: Concatenate the external features with the first feature vector to obtain the concatenated feature vector.

[0057] Specifically, the external feature is concatenated with the first feature vector to obtain the concatenated feature vector. For example, the concatenated feature vector is [first feature vector, external feature], such as [Pred_l, Pred_t, Pred_c, "The next 2 hours is a holiday"].

[0058] Step S234: Input the concatenated feature vector into the XGBoost model to obtain the fused prediction result.

[0059] Specifically, the concatenated feature vectors are used as the input to the second-layer meta-learner, namely the XGBoost model, to obtain the fused prediction result through the XGBoost model.

[0060] Understandably, the XGBoost model is a gradient boosting decision tree model. XGBoost has a strong ability to fit tabular structured data and can efficiently learn when different base models are more reliable. For example, the XGBoost model can learn: "When Pred_t and Pred_c are close but Pred_l deviates significantly, the consensus of Pred_t and Pred_c should be trusted more," or "In predictions involving holidays, the weight of Pred_t should be increased." The XGBoost model integrates multiple weak decision trees, weighting and correcting the predictions output by these three base models to output a final, more accurate fused prediction result, Final_Prediction.

[0061] Please refer to the following: Figure 4 , Figure 4 yes Figure 3 A detailed flowchart of step S234 in the process.

[0062] like Figure 4 As shown, step S234 involves inputting the concatenated feature vector into the XGBoost model to obtain the fused prediction result, including the following steps S2341-S2342: Step S2341: Weight the first prediction result, the second prediction result and the third prediction result to determine the first weight corresponding to the first prediction result, the second weight corresponding to the second prediction result and the third weight corresponding to the third prediction result; Specifically, initialize the first weight, second weight, and third weight. For example, set the first weight, second weight, and third weight to be the same, all of which are 1 / 3.

[0063] Step S2342: Based on the prediction features of the first prediction result, the second prediction result, and the third prediction result, and combined with external features, adjust the first weight and / or adjust the second weight and / or adjust the third weight through the XGBoost model to obtain the fused prediction result.

[0064] Specifically, the first prediction result includes the first task volume for each future time period, the second prediction result includes the second task volume for each future time period, and the third prediction result includes the third task volume for each future time period. The prediction features include the quantitative relationships between the first task quantity, the second task quantity, and the third task quantity. For example, the prediction feature is "the values ​​of Pred_l, Pred_t, and Pred_c are close"; or, "Pred_t and Pred_c are close, but Pred_l deviates significantly"; or, "Pred_t predicts a very high peak value, while Pred_l and Pred_c predict lower values".

[0065] External features include time period features, weekday features, holiday features, and promotional features. For example, external features could be "is_holiday=0, day_of_week=1 (Monday)" or "no special events".

[0066] By integrating multiple weak decision trees through the XGBoost model, the prediction results output by the three base models are weighted and corrected to adjust the first weight, and / or the second weight, and / or the third weight, to output the final, more accurate fusion prediction result Final_Prediction.

[0067] Table 1 below is an enumeration table of decision scenarios for an XGBoost model provided in an embodiment of this application.

[0068]

[0069] Table 1 In this embodiment, stacking integration using the XGBoost model can intelligently weigh the prediction results of each base model and effectively correct the residuals, thereby obtaining more stable and accurate prediction results in various complex business scenarios.

[0070] Furthermore, through a fusion architecture with an inherent fault-tolerance mechanism, even if a base model (such as TCN) fails to predict properly for some reason during a specific time period, the other two models and the XGBoost meta-learner can still make a relatively reliable final judgment based on the correct opinions of other "experts," effectively preventing the collapse of the entire system's scheduling decision due to the failure of a single model.

[0071] Step S204: Calculate the number of devices required for each future time period based on the fusion prediction results.

[0072] Specifically, based on the fusion prediction results, the number of devices required in various future time periods is calculated, including: Based on the fusion prediction results and combined with the efficiency coefficient, the number of devices required in each future time period is calculated. The number of devices required in each future time period = fusion prediction result / efficiency coefficient, where the efficiency coefficient represents the amount of tasks that one device can handle per unit time.

