Warehouse resource adaptive collaborative allocation system based on multi-time scale deep reinforcement learning

Abstract

The application relates to the technical field of collaborative allocation of warehouse resources, and discloses a warehouse resource adaptive collaborative allocation system based on multi-time scale deep reinforcement learning, which comprises a warehouse data acquisition unit, which acquires equipment operation data and environmental parameters and standardizes the same; a warehouse resource state feature extraction unit, which extracts resource state features to form abnormal features; a reinforcement learning state space construction unit, which analyzes the features to form state space construction parameters; a resource allocation strategy generation unit, which constructs a model to generate a strategy and divides influence factors; and a strategy adaptive optimization unit, which updates and optimizes the strategy by using an algorithm. In addition, the system is provided with a multi-source information fusion unit and a task queue dynamic adjustment module. The system realizes adaptive collaborative allocation of warehouse resources, improves warehouse operation efficiency, adapts to dynamic changes of a warehouse system, and meets the demand of modern warehouse intelligent operation.

Description

Technical Field

[0001] This invention relates to the field of collaborative allocation technology for warehousing resources, specifically to an adaptive collaborative allocation system for warehousing resources based on deep reinforcement learning across multiple time scales. Background Technology

[0002] With the booming development of the e-commerce industry and the rapid progress of the logistics industry, the scale and complexity of warehousing operations are increasing day by day. Warehousing operations involve the collaborative operation of multiple types of equipment, such as automated guided vehicles (AGVs), stacker cranes, and forklifts. At the same time, multiple links such as goods storage, handling, and sorting need to be considered, which places extremely high demands on the rational allocation of warehousing resources.

[0003] Traditional warehouse resource allocation methods largely rely on human experience or simple rule-based algorithms. When allocating resources manually, warehouse managers make decisions about equipment tasks and storage location arrangements based on past experience. However, in large-scale, high-frequency operation scenarios, manual judgment is inefficient and prone to errors. For example, during shopping festivals, order volumes surge, making it difficult for manual allocation to quickly and accurately schedule equipment and plan storage locations. This can easily lead to a coexistence of equipment idleness and overuse, with some storage locations crowded while others are underutilized. While simple rule-based algorithms can improve allocation efficiency to some extent, they lack adaptability to dynamic changes in the warehouse system. These algorithms are typically based on fixed parameters and logic, unable to perceive real-time changes in equipment operating status, environmental parameters, and task priority adjustments. When equipment malfunctions, storage conditions change, or urgent tasks are added, resource allocation schemes based on simple rule-based algorithms often cannot be adjusted in time, resulting in decreased warehouse operation efficiency and significant resource waste.

[0004] Some existing warehousing resource allocation technologies have attempted to incorporate intelligent methods, such as using single machine learning algorithms for resource scheduling. However, most of these methods only consider resource allocation at a single time scale and cannot comprehensively handle the dynamic changes across different time dimensions in a warehousing system. Task cycles, equipment maintenance cycles, and cargo storage cycles in warehousing operations all exhibit differences in time scale. Algorithms based on a single time scale struggle to coordinate resource demands across these different time scales, making it difficult to achieve optimal resource allocation. Furthermore, existing technologies often lack in-depth mining and effective integration of multi-dimensional data in terms of data processing and feature extraction. They fail to fully extract key features reflecting the status of warehousing resources, resulting in a lack of comprehensive and accurate data support for resource allocation strategy formulation. This hinders the adaptive and collaborative allocation of warehousing resources and fails to meet the demands of efficient and intelligent modern warehousing operations. Summary of the Invention

[0005] The purpose of this invention is to provide a warehouse resource adaptive collaborative allocation system based on deep reinforcement learning across multiple time scales, in order to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a multi-timescale deep reinforcement learning-based adaptive collaborative allocation system for warehouse resources, the system comprising:

[0007] The warehouse data acquisition unit is used to collect real-time operating data and environmental parameters of various types of equipment in the warehouse scenario, and to standardize the data to form structured data with a unified format.

[0008] The warehouse resource status feature extraction unit is used to complete the resource collaborative allocation task within the warehouse space and extract the resource status features of different storage locations using the structured data with the unified format and the multi-dimensional resource status feature extraction method in the warehouse management center, and to form abnormal features. The resource status features include: inventory turnover rate, storage location utilization rate, equipment idle rate, handling path length, and task waiting time.

[0009] The reinforcement learning state space construction unit is used to analyze and cluster the resource collaborative allocation tasks and resource state characteristics of different storage locations in the storage space based on the abnormal characteristics of each unit time step. At the same time, it analyzes the device interaction frequency under the dynamic load of different storage locations to form the state space construction parameters of different storage location dynamic loads.

[0010] The resource allocation strategy generation unit is used to model different storage locations in the warehouse space to form a resource allocation model. The resource allocation model includes a global strategy model, a local optimization model, and a multi-objective collaborative model based on deep reinforcement learning. Based on the resource allocation model and the structured data with the unified format, the unit generates resource allocation strategies for corresponding inventory scheduling, equipment collaboration, and path planning behaviors, and classifies the influencing factors of resource allocation, thereby completing the adaptive collaborative allocation of resources in the warehouse space.

[0011] The strategy adaptive optimization unit is used to model and identify the resource allocation strategy of the storage space using deep reinforcement learning algorithm through resource adaptive collaborative allocation parameters of the storage space, and to update the parameters of the deep reinforcement learning algorithm using experience replay mechanism and policy gradient optimization.

[0012] Preferably, the warehouse data acquisition unit includes:

[0013] The multi-source data acquisition module is used to collect real-time operating data and environmental parameters of various types of equipment generated by different storage locations in the storage space per unit time step;

[0014] The data standardization module is used to standardize the real-time operating data and environmental parameters of the various types of devices into corresponding resource parameters according to the preset Z-score transformation;

[0015] The outlier handling module is used to process outliers in the standardized resource parameters to obtain interference-free observation data. Outlier handling includes deletion, interpolation, and statistical learning methods.

[0016] The dynamic load source tracking module is used to track the interaction influence data of different devices from the interference-free observation data, and to standardize the interaction influence data of different devices according to the dynamic load source of the storage location to obtain the dynamic load source information of the device that matches the dynamic load source of the storage location.

