Four-way shuttle automatic control system and method

Abstract

The present application discloses a four-way shuttle automatic control system and method, which relates to the field of intelligent control. It uses data analysis and coding technology based on deep learning to embed and code the current power, the distance traveled by the shuttle to the target point, and the number of reversing nodes of the shuttle, and semantically embeds the natural language description of the current demand, so as to intelligently generate the current power weight ratio, the travel distance weight ratio, and the reversing node weight ratio according to the external dynamic collaborative interaction representation between the low-dimensional embedding coding features of the current power, the low-dimensional embedding coding features of the travel distance, the embedding coding features of the number of reversing nodes, and the semantic embedding coding features of the current demand, and obtain the score value. In this way, the dependence on manual experience is reduced, and the problem of waste of resources in other aspects due to excessive emphasis on certain factors is avoided.

Description

Technical Field

[0001] The present application relates to the field of intelligent control, and more specifically, to an automatic control system and method for a four-way shuttle vehicle. Background Art

[0002] The four-way shuttle is an automated vehicle that can move freely in multiple directions and is usually used for cargo handling and storage operations in warehouses and logistics centers. Due to its high efficiency and flexibility, the four-way shuttle is widely used in large warehouses, e-commerce logistics distribution, etc.

[0003] Patent CN115140481A proposes a four-way shuttle dynamic avoidance method and device, which presets a BLOCK mechanism according to the characteristics of the warehouse, divides the area and sets driving rules to avoid deadlock. The BLOCK mechanism is loaded at startup and deadlock monitoring is started. After receiving the task, the system selects the optimal vehicle based on the shuttle's power, driving distance and number of changes, and calculates the minimum cost path to send instructions. It continuously monitors deadlocks during driving and replans the route if deadlocks occur.

[0004] The specific method of selecting the optimal shuttle in this patent is to comprehensively evaluate the current power, driving distance and number of switching nodes, and manually assign a certain weight ratio to each parameter according to specific needs and situations, and then calculate the comprehensive score to determine. However, the manually set weight ratio is easily affected by personal experience and preferences, resulting in different operators setting different weight values. This subjectivity may result in a configuration that is not scientific or reasonable enough and cannot optimally reflect actual needs. In addition, it is difficult to fully consider the interactions between all factors and find the global optimal solution based on manual experience alone. Especially when multiple parameters are involved, the manually set weight ratio may only optimize certain local indicators, while ignoring the maximization of the overall system performance.

[0005] Therefore, an automatic control scheme for a four-way shuttle is desired, which can automatically adjust the weight ratio according to the actual situation to improve the intelligence and adaptability of the system. Summary of the invention

[0006] In order to solve the above technical problems, this application is proposed.

[0007] According to one aspect of the present application, a method for automatically controlling a four-way shuttle is provided, which includes: presetting a BLOCK mechanism of a shuttle according to a warehouse layout; loading the preset BLOCK mechanism and starting deadlock monitoring when the system starts; in response to receiving a task requirement, obtaining a shuttle that currently meets the scheduling conditions and determining an optimal shuttle therefrom; calculating a minimum driving route of the optimal shuttle and sending the minimum driving route to the optimal shuttle; monitoring the operating state of the optimal shuttle in real time through the deadlock monitoring and detecting whether a deadlock occurs with other shuttles; in response to a deadlock between the optimal shuttle and other shuttles, replanning a route for the affected shuttle and sending the newly planned driving route to the affected shuttle; wherein, in response to receiving a task requirement, obtaining a shuttle that currently meets the scheduling conditions and determining an optimal shuttle therefrom includes: Calculate the score of each shuttle vehicle among the shuttle vehicles that currently meet the scheduling conditions, and use the shuttle vehicle with the smallest score as the optimal shuttle vehicle, including: obtaining the current power of the target shuttle vehicle, the driving distance to the target point, and the number of switching nodes; obtaining the natural language description of the current demand; embedding and coding the current power of the target shuttle vehicle, the driving distance to the target point, the number of switching nodes, and the natural language description of the current demand to obtain low-dimensional embedded coding features of current power, low-dimensional embedded coding features of driving distance, embedded coding features of the number of switching nodes, and current demand semantic embedded coding features; respectively perform dynamic memory collaborative interaction and weight calculation on the low-dimensional embedded coding features of current power, the low-dimensional embedded coding features of driving distance, and the embedded coding features of the number of switching nodes with the current demand semantic embedded coding features to obtain the score of the target shuttle vehicle.