[0073] It is understandable that the number of devices required in each future time period is the minimum number of devices required in each future time period, denoted as N_T. Then, N_T = Final_Prediction(T) / efficiency coefficient. It should be noted that the efficiency coefficient represents the amount of work that one device can handle per unit of time.

[0074] In the embodiments of this application, the equipment includes Automated Guided Vehicles (AGVs), Autonomous Mobile Robots (AMRs), stacker cranes, etc., wherein the equipment in the embodiments of this application is an Automated Guided Vehicle.

[0075] It should be noted that the efficiency factor (EF) is defined as the amount of work that a single device can handle per unit of time. The efficiency factor comprehensively reflects multiple factors such as device speed, operator proficiency, and process smoothness.

[0076] In this embodiment of the application, the formula for calculating the efficiency coefficient is: Efficiency coefficient EF = Total completed tasks / (Total equipment operating hours) Specifically, the calculation of the efficiency system includes the following steps (1)-(3): Step (1): Obtain historical time.

[0077] Specifically, choose a representative historical period, such as the past 3 months. It's understood that this historical period includes multiple historical time intervals, for example, each historical time interval being 1 hour.

[0078] Step (2): Filter multiple preset historical time periods and obtain the total number of tasks completed within the multiple preset historical time periods.

[0079] Specifically, data is filtered across multiple historical time periods. For example, only periods when the equipment operates at full capacity are selected, such as excluding idle periods in the early morning and analyzing only data from peak working hours (e.g., 9-12 AM and 2-5 PM) to calculate the maximum efficiency of the equipment under normal conditions. Based on the selected historical time periods, the total workload completed by multiple devices within those periods is calculated.

[0080] Step (3): Calculate the efficiency coefficient based on the total amount of tasks completed within multiple preset historical time periods.

[0081] Specifically, the total number of tasks completed includes the total number of tasks completed (e.g., the number of boxes moved) recorded in the warehouse WMS system within the selected historical time period. The total equipment operating hours include the sum of the operating hours of all equipment (AGV / AMR) involved in the operation within the same time period. (For example, if 10 devices run for 3 hours, the total operating hours are 30).

[0082] Therefore: Efficiency coefficient EF = Total completed tasks / Total equipment operating hours For example: During peak weekday hours over the past three months, a total of 600,000 tasks were completed.

[0083] During the same period, the total operating time of all equipment was 15,000 equipment-hours.

[0084] Therefore: Efficiency coefficient EF = 600,000 / 15,000 = 40.

[0085] Step S205: Based on the number of devices required in each future time period and in combination with the current online devices, perform charging scheduling.

[0086] Please refer to the following: Figure 5 , Figure 5 yes Figure 2 A detailed flowchart of step S205 in the process.

[0087] like Figure 5 As shown, step S205 involves scheduling charging based on the number of devices required in future time periods and the current number of online devices, including the following steps S251-S255: Step S251: Obtain the current status information of the online device, including power information.

[0088] It is understandable that the current online devices refer to the devices used to perform tasks. Each device corresponds to a status information, including battery information, for example: the current battery level of a certain device is 60%.

[0089] Step S252: Based on the power information, predict the number of devices available in each future time period.

[0090] Specifically, based on the current power information of each device and its power consumption rate, the power of each device in future time periods is calculated. If the power is less than or equal to a preset power threshold, the device is determined to be unusable.

[0091] For example: Assuming the preset battery threshold is 20%, the current time is 7 AM, a device's current battery is 60%, and the battery consumption rate is 10% per hour, then the device's battery information for various future time periods is as follows: 8:00 (1 hour): 50%; 9:00 (2 hours): 40%; 10:00 (3 hours): 30% 11:00 (4 hours): 20% It can be seen that if the device's battery level is greater than 20% before 11:00, then the device is considered usable; if the device's battery level is less than or equal to 20% after 11:00, then the device is considered unusable.