[0017] The environmental impact dynamic load information parsing module is used to analyze the environmental impact of the dynamic load source information of the equipment, and form different environmental dynamic load source parameters as structured data with a unified format.

[0018] Preferably, the warehouse resource status feature extraction unit includes:

[0019] The edge computing node adaptation module is used to preset edge computing node parameters, adapt to warehouse throughput and operational complexity, and establish an adaptation statistics table.

[0020] The warehouse management center module is used to adjust the extraction steps and preset the extraction parameters of the multi-dimensional resource status feature extraction method in the warehouse management center.

[0021] The multi-warehouse location resource status feature extraction module is used to periodically extract resource status features of different warehouse locations based on the multi-dimensional resource status feature extraction method, the uniformly formatted structured data, and the adaptation statistical table, and generate abnormal resource status features of different warehouse locations.

[0022] The resource status feature extraction module is used to analyze the execution status of resource collaborative allocation tasks by edge computing nodes. For resource collaborative allocation tasks with discovered resource status features, the module performs judgment and analysis, and generates collaborative allocation resource status anomaly features. The collaborative allocation resource status anomaly features and the resource status anomaly features of different storage locations form the anomaly features.

[0023] Preferably, the reinforcement learning state space construction unit includes:

[0024] The warehouse resource status parameter parsing module is used to locate the storage location based on the abnormal characteristics of the collaboratively allocated resource status and the abnormal characteristics of the resource status of different storage locations. This includes warehouse resource status parameter load parsing, warehouse resource status parameter environment parsing, and warehouse resource status parameter anomaly location.

[0025] The resource status parameter classification and processing module is used to classify and form the warehouse resource status parameter information in the collaborative allocation resource status anomaly features and different warehouse location resource status anomaly features, and obtain the data processing flow corresponding to the warehouse resource status parameter information, forming the warehouse resource status parameter reference benchmark, warehouse resource status parameter calculation model optimization coefficients and noise types;

[0026] The dynamic load module is used to detect the device interaction of the dynamic load of different storage locations at all time steps, and to statistically analyze the sources of the dynamic load of different storage locations to form the state space construction parameters of the dynamic load of different storage locations.

[0027] Preferably, the resource allocation strategy generation unit includes:

[0028] The resource allocation model management module is used to build and manage global policy models, local optimization models, and multi-objective collaborative models based on deep reinforcement learning.

[0029] The allocation strategy generation module is used to determine the resource allocation security of different storage locations, systems and management terminals based on real-time operating data and environmental parameters of multiple types of equipment, global strategy model, local optimization model and multi-objective collaborative model, and adopts a preset state coding mechanism and strategy generation algorithm to obtain allocation strategy parameters.

[0030] The task difficulty assessment module is used to quantitatively assess the task difficulty of different storage locations based on the resource allocation safety assessment parameters and the dynamic load source of equipment task complexity, and obtain task difficulty assessment parameters.

[0031] The resource allocation influencing factor classification module is used to automatically complete the resource allocation security assessment for different storage locations, systems, and management terminals based on the allocation strategy parameters and operation difficulty assessment parameters, and to generate a resource allocation influencing factor record.

[0032] The allocation process visualization module is used to visualize the system resource allocation process.

[0033] Preferably, the strategy adaptive optimization unit includes:

[0034] The operation status recording module is used to continuously monitor the working status of different storage locations and the dynamic load sources of system equipment operation complexity, forming records of different storage location statuses and storage system operation complexity, providing a reference for subsequent resource allocation and collaborative scheduling;

[0035] The strategy effectiveness evaluation module is used to determine the resource allocation security of different storage locations, obtain resource allocation security test parameters, evaluate the duration of different storage locations or allocation strategy levels, and evaluate the accuracy of different storage locations in allocating future resource allocation tasks.

[0036] The reinforcement learning network optimization module is used to update the parameters of the deep reinforcement learning algorithm based on the resource allocation security of different storage locations, the allocation accuracy of different storage locations for future resource allocation tasks, and the experience replay mechanism and policy gradient optimization.

[0037] Preferably, the system further includes a multi-source information fusion unit, used to integrate equipment sensing data within the warehouse space, management system instruction data, and external environmental data to form a fused information set;

[0038] The multi-source information fusion unit includes a time synchronization module for timestamp alignment of data from different sources; a spatial calibration module for spatial positioning calibration of equipment sensing data based on warehouse location coordinates; and a conflict resolution module for identifying conflicting information from different data sources and fusion processing through a voting mechanism or credibility weight. The fused information set serves as a supplementary input to the uniformly formatted structured data.

[0039] Preferably, the multi-objective collaborative model includes an inventory balancing sub-model, an equipment load balancing sub-model, and a path overlap avoidance sub-model; the inventory balancing sub-model is used to dynamically adjust the allocation ratio of storage locations to make the inventory level of each storage location approach a preset average value; the equipment load balancing sub-model is used to make the difference in the operating load of different equipment less than a threshold through task allocation strategies; and the path overlap avoidance sub-model is used to reduce the probability of intersection and overlap of equipment handling paths through path planning algorithms.

[0040] Preferably, the resource allocation strategy generation unit further includes a task queue dynamic adjustment module, used to dynamically adjust the generated task execution sequence according to the real-time resource status and strategy generation parameters; the task queue dynamic adjustment module includes a status awareness submodule, used to obtain the inventory status of each storage location, equipment operation status and task completion progress in real time; an impact assessment submodule, used to analyze the degree of impact of the current task queue status change on the task sequence execution efficiency; and a sequence rearrangement submodule, used to perform local or global rearrangement of the task queue according to the impact assessment results.

[0041] Preferably, the local or global reordering operations preserve the execution order of high-priority tasks.

[0042] Compared with the prior art, the beneficial effects of the present invention are:

[0043] This invention presents a multi-timescale deep reinforcement learning-based adaptive collaborative allocation system for warehouse resources. The system collects real-time operational data and environmental parameters from various types of equipment through a warehouse data acquisition unit, and performs standardized processing to form structured data, providing an accurate and unified data foundation for subsequent operations. The warehouse resource status feature extraction unit utilizes structured data and multi-dimensional resource status feature extraction methods to comprehensively and accurately extract resource status features such as inventory turnover rate and storage space utilization rate, and to generate anomaly features, enabling the system to promptly detect potential problems in the warehouse resource allocation process.