[0008] According to another aspect of the present application, a four-way shuttle automatic control system is provided, which includes: a system for executing the four-way shuttle automatic control method as described above, wherein the four-way shuttle automatic control system is used to: calculate the score value of each shuttle among the shuttles that currently meet the scheduling conditions, and use the shuttle with the smallest score as the optimal shuttle. The four-way shuttle automatic control system also includes: a shuttle working parameter acquisition module for acquiring the current power of the target shuttle, the travel distance to the target point, and the number of switching nodes; a current demand acquisition module for acquiring a natural language description of the current demand; a data embedding module A block is used to embed the current power of the target shuttle, the driving distance to the target point, the number of switching nodes and the natural language description of the current demand to obtain low-dimensional embedded coding features of the current power, low-dimensional embedded coding features of the driving distance, embedded coding features of the number of switching nodes and current demand semantic embedded coding features; a feature collaborative interaction module is used to perform dynamic memory collaborative interaction and weight calculation on the low-dimensional embedded coding features of the current power, the low-dimensional embedded coding features of the driving distance and the embedded coding features of the number of switching nodes with the current demand semantic embedded coding features to obtain the score value of the target shuttle.

[0009] Compared with the prior art, the present application provides a four-way shuttle automatic control system and method, which obtains the current power of the target shuttle, the driving distance to the target point, and the number of reversing nodes, and obtains the natural language description of the current demand, and uses data analysis and coding technology based on deep learning to embed the current power, the driving distance of the shuttle to the target point, and the number of reversing nodes of the shuttle, and at the same time, the natural language description of the current demand is semantically embedded, so as to intelligently generate the current power weight ratio, driving distance weight ratio and reversing node weight ratio according to the external dynamic collaborative interaction representation between the low-dimensional embedded coding features of the current power, the low-dimensional embedded coding features of the driving distance, the embedded coding features of the number of reversing nodes, and the current demand semantic embedded coding features, and obtain the score value. In this way, the dependence on manual experience is reduced, and at the same time, a more intelligent weight ratio is generated to better balance factors such as power, driving distance and number of reversing times, so as to avoid the problem of waste of resources in other aspects due to excessive emphasis on certain factors. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0011] Figure 1 The present invention is a flowchart for calculating the score value of each shuttle vehicle among the shuttle vehicles that currently meet the scheduling conditions in the automatic control method of the four-way shuttle vehicles according to the embodiment of the present application.

[0012] Figure 2 Flow chart of step S140 in the automatic control method of a four-way shuttle according to an embodiment of the present application.

[0013] Figure 3 Flow chart of step S210 in the automatic control method of a four-way shuttle according to an embodiment of the present application.

[0014] Figure 4 Flow chart of step S320 in the automatic control method of a four-way shuttle according to an embodiment of the present application.

[0015] Figure 5 4 is a block diagram of a four-way shuttle automatic control system according to an embodiment of the present application. DETAILED DESCRIPTION

[0016] The embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein, but rather these embodiments are provided for a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0017] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0018] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0019] Therefore, in response to the problems in the above-mentioned background technology and based on patent CN115140481A, the present application proposes a four-way shuttle automatic control method, including: presetting the BLOCK mechanism of the shuttle according to the warehouse layout; loading the preset BLOCK mechanism and starting deadlock monitoring when the system starts; in response to receiving the task requirements, obtaining the shuttles that currently meet the scheduling conditions and determining the optimal shuttle from them; calculating the minimum driving route of the optimal shuttle, and sending the minimum driving route instruction to the optimal shuttle; monitoring the operating status of the optimal shuttle in real time through the deadlock monitoring, and detecting whether deadlock occurs with other shuttles; in response to the optimal shuttle deadlocking with other shuttles, re-planning the route for the affected shuttle, and sending the newly planned driving route instruction to the affected shuttle.