[0092] It is understandable that the power consumption rate of each device may be the same or different. Therefore, by calculating the power consumption of each device, it is possible to determine whether the device is available in different time periods in the future, thereby determining the number of devices available in the warehousing system in different time periods in the future, and thus enabling better charging scheduling.

[0093] Step S253: Whether the number of devices required in each future time period is greater than the number of available devices.

[0094] Specifically, it involves determining whether the number of devices required by the warehousing system in various future time periods exceeds the number of available devices.

[0095] Step S254: Update the charging strategy decided in the previous moment, including: generating an active charging scheduling instruction to schedule the equipment control system to execute the active charging scheduling instruction.

[0096] Specifically, if the number of devices required in future time periods exceeds the number of available devices, the charging strategy decided in the previous time period will be updated, including: generating active charging scheduling instructions to schedule the device control system to execute the active charging scheduling instructions.

[0097] In this embodiment of the application, the active charging scheduling command is used for: Mark future periods of high demand; and, Before the peak demand period begins, the first device to enter the charging queue must not be allowed to charge; and, Prioritize charging the second power source during the current idle period; Among them, the power of the first power device is greater than the first power threshold, the power of the second power device is less than the second power threshold, and the first power threshold is greater than the second power threshold.

[0098] Specifically, a high-demand period refers to a time period in which the number of devices required exceeds a preset device number threshold. For example, if the preset device number threshold is 50, and the number of devices required in a certain time period is greater than 50, then that time period is defined as a high-demand period.

[0099] Step S255: Maintain the charging strategy decided in the previous moment.

[0100] Specifically, if the number of devices required in each future time period is less than or equal to the number of available devices, the charging strategy decided in the previous time period will be maintained.

[0101] In this embodiment, the charging strategy includes: retaining the currently needed number of devices and controlling other devices besides the currently needed devices to be charged through the charging station. For example, the charging strategy decided at the previous moment includes: retaining the needed N devices and charging the remaining devices through the charging station, where N is a positive integer.

[0102] In the embodiments of this application, the charging scheduling system can more accurately predict future capacity gaps through more accurate and robust prediction results, thereby achieving more refined and forward-looking charging management. This can effectively avoid capacity shortages during peak task periods, reduce order backlogs, and also reasonably arrange charging during idle periods, improving the overall utilization rate of equipment and energy efficiency, ultimately achieving cost reduction and efficiency improvement.

[0103] Please see Figure 6 , Figure 6 This is a schematic diagram of the overall process of a charging scheduling method based on multi-model fusion provided in an embodiment of this application.

[0104] like Figure 6 As shown, the overall process of this charging scheduling method based on multi-model fusion includes the following steps S601-S606: Step S601: Obtain historical data.

[0105] Step S602: Construct a time series dataset.

[0106] Step S603: Perform multi-model prediction based on the time series dataset.

[0107] Step S604: Merge multiple prediction results to obtain a fused prediction result.

[0108] Step S605: Calculate the number of devices required for each future time period based on the fusion prediction results.

[0109] Step S606: Based on the number of devices required in each future time period and in combination with the current online devices, perform charging scheduling.

[0110] The following example illustrates the overall process of the charging scheduling method based on multi-model fusion in this application.

[0111] Background setting: Current time: 23:00 on November 9th Forecast objective: Forecast the total task volume per hour for the next 24 hours (11-10 00:00 to 11-10 23:00).

[0112] Efficiency coefficient: Based on historical data analysis, we determined that a single device can process an average of 40 tasks per hour. (EF = 40) Current equipment status: The warehouse has a total of 150 AGVs. At this moment, 120 AGVs are online, of which 30 have less than 30% battery power; 30 AGVs are offline charging.

[0113] Step 1: Data Preparation and Base Model Input The system extracts historical data from the database for the past 7 days (from 00:00 on November 2 to 23:00 on November 9, a total of 192 hours) to construct input features.

[0114] Table 2 below is a schematic table of historical data provided in an embodiment of this application.

[0115]

[0116] Table 2 Known features for the next 24 hours (for XGBoost): The system knows that November 10th is a Friday and a pre-sale promotion day.