[0044] The reinforcement learning state space construction unit analyzes and clusters resource state features based on anomaly characteristics, and simultaneously analyzes equipment interaction frequency to form state space construction parameters, providing deep reinforcement learning algorithms with inputs that closely reflect actual warehouse operations. The resource allocation strategy generation unit constructs a global strategy model, a local optimization model, and a multi-objective collaborative model, which, combined with structured data, generate resource allocation strategies and identify influencing factors. This achieves comprehensive optimization of behaviors such as inventory scheduling, equipment collaboration, and path planning, enabling the rational allocation of resources based on the actual needs of different storage locations and the overall system objectives.

[0045] The adaptive strategy optimization unit updates the parameters of the deep reinforcement learning algorithm through experience replay and strategy gradient optimization, enabling the system to continuously adapt to the dynamic changes in the warehousing environment and optimize resource allocation strategies. Furthermore, the multi-source information fusion unit integrates equipment perception data, management system instruction data, and external environment data, further improving data integrity and accuracy. The task queue dynamic adjustment module adjusts the task execution sequence based on real-time resource status, ensuring the execution of high-priority tasks and improving the overall efficiency and flexibility of warehousing operations. This system forms a complete closed loop from data acquisition, feature extraction, model building, strategy generation to strategy optimization, effectively solving the problems of low efficiency and poor adaptability in traditional warehousing resource allocation. It achieves adaptive and collaborative allocation of warehousing resources, significantly improving warehousing operation efficiency, reducing resource waste, and enhancing the warehousing system's ability to cope with complex business scenarios. Attached Figure Description

[0046] Figure 1 This is a schematic diagram illustrating the working principle of the multi-timescale deep reinforcement learning-based adaptive collaborative allocation system for warehouse resources described in this invention.

[0047] Figure 2 A flowchart for data acquisition and standardized processing in the warehouse data acquisition unit;

[0048] Figure 3 Flowchart for feature extraction and anomaly generation in the warehouse resource status feature extraction unit;

[0049] Figure 4A flowchart for the state parameter parsing and space construction of the state space construction unit for reinforcement learning. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] Please see Figures 1-4 The multi-timescale deep reinforcement learning-based adaptive collaborative allocation system for warehouse resources described in this invention is specifically implemented as follows:

[0052] During actual operation, the warehouse data acquisition unit collects real-time operating data and environmental parameters of various types of equipment in the warehouse environment. This data includes, but is not limited to, forklift trajectories, stacker crane operating status, and warehouse temperature and humidity. After data collection, the data is standardized using a preset Z-score transform, converting the real-time operating data and environmental parameters of various equipment types into corresponding resource parameters. Simultaneously, outlier handling is performed on the standardized resource parameters using deletion, interpolation, and statistical learning methods to obtain interference-free observation data. From this interference-free observation data, the interaction and impact of different equipment are tracked. After standardization, dynamic load source information matching the dynamic load source of the warehouse location is obtained, followed by environmental impact analysis to form structured data with a unified format.

[0053] After the warehouse resource status feature extraction unit acquires structured data in a uniform format, the edge computing node adaptation module pre-sets the edge computing node parameters to adapt them to warehouse throughput and operational complexity, and establishes an adaptation statistics table. The warehouse management center module adjusts the extraction steps of the multi-dimensional resource status feature extraction method and presets the extraction parameters. Based on the multi-dimensional resource status feature extraction method, structured data, and adaptation statistics table, the multi-warehouse location resource status feature extraction module performs periodic resource status feature extraction on different warehouse locations, generating abnormal resource status features for different warehouse locations. The resource status feature extraction module analyzes the execution status of resource collaborative allocation tasks through edge computing nodes, judges and analyzes resource collaborative allocation tasks that discover resource status features, and generates abnormal collaborative allocation resource status features. The two together form abnormal features.

[0054] The reinforcement learning state space construction unit analyzes and clusters the resource collaborative allocation tasks and resource state characteristics of different storage locations within the storage space based on the abnormal characteristics of the collaborative allocation resource state and the abnormal characteristics of resource state of different storage locations. The storage resource state parameter analysis module performs location positioning on the abnormal characteristics of the collaborative allocation resource state and the abnormal characteristics of resource state of different storage locations, including storage resource state parameter load analysis, storage resource state parameter environment analysis, and storage resource state parameter anomaly positioning; the resource state parameter classification and processing module classifies and forms the storage resource state parameter information in the abnormal characteristics, obtains the corresponding data processing flow, and forms the storage resource state parameter reference benchmark, the coefficients to be optimized in the storage resource state parameter calculation model, and the noise type; the dynamic load module detects the equipment interaction of the dynamic load of different storage locations at all time steps, statistically analyzes the sources of the dynamic load of equipment in different storage locations, and forms the state space construction parameters of the dynamic load of different storage locations.

[0055] The resource allocation strategy generation unit models different storage locations within the warehouse space, constructing a global strategy model, a local optimization model, and a multi-objective collaborative model based on deep reinforcement learning. Based on real-time operating data and environmental parameters from various equipment types, and each model, a pre-defined state coding mechanism and strategy generation algorithm are used to assess the security of resource allocation for different storage locations, systems, and management terminals, yielding allocation strategy parameters. Based on the resource allocation security assessment parameters and the dynamic load sources of equipment operational complexity, the operational difficulty of different storage locations is quantitatively evaluated, yielding operational difficulty assessment parameters. The unit automatically completes the resource allocation security assessment for different storage locations, systems, and management terminals based on the allocation strategy parameters and operational difficulty assessment parameters, forming a record of resource allocation influencing factors and visualizing the system resource allocation process.

[0056] The strategy adaptive optimization unit continuously monitors the working status of different storage locations and the dynamic load sources of system equipment operational complexity, generating records. It determines the resource allocation security of different storage locations, obtains resource allocation security test parameters, evaluates the duration of different storage locations or allocation strategy levels, and assesses the accuracy of different storage locations in allocating future resource allocation tasks. Based on the resource allocation security of different storage locations, combined with the accuracy of different storage locations in allocating future resource allocation tasks, and utilizing an experience replay mechanism and strategy gradient optimization, the deep reinforcement learning algorithm's parameters are updated.

[0057] Example 1: The warehouse data acquisition unit completes data acquisition and processing through the collaboration of multiple modules. The multi-source data acquisition module collects real-time operating data and environmental parameters of various types of equipment generated at different storage locations within the warehouse space at fixed unit time steps. In the warehouse operation site, various equipment such as Automated Guided Vehicles (AGVs), cranes, and forklifts generate a large amount of data during their operation. AGV operating data includes travel speed, travel direction, current position coordinates, and cargo loading status; crane operating data includes lifting height, travel trajectory, and hook status; forklifts generate data such as the weight of transported goods, travel distance, and working time. For environmental parameter acquisition, information such as light intensity, air pressure, temperature, humidity, and air quality within the warehouse is collected in real time. This data is gathered from various sensors at different storage locations and the built-in monitoring systems of the equipment to the multi-source data acquisition module.

[0058] After collecting the raw data, the data standardization module processes it. Data standardization is based on preset rules and methods, aiming to transform data of different types and magnitudes into a unified and comparable format. Each data dimension is processed according to a predetermined transformation method, ensuring that data with different measurement standards and value ranges are placed under the same standard for subsequent analysis and use. For example, for equipment operating speed data, some equipment speeds may be measured in kilometers per hour, while others are measured in meters per second. The data standardization module will convert them to the same unit and map the data to a reasonable range according to a specific mapping relationship.

[0059] The outlier handling module works on the standardized data. The deletion method handles data that significantly deviates from the normal range. In actual warehouse operations, extreme data may occur due to sensor malfunctions, data transmission errors, etc. For example, if an AGV's speed record shows a value far exceeding its normal operating speed range, this data will be directly removed by the deletion method. The interpolation method is suitable for handling missing data or minor anomalies. When equipment operation data is missing within a certain time period, or when some data has slight anomalies, the interpolation method estimates and fills in the missing or abnormal data based on the distribution of adjacent data and using reasonable calculation methods. For example, if the lifting height data of a crane is missing values ​​for several time points within a certain period, the interpolation method will calculate a reasonable estimate of the missing data based on the lifting height data of the preceding and following time points using linear interpolation or other suitable interpolation algorithms. Statistical learning methods, on the other hand, construct data distribution models to conduct in-depth data analysis, thereby identifying and handling outliers. It learns the distribution patterns of data, builds mathematical models, and when new data comes in, it uses the model to determine whether the data belongs to the normal distribution range. If it does not, it is identified as an outlier and processed accordingly, thus obtaining interference-free observation data.

[0060] The dynamic load source tracking module acquires information from interference-free observation data. It tracks the impact of interactions between different devices by analyzing the strength and frequency of signals exchanged between them. During warehousing operations, frequent interactions occur between devices, such as the collaboration between AGVs and shelves, or cranes and forklifts. These interactions generate various signals. By capturing and analyzing these signals, the dynamic load source tracking module understands the interactions between different devices and determines which interactions affect the dynamic load of the storage location and to what extent. Then, based on the dynamic load source of the storage location, this data is standardized to obtain device dynamic load source information that matches the dynamic load source of the storage location. For example, if the load of a storage location suddenly increases within a specific time period, the dynamic load source tracking module, through analysis, finds that it is due to frequent cargo handling interactions by several AGVs at that location. It then standardizes the relevant device interaction impact data to clarify the relationship between these devices and the dynamic load of that storage location.

[0061] The Environmental Impact Dynamic Load Information Analysis Module further analyzes the dynamic load source information of equipment. It considers the impact of environmental factors on equipment operation; for example, temperature changes may affect equipment operating speed and performance, humidity may affect cargo storage and equipment metal components, and light intensity may affect operator efficiency, thus indirectly affecting equipment operation. Through analyzing the relationship between these environmental factors and equipment operation, the module can perform environmental impact analysis on the equipment dynamic load source information, forming parameters for different environmental dynamic load sources. These parameters, ultimately presented as structured data in a standardized format, provide data support for subsequent tasks such as extracting warehousing resource status characteristics. This enables the system to perform more rational analysis and allocation of warehousing resources based on this accurate and standardized data.

[0062] Example 2: In the actual operation of the warehouse resource status feature extraction unit, the edge computing node adaptation module begins to function during the system initialization phase. Upon system startup, staff, based on accumulated historical warehouse data covering different time periods, the complexity of various operations, and equipment load conditions, and combined with extensive operational experience, preset relevant parameters for the edge computing nodes. These parameters include, but are not limited to, the processing capacity of the edge computing nodes, such as the amount of data that can be processed per second; storage capacity, i.e., the total amount of data that can be stored; and data processing priority. By appropriately setting these parameters, the edge computing nodes can adapt to the needs of warehouse operations of varying scales and complexities. After parameter presets are completed, the edge computing node adaptation module establishes a detailed adaptation statistics table. This table records the processing effects of the edge computing nodes on various types of warehouse data under different parameter settings, such as processing time and processing accuracy, for subsequent parameter adjustments and optimizations.

[0063] The warehouse management center module provides managers with a user-friendly human-computer interface. Within this interface, managers can flexibly adjust the extraction steps of the multi-dimensional resource status feature extraction method. For example, based on the actual situation of warehouse operations, the feature extraction time interval can be changed. During periods of frequent goods movement, the time interval can be shortened to obtain resource status features more promptly; during periods of relatively stable operations, the time interval can be appropriately extended to reduce computational resource consumption. Simultaneously, managers can set data filtering conditions, such as selecting only data from specific types of equipment or data within specific time periods, to improve the targeting and effectiveness of feature extraction. Furthermore, this module allows for the preset of extraction parameters, such as weighting coefficients to determine the importance of different resource status features in the comprehensive evaluation, and thresholds to determine whether the resource status is within a normal range. These settings enable the multi-dimensional resource status feature extraction method to better meet the actual needs of warehouse operations.