[0117] For example, for the future time point 11-10 14:00, its characteristics are: hour_of_day=14, day_of_week=4 (Friday), is_holiday=0, is_promotion=1.

[0118] Step 2: Parallel prediction using the three fundamental models Input the matrix [168, 5] above into the three pre-trained base models simultaneously.

[0119] Table 3 below is a schematic table of the outputs (predictions for the next 24 hours) of the three base models provided in the embodiments of this application.

[0120]

[0121] Table 3 Step 3: XGBoost Integrated Meta-Learning Now, XGBoost will make the final decision based on the opinions of these three "experts." Let's take 10:00 and 20:00 as examples.

[0122] XGBoost input feature vector construction: For 11-10 10:00: Baseline model predictions: [3800, 4500, 3700] Time characteristics: [10, 4, 0, 1] (hour=10, Friday, non-holiday, promotion) Complete feature vector: [3800, 4500, 3700, 10, 4, 0, 1] For 11-10 20:00: Baseline model predictions: [4200, 5200, 4100] Time characteristics: [20, 4, 0, 1] (hours = 20, Friday, non-holiday, promotional) Complete feature vector: [4200, 5200, 4100, 20, 4, 0, 1] Decision logic of the XGBoost model (based on a trained model): The model has learned that when is_promotion=1, the prediction accuracy of Transformer (Pred_t) is much higher than that of other models.

[0123] Therefore, for these two time points, the XGBoost model will give Pred_t extremely high weights, and the final output will be very close to the Transformer's prediction.

[0124] The final output of the XGBoost model (Final_Prediction): Final_Prediction(10:00) = 4480 (very close to Pred_t=4500) Final_Prediction(20:00) = 5150 (very close to Pred_t=5200) Step 4: Intelligent charging scheduling decision The charging scheduling system receives the complete 24-hour Final_Prediction sequence.

[0125] 1. Calculate the number of devices required in each future time period (N_T): N_T = Final_Prediction(T) / Efficiency coefficient (40) N_10:00 = 4480 / 40 = 112 units N_20:00 = 5150 / 40 = 129 units 2. Assess the number of future available devices (M_T): The system simulates device power consumption and charging status for the next 24 hours. Assuming there are currently 120 online devices, based on their current power consumption and task consumption model, the charging scheduling system predicts: M_10:00 = 110 units (10 devices were offline charging before 10:00 due to depleted battery power) M_20:00 = 105 units (After a day of consumption, more devices have battery levels below the threshold). 3. Supply and demand comparison and decision generation: Table 4 below is a schematic table of an active charging scheduling instruction provided in an embodiment of this application.

[0126]

[0127] Table 4 In this embodiment, a charging scheduling method based on multi-model fusion is provided. The method includes: constructing a time-series dataset, wherein the time-series dataset includes historical task volumes corresponding to each historical time period; simultaneously inputting the time-series dataset into an LSTM base model, a Transformer base model, and a TCN base model to obtain a first prediction result output by the LSTM base model, a second prediction result output by the Transformer base model, and a third prediction result output by the TCN base model; fusing the first prediction result, the second prediction result, and the third prediction result to obtain a fused prediction result; calculating the number of devices required for each future time period based on the fused prediction result; and performing charging scheduling based on the number of devices required for each future time period, combined with the current online devices.

[0128] By fusing the prediction results of LSTM, Transformer, and TCN base models, the system comprehensively captures long-range dependencies, short-term memory, and local patterns in the task sequence. Based on the fused prediction results and combined with the current online equipment, the system can dynamically optimize the charging strategy and improve the robustness of charging scheduling.

[0129] Example 2 Please see Figure 7 , Figure 7 This is a schematic diagram of a charging scheduling system based on multi-model fusion provided in an embodiment of this application.