[0064] The multi-warehouse location resource status feature extraction module extracts resource status features for each warehouse location according to a pre-set cycle, based on a multi-dimensional resource status feature extraction method, standardized structured data, and adapted statistical tables. During the extraction process, several key indicators are calculated to evaluate the resource status of the warehouse location, including: inventory turnover rate (the number of times inventory turns over within a certain period, reflecting the speed of goods flow); location utilization rate (measuring the utilization of warehouse space); equipment idle rate (the proportion of equipment not in use within a certain period); handling path length (recording the distance traveled during goods handling); and task waiting time (the time from task assignment to execution). These calculated indicators are compared with normal ranges. If any indicator exceeds or falls below the normal range, the resource status of the corresponding warehouse location is identified as abnormal, and different abnormal resource status features for different warehouse locations are generated. For example, if the location utilization rate of a warehouse location is consistently lower than the normal level, it indicates that the space of that warehouse location is not being fully utilized, and it will be marked as abnormal.

[0065] The resource status feature extraction module leverages the powerful computing capabilities of edge computing nodes to analyze the execution status of resource collaborative allocation tasks in real time. During task execution, it comprehensively analyzes and judges the relevant resource status features. If an excessively long task execution time is detected, it may indicate unreasonable resource allocation or equipment malfunction; unreasonable resource utilization, such as some equipment being idle for extended periods while others are overworked, will generate abnormal collaborative allocation resource status features. For example, in a cargo handling task, if unreasonable path planning causes the AGV to travel an excessively long distance, exceeding the expected task execution time, the resource status feature extraction module will identify this anomaly and generate corresponding abnormal collaborative allocation resource status features. The abnormal resource status features of different storage locations and the abnormal collaborative allocation resource status features together constitute a complete set of anomaly features. These features provide crucial information for subsequent reinforcement learning state space construction and resource allocation strategy generation, enabling the system to promptly adjust and optimize for problems arising in warehousing operations, achieving rational allocation and efficient utilization of warehousing resources.

[0066] Example 3: The reinforcement learning state space construction unit plays a crucial role in the entire adaptive collaborative allocation system of warehouse resources, transforming the state characteristics of warehouse resources into state space construction parameters suitable for reinforcement learning processing. This unit consists of a warehouse resource state parameter parsing module, a resource state parameter classification and processing module, and a dynamic load module that work together to complete the task.

[0067] The warehouse resource status parameter analysis module processes abnormal features of collaboratively allocated resource status and abnormal features of different storage locations. In actual warehousing environments, various types of equipment are equipped with positioning devices. For example, automated guided vehicles (AGVs) are equipped with high-precision positioning sensors that can acquire their own location information in real time. When abnormal features occur, this module first uses the equipment's positioning information, combined with task-related information, such as the storage location of goods, the start and end points of the handling task, etc., to locate the abnormal feature in the storage location.

[0068] In terms of load analysis of warehousing resource status parameters, taking AGVs as an example, data such as their transport weight and running time are analyzed. The transport weight can be obtained through weighing sensors installed on the AGV, while the running time is recorded by the system from the start to the end of the task. Analyzing this data helps understand the load situation of the equipment under different time periods and tasks. When performing environmental analysis, the impact of humidity in the warehouse on goods storage is considered. Humidity sensors are installed in the warehouse to monitor ambient humidity in real time. Different types of goods have different humidity requirements; when humidity exceeds the suitable storage range, it may affect the quality of the goods, thus affecting the status of warehousing resources. By analyzing the relationship between humidity data and the storage status of goods, the impact of environmental factors on resource status is assessed. During anomaly localization, combining equipment location information and timestamps, the specific location and time of the anomaly are accurately determined. For example, when an anomaly is found in goods in a certain storage location, the equipment's operating trajectory and time records determine that the anomaly occurred during a specific handling process, identifying the specific equipment and time point.

[0069] The resource status parameter classification and processing module categorizes warehouse resource status parameter information from abnormal characteristics. It classifies data according to data type, such as numerical data (equipment operating speed, cargo weight, etc.), character data (equipment model, cargo name, etc.), and time-based data (task start time, equipment failure time, etc.); it also further subdivides data based on its source, such as data from sensors or management systems. Corresponding data processing procedures are developed for different data types.

[0070] Statistical analysis methods are used to establish reference standards for warehouse resource status parameters. For example, for the operating speed data of a certain piece of equipment, a large number of data samples are collected over a period of time. By calculating the average value ,in Indicates the number of data samples. Indicates the first We obtain an average operating speed of the equipment from a data sample, using this as a reference benchmark. Based on the data's trend, we determine the coefficients to be optimized in the calculation model for warehouse resource status parameters. For example, if we find that the equipment's operating speed gradually decreases with increasing years of use, we establish a regression model to analyze the relationship between the two, determine the coefficients in the model, and thus more accurately predict the equipment's operating status. We identify the types of noise in the data; common noise types include random noise and periodic noise. Random noise is usually caused by minor sensor errors or random environmental interference; periodic noise may be related to the periodic operation of the equipment or the periodic operation of the warehouse's ventilation system. By identifying the noise type, we adopt corresponding filtering and denoising methods to improve data quality.

[0071] The dynamic load module monitors the real-time dynamic load interactions of equipment at different storage locations using various sensors deployed throughout the storage space. These sensors capture interaction signals between devices, recording information such as interaction time and content. In actual warehousing operations, frequent interactions occur between equipment, such as the handover of goods between AGVs and stacker cranes, and the collaborative operation between cranes and forklifts. The dynamic load module analyzes this interaction information to statistically analyze the sources of dynamic load at different storage locations. When a large number of goods enter and exit a storage location in a short period, causing multiple devices to operate frequently and increasing the load, the dynamic load module can identify this uneven load distribution caused by concentrated task allocation. If a device malfunctions, affecting the entire workflow and requiring other devices to undertake additional tasks, thus causing load changes, the module can also accurately analyze that the device malfunction is the source of the increased load. By processing and analyzing this information, state space construction parameters for the dynamic load of different storage locations are formed, providing key data support for the generation of subsequent resource allocation strategies. This enables the system to formulate more reasonable resource allocation strategies based on the actual load of the storage locations.

[0072] Example 4: The resource allocation strategy generation unit is responsible for generating specific resource allocation strategies in the warehouse resource adaptive collaborative allocation system. Its multiple modules work together to achieve this goal.

[0073] The resource allocation model management module builds various models based on a deep learning framework. In a smart warehousing center scenario of a large e-commerce enterprise, the warehouse contains tens of thousands of storage locations, storing goods of different categories and specifications, and dozens of automated guided vehicles (AGVs), stacker cranes, and other equipment working collaboratively. The global strategy model considers all resources within the warehouse from the perspective of the entire warehousing system. For example, during shopping festivals, when a large number of orders flood in, the global strategy model coordinates the storage arrangements of goods in all storage locations, the task allocation of various equipment, and personnel scheduling, enabling the entire warehousing system to efficiently cope with the surge in business volume, allocate resources holistically, and ensure stable and efficient system operation. The local optimization model addresses specific needs. If a storage location is specifically used to store easily damaged electronic products, the local optimization model optimizes the placement of goods, storage density, and equipment operation paths within that location based on its size, temperature and humidity control conditions, the storage requirements of the electronic products, and the frequency of inbound and outbound operations. This improves the operational efficiency and storage security of that specific storage location.

[0074] The multi-objective collaborative model comprises three sub-models: inventory balancing, equipment load balancing, and path overlap avoidance. These sub-models work in tandem. The inventory balancing sub-model dynamically adjusts the allocation ratio of storage locations depending on the sales season. For example, in summer, when cooling products like air conditioners and fans are in high demand, the model increases the inventory allocation ratio for these products in their respective storage locations while decreasing the inventory ratio for winter clothing, bringing the inventory level of each storage location closer to the preset average. This avoids situations where some storage locations are overstocked while others require frequent restocking due to shortages. The equipment load balancing sub-model allocates tasks based on the performance and current operating status of different equipment when handling goods handling tasks. When a batch of goods needs to be moved from the storage area to the sorting area, if an AGV has been working continuously for a long time and is approaching its load limit, the model will allocate some handling tasks to other idle or less loaded AGVs. This ensures that the difference in workload between different devices is less than a threshold, preventing equipment fatigue and malfunctions, and improving the overall lifespan and operational stability of the equipment. The path overlap avoidance sub-model plays a crucial role when equipment is operating intensively within the warehouse. When multiple AGVs perform transport tasks simultaneously, the path overlap avoidance sub-model uses a path planning algorithm to plan an optimal path for each AGV, reducing the probability of overlapping transport paths, avoiding collisions between AGVs, and improving the safety and smoothness of operations.

[0075] The allocation strategy generation module integrates real-time operating data from various types of equipment, environmental parameters, and different models. For example, suppose at a certain moment, the warehouse temperature rises, affecting the operating speed of some equipment, while some AGVs have low battery levels, and there are urgent orders to process. The allocation strategy generation module first uses a preset state encoding mechanism to convert the state information of these warehousing systems, such as equipment operating speed, battery level, and task urgency, into an input format acceptable to each model. Then, it uses a strategy generation algorithm to comprehensively consider the overall planning of the global strategy model, the requirements of the local optimization model for specific areas, and the objectives of the multi-objective collaborative model. This process assesses the resource allocation safety of different storage locations, systems, and management terminals, resulting in allocation strategy parameters. For instance, it prioritizes allocating urgent order handling tasks to AGVs with sufficient battery power and plans appropriate routes to ensure the safety of both equipment and goods while completing the task.

[0076] The task difficulty assessment module evaluates task difficulty based on resource allocation safety assessment parameters and dynamic load sources related to equipment operation complexity. For goods stored on high-rise shelves that are bulky and heavy, handling not only requires specific large equipment but also presents significant operational challenges and safety risks. The task difficulty assessment module comprehensively considers factors such as task urgency, equipment operation complexity, and potential risks to quantitatively evaluate the task difficulty of the storage location, obtaining task difficulty assessment parameters to provide a reference for subsequent resource allocation.

[0077] The resource allocation influencing factor classification module automatically assesses the safety of resource allocation for different storage locations, systems, and management terminals based on allocation strategy parameters and operational difficulty assessment parameters. In the above scenario, analysis reveals factors influencing resource allocation, including equipment performance (such as AGV battery level and operating speed), task priority (urgent orders take precedence), and cargo characteristics (volume, weight, storage requirements), generating a record of these influencing factors. The allocation process visualization module graphically displays the system's resource allocation process. Managers can intuitively see changes in AGV operating paths and the transfer of goods between storage locations through a visual interface, facilitating real-time monitoring and management of warehousing operations.

[0078] Example 5: In the actual operation of the strategy adaptive optimization unit, the operation status recording module plays a crucial role in continuous monitoring and data recording. In large-scale automated warehousing scenarios, the warehouse contains a large number of automated guided vehicles (AGVs), stacker cranes, sorting robots, and other equipment, as well as numerous storage locations with different functions. The operation status recording module utilizes sensors distributed throughout the warehouse, such as displacement sensors, speed sensors, and current sensors installed on the equipment, and inventory sensors deployed in the storage locations, to continuously monitor the working status of different storage locations and the dynamic load sources of the system equipment's operational complexity.

[0079] Taking AGVs as an example, displacement sensors track their travel trajectory in real time, speed sensors report changes in operating speed, and current sensors monitor the motor's operating current. This data can intuitively reflect the AGV's operating status, including whether it is traveling normally, whether there are any abnormal jams, and whether the load is overloaded. For storage locations, inventory sensors perceive changes in the quantity of goods stored in real time. Combined with inbound and outbound time information, the usage dynamics of storage locations can be clearly understood. At the same time, by analyzing the interaction data between equipment, such as the time and frequency of goods handover between AGVs and stacker cranes, as well as the task allocation records in the task scheduling system, the source of dynamic load on the system's equipment operation complexity can be identified. For example, is it due to a surge in orders leading to excessive task volume, or is it due to load imbalance caused by equipment failure? This information is recorded in real time, forming a detailed operating status log file.

[0080] The strategy effectiveness evaluation module assesses the safety of resource allocation for different storage locations based on pre-defined evaluation criteria. These criteria cover multiple dimensions, including equipment operational safety (e.g., whether there is a risk of collision or whether the equipment exceeds its safe load range during task execution); cargo storage safety (e.g., whether the cargo is in a suitable storage environment and whether there is a possibility of damage due to improper storage); and task execution timeliness (e.g., whether the task is completed within the specified time). By comparing and analyzing these criteria one by one, resource allocation safety test parameters are obtained.