[0130] like Figure 7 As shown, the charging scheduling system 70 based on multi-model fusion includes: The data preprocessing module 701 is used to construct a time series dataset, which includes the historical task volume corresponding to each historical time period. The multi-model parallel prediction module 702 is used to simultaneously input the time series dataset into the LSTM base model, the Transformer base model, and the TCN base model to obtain the first prediction result output by the LSTM base model, the second prediction result output by the Transformer base model, and the third prediction result output by the TCN base model. The integration and meta-learning module 703 is used to fuse the first prediction result, the second prediction result, and the third prediction result to obtain the fused prediction result; based on the fused prediction result, the number of devices required in each future time period is calculated. The charging scheduling decision module 704 is used to schedule charging based on the number of devices required in various future time periods, combined with the current online devices.

[0131] In the embodiments of this application, the charging scheduling system 70 based on multi-model fusion can also be constructed from hardware devices. For example, the charging scheduling system 70 based on multi-model fusion can be constructed from one or more chips, and the chips can work together to complete the charging scheduling method based on multi-model fusion described in the above embodiments. Furthermore, the charging scheduling system 70 based on multi-model fusion can also be constructed from various logic devices, such as general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), microcontrollers, ARM (Acorn RISC Machine) or other programmable logic devices, discrete gate or transistor logic, discrete hardware components, or any combination of these components.

[0132] The charging scheduling system 70 based on multi-model fusion in this application embodiment can be a device, or a component, integrated circuit, or chip in a terminal. The device can be a mobile electronic device or a non-mobile electronic device. For example, mobile electronic devices can be mobile phones, tablets, laptops, PDAs, in-vehicle electronic devices, wearable devices, ultra-mobile personal computers (UMPCs), netbooks, or personal digital assistants (PDAs), etc., while non-mobile electronic devices can be servers, network-attached storage (NAS), personal computers (PCs), televisions (TVs), ATMs, or self-service machines, etc. This application embodiment does not impose specific limitations.

[0133] The charging scheduling system 70 based on multi-model fusion in this embodiment can be a device with an operating system. This operating system can be Android, iOS, or other possible operating systems; this embodiment does not specifically limit its application.

[0134] The charging scheduling system 70 based on multi-model fusion provided in this application embodiment can achieve... Figure 2 To avoid repetition, the various processes involved will not be detailed here.

[0135] It should be noted that the charging scheduling system 70 based on multi-model fusion described above can execute the charging scheduling method based on multi-model fusion provided in the above embodiments, and has the corresponding functional modules and beneficial effects of the method. Technical details not described in detail in the embodiments of the charging scheduling system based on multi-model fusion can be found in the charging scheduling method based on multi-model fusion provided in the above embodiments.

[0136] In this embodiment, a charging scheduling system based on multi-model fusion is provided, comprising: a data preprocessing module for constructing a time-series dataset, wherein the time-series dataset includes historical task volumes corresponding to each historical time period; a multi-model parallel prediction module for simultaneously inputting the time-series dataset into an LSTM base model, a Transformer base model, and a TCN base model to obtain a first prediction result output by the LSTM base model, a second prediction result output by the Transformer base model, and a third prediction result output by the TCN base model; an integration and meta-learning module for fusing the first, second, and third prediction results to obtain a fused prediction result; calculating the number of devices required for each future time period based on the fused prediction result; and a charging scheduling decision module for performing charging scheduling based on the number of devices required for each future time period, combined with the current online devices.

[0137] By fusing the prediction results of LSTM, Transformer, and TCN base models, the system comprehensively captures long-range dependencies, short-term memory, and local patterns in the task sequence. Based on the fused prediction results and combined with the current online equipment, the system can dynamically optimize the charging strategy and improve the robustness of charging scheduling.

[0138] Example 3 Please see Figure 8 , Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0139] like Figure 8 As shown, the electronic device 80 includes one or more processors 801 and a memory 802. Wherein, Figure 8 Take the 801 processor as an example.