[0081] For example, during an evaluation of a storage location holding precision instruments, it was found that the shelves storing these instruments posed a risk of tipping over due to adjacent goods being stacked too high. This lowered the resource allocation safety score for that storage location. Conversely, for a regular goods storage location, if tasks are completed on time, equipment is operated correctly, and goods are stored in good condition, a higher safety score is given. Simultaneously, the strategy effectiveness evaluation module analyzes the duration of different storage locations or allocation strategy levels. By statistically analyzing the duration of different safety levels, the stability of the current allocation strategy is understood. Furthermore, by combining historical data and current storage conditions, the accuracy of different storage locations in allocating future resource tasks is assessed, determining whether the current resource allocation model can effectively cope with potential future task demands.

[0082] Based on the data provided by the runtime status recording module and the policy effectiveness evaluation module, the reinforcement learning network optimization module optimizes the deep reinforcement learning algorithm. When the system detects that the resource allocation security of a certain storage location is low and it remains unstable for a period of time, the reinforcement learning network optimization module combines the evaluation results of the accuracy of future resource allocation tasks for that storage location with the experience replay mechanism. It randomly extracts experience data related to that storage location from the historical runtime status recording file, including the runtime status, task execution, and resource utilization efficiency under different resource allocation strategies in the past.

[0083] This empirical data was reused to train the deep reinforcement learning algorithm, enabling it to learn from past experiences and avoid repeating erroneous allocation strategies. Simultaneously, policy gradient optimization methods were employed to adjust and update the parameters of the deep reinforcement learning algorithm based on the goals of current resource allocation security and future allocation accuracy. For example, the allocation weight parameters for different resource types in the global policy model were adjusted, or the optimization policy parameters for specific storage locations in the local optimization model were modified. This allows the resource allocation strategy generated by the deep reinforcement learning algorithm to better adapt to the constantly changing dynamic environment of the warehousing system, improving the performance and efficiency of the entire adaptive collaborative allocation system for warehousing resources.

[0084] Furthermore, the multi-source information fusion unit in the system also plays a crucial role. The time synchronization module aligns the timestamps of data from different sources, including equipment sensors, the management system, and external environmental monitoring equipment. For example, the frequency at which equipment sensors collect data may differ from the frequency at which the management system records task times. The time synchronization module unifies these data to the same time reference, ensuring data consistency and preventing data corruption caused by time discrepancies. The spatial calibration module performs spatial location calibration on the equipment's sensing data based on the precise coordinates of the storage location.

[0085] When there is a discrepancy between the AGV's positioning data and the actual location of the storage location, the spatial calibration module corrects the AGV's position information based on the warehouse layout and coordinate system, ensuring the accuracy of the equipment position data and providing a reliable spatial reference for resource allocation. When the conflict resolution module identifies conflicting information from different data sources, such as discrepancies between the quantity of goods reported by equipment sensors and the quantity recorded by the management system, it fuses the conflicting data through a voting mechanism or credibility weighting. If data from multiple sensors corroborate each other, while management system data contradicts them, the sensor data is given higher credibility, and fusion is performed based on the sensor data. The integrated fused information set serves as supplementary input to structured data with a unified format, providing more comprehensive and accurate data support for all units of the system.

[0086] The task queue dynamic adjustment module in the resource allocation strategy generation unit, and its status awareness submodule, continuously acquire information on the inventory status, equipment operating status, and task completion progress of each storage location through a real-time communication interface. During a concentrated order processing period, the status awareness submodule can quickly obtain information such as the current location, remaining power, and operational status of each AGV, as well as changes in the amount of goods stored in each storage location and the number of tasks awaiting processing. Based on this real-time data, the impact assessment submodule uses a specific evaluation algorithm to analyze the degree of impact of changes in the current task queue status on the execution efficiency of the task sequence.

[0087] If a critical piece of equipment experiences a sudden failure, the impact assessment submodule will evaluate the scope and extent of its impact on the overall task sequence execution efficiency based on the type and priority of the tasks performed by the failed equipment, as well as the substitutability of other equipment. The sequence reordering submodule will then perform partial or global reordering of the task queue based on the impact assessment results. For high-priority tasks, such as the sorting of urgent orders, the sequence reordering submodule will prioritize ensuring their execution order remains unaffected. By adjusting the execution order of other low-priority tasks, it will replan the equipment's operational paths and task allocation to optimize the warehousing workflow and ensure that the warehousing system can still operate as efficiently as possible in the event of an emergency.

[0088] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0089] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A warehouse resource self-adaptive collaborative allocation system of multi-time scale deep reinforcement learning, characterized in that , comprising: a warehouse data acquisition unit for acquiring real-time operation data and environmental parameters of multiple types of equipment in a warehouse scene, and performing standardized processing on the data to form structured data with uniform format; a warehouse resource state feature extraction unit for using the structured data with uniform format and a multi-dimensional resource state feature extraction method in a warehouse management center to complete resource collaborative allocation tasks in a warehouse space and extract resource state features of different warehouse locations and form abnormal features, the resource state features including: inventory turnover rate, storage space utilization rate, equipment idle rate, handling path length, task waiting time; a reinforcement learning state space construction unit for analyzing and clustering resource collaborative allocation tasks in a warehouse space and resource state features of different warehouse locations based on abnormal features at a unit time step, and analyzing equipment interaction frequency under dynamic load of different warehouse locations to form state space construction parameters of dynamic load of different warehouse locations; a resource allocation strategy generation unit for modeling different warehouse locations in a warehouse space to form a resource allocation model, the resource allocation model including a global strategy model, a local optimization model and a multi-objective collaborative model based on deep reinforcement learning; generating resource allocation strategies for corresponding inventory scheduling, equipment collaboration and path planning behaviors based on the resource allocation model and the structured data with uniform format, and dividing resource allocation influencing factors to complete resource adaptive collaborative allocation of the warehouse space; a strategy adaptive optimization unit for modeling and allocating strategies of resource allocation strategies of a warehouse space using a deep reinforcement learning algorithm based on resource adaptive collaborative allocation parameters of the warehouse space, and updating parameters of the deep reinforcement learning algorithm using an experience replay mechanism and a policy gradient optimization; The strategy adaptive optimization unit comprises: an operating state recording module for continuously monitoring the working state of different warehouse locations and the dynamic load source of system equipment operation complexity, forming a record of the state of different warehouse locations and the operation complexity of the warehouse system, and providing a reference basis for subsequent resource allocation and collaborative scheduling; a strategy effect evaluation module for determining the resource allocation safety of different warehouse locations, obtaining resource allocation safety test parameters, evaluating the duration of different warehouse locations or allocation strategy levels, and evaluating the allocation accuracy of different warehouse locations for future resource allocation tasks; a reinforcement learning network optimization module for updating parameters of the deep reinforcement learning algorithm using an experience replay mechanism and a policy gradient optimization based on the resource allocation safety of different warehouse locations and the allocation accuracy of different warehouse locations for future resource allocation tasks.

2. The warehouse resource adaptive collaborative allocation system of claim 1, wherein The warehouse data acquisition unit comprises: a multi-source data acquisition module for acquiring real-time operation data and environmental parameters of multiple types of equipment generated by different warehouse locations in a warehouse space at a unit time step; a data standardization module for standardizing the real-time operation data and environmental parameters of multiple types of equipment into corresponding resource parameters according to a preset Z-score transformation. An outlier processing module is configured to perform outlier processing on the standardized resource variables to obtain non-interference observation data; the outlier processing includes deletion method, interpolation method and statistical learning method processing; A dynamic load source tracking module is configured to track different device interaction influence degree data from the non-interference observation data, and to standardize the different device interaction influence degree data according to the warehouse site dynamic load source to obtain device dynamic load source information matched with the warehouse site dynamic load source. An environmental impact dynamic load information analysis module is configured to analyze the device dynamic load source information according to the environmental impact to form different environmental dynamic load source variables as the format-unified structured data.

3. The warehouse resource adaptive collaborative allocation system of claim 1, wherein The warehouse resource state feature extraction unit includes: An edge computing node adaptation module is configured to preset edge computing node parameters, adapt warehouse throughput and job complexity, and establish an adaptation statistical table. A warehouse management center module is configured to adjust extraction steps and preset extraction parameters of a multi-dimensional resource state feature extraction method in a warehouse management center. A multi-warehouse site resource state feature extraction module is configured to periodically extract resource state features of different warehouse sites according to the multi-dimensional resource state feature extraction method, the format-unified structured data and the adaptation statistical table, and generate different warehouse site resource state abnormal features. A resource state feature extraction module is configured to analyze the execution state of resource collaborative allocation tasks by edge computing nodes, judge and analyze resource collaborative allocation tasks with discovered resource state features, and generate collaborative allocation resource state abnormal features; the collaborative allocation resource state abnormal features and different warehouse site resource state abnormal features form the abnormal features.

4. The warehouse resource adaptive co-allocation system of claim 1, wherein The reinforcement learning state space construction unit includes: A warehouse resource state parameter analysis module is configured to locate the collaborative allocation resource state abnormal features and the different warehouse site resource state abnormal features, including warehouse resource state parameter load analysis, warehouse resource state parameter environment analysis and warehouse resource state parameter abnormal positioning. A resource state parameter classification and processing flow module is configured to classify and form warehouse resource state parameter information in the collaborative allocation resource state abnormal features and the different warehouse site resource state abnormal features, obtain data processing flows corresponding to the warehouse resource state parameter information, form warehouse resource state parameter reference benchmarks, warehouse resource state parameter calculation model optimization coefficients and noise types. A dynamic load module is configured to detect device interaction of full time step dynamic loads of different warehouse sites, and statistically analyze device dynamic load sources of different warehouse sites to form state space construction parameters of different warehouse site dynamic loads.

5. The warehouse resource adaptive co-allocation system of claim 1, wherein The resource allocation strategy generation unit includes: A resource allocation model management module is configured to construct and manage global strategy models, local optimization models and multi-objective collaborative models based on deep reinforcement learning. The allocation strategy generation module is configured to determine the allocation strategy parameters of the resource allocation security of different storage locations, systems and management terminals according to the real-time operation data and environmental parameters of the multi-type devices, the global strategy model, the local optimization model and the multi-objective coordination model, and by using a preset state coding mechanism and a strategy generation algorithm. The operation difficulty determination module is configured to determine the operation difficulty of different storage locations according to the resource allocation security determination parameters and the device operation complexity dynamic load source, and to obtain operation difficulty determination parameters. The resource allocation influence factor division module is configured to automatically complete the resource allocation security determination of different storage locations, systems and management terminals according to the allocation strategy parameters and the operation difficulty determination parameters, and to form a resource allocation influence factor record. The allocation process visualization module is configured to visualize the resource allocation process of the system.

6. The warehouse resource adaptive co-allocation system of claim 1, wherein The system further comprises a multi-source information fusion unit configured to integrate the device perception data, the management system instruction data and the external environment data in the storage space to form a fusion information set. The multi-source information fusion unit comprises a time synchronization module configured to perform timestamp alignment processing on different source data; a space calibration module configured to perform spatial position calibration on the device perception data according to the storage location coordinate information; and a conflict resolution module configured to identify conflict information of different data sources and perform fusion processing by using a voting mechanism or a credibility weight, wherein the fusion information set is used as a supplementary input of the format-unified structured data.

7. The warehouse resource adaptive co-allocation system of claim 5, wherein The multi-objective coordination model comprises an inventory balance sub-model, a device load balance sub-model and a path overlap avoidance sub-model; the inventory balance sub-model is configured to make the inventory levels of each storage location tend to a preset mean value by dynamically adjusting the allocation proportion of the storage locations; the device load balance sub-model is configured to make the operation load differences of different devices less than a threshold value by using a task allocation strategy; and the path overlap avoidance sub-model is configured to reduce the cross-overlap probability of the device carrying paths by using a path planning algorithm.

8. The warehouse resource adaptive co-allocation system of claim 5, wherein The resource allocation strategy generation unit further comprises a task queue dynamic adjustment module configured to dynamically adjust the generated task execution sequence according to the real-time resource state and the strategy generation parameters; the task queue dynamic adjustment module comprises a state perception sub-module configured to obtain the inventory state of each storage location, the device operation state and the task completion progress in real time; an influence evaluation sub-module configured to analyze the influence degree of the current task queue state change on the task sequence execution efficiency; and a sequence rearrangement sub-module configured to rearrange the task queue locally or globally according to the influence evaluation result.

9. The warehouse resource adaptive co-allocation system of claim 8, wherein The local rearrangement or global rearrangement operation retains the execution order of high-priority tasks.

Images