[0140] The processor 801 and the memory 802 can be connected via a bus or other means. Figure 8 Taking the example of a connection via a bus, processor 801 provides computing and control capabilities to perform the aforementioned tasks. Figure 2The proposed charging scheduling method based on multi-model fusion includes: constructing a time-series dataset, which includes the historical task volume corresponding to each historical time period; simultaneously inputting the time-series dataset into an LSTM base model, a Transformer base model, and a TCN base model to obtain a first prediction result output by the LSTM base model, a second prediction result output by the Transformer base model, and a third prediction result output by the TCN base model; fusing the first, second, and third prediction results to obtain a fused prediction result; calculating the number of devices required for each future time period based on the fused prediction result; and performing charging scheduling based on the number of devices required for each future time period, combined with the current online devices.

[0141] By fusing the prediction results of LSTM, Transformer, and TCN base models, the system comprehensively captures long-range dependencies, short-term memory, and local patterns in the task sequence. Based on the fused prediction results and combined with the current online equipment, the system can dynamically optimize the charging strategy and improve the robustness of charging scheduling.

[0142] Processor 801 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), a hardware chip, or any combination thereof; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The aforementioned PLD can be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof.

[0143] Memory 802, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as the program instructions / modules corresponding to the multi-model fusion-based charging scheduling method in the embodiments of this application. Processor 801 can implement the multi-model fusion-based charging scheduling method in any of the above method embodiments by running the non-transitory software programs, instructions, and modules stored in memory 802. Specifically, memory 802 may include volatile memory (VM), such as random access memory (RAM); memory 802 may also include non-volatile memory (NVM), such as read-only memory (ROM), flash memory, hard disk drive (HDD), solid-state drive (SSD), or other non-transitory solid-state storage devices; memory 802 may also include combinations of the above types of memory.

[0144] Memory 802 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, memory 802 may optionally include memory remotely located relative to processor 801, which can be connected to processor 801 via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0145] One or more modules are stored in memory 802. When executed by one or more processors 801, they execute the charging scheduling method based on multi-model fusion in any of the above method embodiments, for example, executing the method described above. Figure 2 The steps shown.

[0146] In the embodiments of this application, the electronic device 80 includes, but is not limited to, a terminal, such as a fixed terminal, like a desktop computer; or, the electronic device 80 includes a mobile terminal, such as a mobile phone, tablet computer, laptop computer, handheld computer, in-vehicle electronic device, wearable device, ultra-mobile personal computer (UMPC), netbook, or personal digital assistant (PDA) terminal.

[0147] In this embodiment, the electronic device 80 may also have wired or wireless network interfaces, keyboards, and input / output interfaces for input and output. The electronic device 80 may also include other components for implementing device functions, which will not be described in detail here.

[0148] This application also provides a non-volatile computer-readable storage medium, such as a memory including program code, which can be executed by a processor to complete the multi-model fusion-based charging scheduling method in the above embodiments. For example, the non-volatile computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CDROM), magnetic tape, floppy disk, and optical data storage device, etc.

[0149] This application also provides a computer program product comprising one or more lines of program code stored in a non-volatile computer-readable storage medium. The processor of an electronic device reads the program code from the non-volatile computer-readable storage medium and executes the program code to complete the method steps of the multi-model fusion-based charging scheduling method provided in the above embodiments.

[0150] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a general-purpose hardware platform, and of course, it can also be implemented by hardware. Those skilled in the art can understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods.

[0151] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and not to limit them; under the concept of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations as described above in different aspects of this application, which are not provided in detail for the sake of brevity; although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A charging scheduling method based on multi-model fusion, characterized in that, The method includes: Construct a time series dataset, wherein the time series dataset includes the historical task volume corresponding to each historical time period; The time series dataset is simultaneously input into an LSTM base model, a Transformer base model, and a TCN base model to obtain a first prediction result output by the LSTM base model, a second prediction result output by the Transformer base model, and a third prediction result output by the TCN base model. The first prediction result, the second prediction result, and the third prediction result are fused to obtain a fused prediction result. Based on the fusion prediction results, calculate the number of devices required for each future time period; Charging schedules are implemented based on the number of devices required in different time periods and the current number of online devices.

2. The method according to claim 1, characterized in that, The process of fusing the first prediction result, the second prediction result, and the third prediction result to obtain a fused prediction result includes: A first feature vector is generated based on the first prediction result, the second prediction result, and the third prediction result; Obtain external features for various future time periods; The external feature is concatenated with the first feature vector to obtain the concatenated feature vector; The concatenated feature vectors are input into the XGBoost model to obtain the fused prediction results.

3. The method according to claim 2, characterized in that, The step of inputting the concatenated feature vector into the XGBoost model to obtain the fused prediction result includes: The first prediction result, the second prediction result, and the third prediction result are weighted to determine the first weight corresponding to the first prediction result, the second weight corresponding to the second prediction result, and the third weight corresponding to the third prediction result. Based on the prediction features of the first prediction result, the second prediction result, and the third prediction result, and combined with the external features, the XGBoost model is used to adjust the first weight, and / or adjust the second weight, and / or adjust the third weight to obtain a fused prediction result.

4. The method according to claim 3, characterized in that, The first prediction result includes the first task volume for each future time period, the second prediction result includes the second task volume for each future time period, and the third prediction result includes the third task volume for each future time period. The prediction features include the quantitative relationship between the first task quantity, the second task quantity, and the third task quantity; The external features include one or more of the following: time period features, weekday features, holiday features, and promotional features.

5. The method according to claim 3, characterized in that, The step of calculating the number of devices required for each future time period based on the fusion prediction results includes: Based on the fusion prediction results and combined with the efficiency coefficient, the number of devices required in each future time period is calculated, where the number of devices required in each future time period = fusion prediction result / efficiency coefficient, and the efficiency coefficient represents the amount of tasks that one device can handle per unit time.

6. The method according to claim 5, characterized in that, The method further includes: Obtain the total number of tasks completed within multiple preset historical time periods; The efficiency coefficient is calculated based on the total amount of tasks completed within multiple preset historical time periods. The efficiency coefficient is calculated as the total number of tasks completed within a preset number of historical time periods divided by the total number of equipment operating hours, where the total number of equipment operating hours is the sum of the operating hours of all equipment involved in the operation.

7. The method according to claim 1, characterized in that, The charging scheduling based on the number of devices required in future time periods, combined with the current number of online devices, includes: Obtain the current status information of online devices, wherein the status information includes battery level information; Based on the power consumption information, predict the number of devices available in various future time periods; If the number of devices required in each future time period is less than or equal to the number of available devices, then the charging strategy decided in the previous time period will be maintained. If the number of devices required in future time periods exceeds the number of available devices, the charging strategy decided in the previous time period is updated, including: generating an active charging scheduling instruction to schedule the device control system to execute the active charging scheduling instruction.

8. The method according to claim 7, characterized in that, The active charging scheduling command is used for: Mark future periods of high demand; and, Before the peak demand period begins, the first device to enter the charging queue must not be allowed to charge; and, Prioritize charging the second power source during the current idle period; Among them, the power of the first power device is greater than the first power threshold, the power of the second power device is less than the second power threshold, and the first power threshold is greater than the second power threshold.

9. A charging scheduling system based on multi-model fusion, characterized in that, include: The data preprocessing module is used to construct a time series dataset, wherein the time series dataset includes the historical task volume corresponding to each historical time period. The multi-model parallel prediction module is used to simultaneously input the time series dataset into the LSTM base model, the Transformer base model, and the TCN base model to obtain the first prediction result output by the LSTM base model, the second prediction result output by the Transformer base model, and the third prediction result output by the TCN base model. An integration and meta-learning module is used to fuse the first prediction result, the second prediction result, and the third prediction result to obtain a fused prediction result; and to calculate the number of devices required for each future time period based on the fused prediction result. The charging scheduling decision module is used to schedule charging based on the number of devices required in various future time periods, combined with the current online devices.

10. An electronic device, characterized in that, include: At least one processor; At least one memory for storing at least one program; When at least one of the programs is executed by at least one of the processors, such that at least one of the processors implements claim 1 8. Any one of the methods described.

11. A non-volatile computer-readable storage medium storing a processor-executable program, characterized in that, The processor-executable program, when executed by the processor, is used to perform as claimed in claim 1.

8. Any one of the methods described.

Images