[0020] It should be understood that in order to optimize the shuttle's movement path and reduce the occurrence of conflicts and deadlocks, the warehouse space is logically divided into multiple BLOCK areas and corresponding driving rules are set. This BLOCK mechanism is a rule for managing the shuttle's driving area, ensuring that only one car is allowed to drive in the same area or multiple cars are allowed to drive in the same direction to avoid multiple shuttles from deadlocking due to competition for resources. When the system is initialized, the preset BLOCK mechanism is correctly applied, and a dedicated main process is activated to monitor the system status in real time to prevent and solve possible deadlock problems. When the system starts, it loads the preset BLOCK mechanism and starts a continuously running deadlock monitoring main process, which is responsible for checking whether a deadlock occurs. After receiving the task requirements for entering and leaving the warehouse, the system first selects the shuttles that currently meet the scheduling conditions, that is, those with sufficient power and suitable locations. Next, the system calculates a score for each candidate shuttle based on a series of parameters (such as current power, driving distance to the target point, and number of switching nodes), and determines the most suitable shuttle for the job. For the selected optimal shuttle, the system calculates the route with the lowest path cost from the current location to the target location. This process involves using algorithms to find the best path while avoiding obstacles and other inaccessible areas. Once the calculation is complete, the optimal route will be sent as an instruction to the optimal shuttle to guide it to follow the planned route. During the entire mission execution process, the deadlock monitoring main process is always in operation, monitoring in real time any vehicle that may be stuck in a deadlock state due to conflicts with other shuttles. If a deadlock is detected, the system will respond immediately, re-plan a new route for the affected shuttle that will not cause a deadlock, and send the new route instruction to the shuttle to ensure that it can continue and successfully complete the mission.

[0021] Specifically, in response to receiving a task requirement, obtaining the shuttles that currently meet the scheduling conditions and determining the optimal shuttle from them, including: calculating the score value of each shuttle among the shuttles that currently meet the scheduling conditions, and taking the shuttle with the smallest score value as the optimal shuttle. Figure 1 FIG. 1 is a flowchart of calculating the score value of each shuttle vehicle among the shuttle vehicles that currently meet the scheduling conditions in the automatic control method of the four-way shuttle vehicle according to the embodiment of the present application. Figure 1 As shown, the score value of each shuttle vehicle among the shuttle vehicles that currently meet the scheduling conditions is calculated, and the shuttle vehicle with the smallest score value is used as the optimal shuttle vehicle, including: S110, obtaining the current power of the target shuttle vehicle, the driving distance to the target point, and the number of switching nodes; S120, obtaining the natural language description of the current demand; S130, respectively embedding and encoding the current power of the target shuttle vehicle, the driving distance to the target point, the number of switching nodes, and the natural language description of the current demand to obtain the current power low-dimensional embedded coding feature, the driving distance low-dimensional embedded coding feature, the switching node number embedded coding feature, and the current demand semantic embedded coding feature; S140, respectively, dynamically memorize the low-dimensional embedded coding feature of the current power, the driving distance low-dimensional embedded coding feature, and the switching node number embedded coding feature and the current demand semantic embedded coding feature to perform collaborative interaction and weight calculation to obtain the score value of the target shuttle vehicle.

[0022] Specifically, in step S110, the current power of the target shuttle, the driving distance to the target point, and the number of switching nodes are obtained. It should be understood that the current power refers to the remaining power of the battery of the target shuttle, which directly affects whether the vehicle can successfully complete the task, especially for long-term or long-distance tasks. The driving distance to the target point refers to the estimated driving distance from the current position of the shuttle to the target position specified by the task. A shorter driving distance means less energy consumption and faster task completion time. The number of switching nodes refers to the number of times the direction needs to be changed in the process of going to the target point. Fewer switching times usually means a more direct path, reducing additional time and energy consumption, and reducing complexity. Therefore, in order to achieve efficient and reliable task allocation and path planning, and ensure that the operation of the shuttle system in the warehouse is both safe and efficient, in the technical solution of the present application, the current power of the target shuttle, the driving distance to the target point, and the number of switching nodes are obtained. That is, by evaluating the current power level, it can be ensured that the selected shuttle vehicle has enough energy to complete the task and avoid stopping midway due to insufficient power; choosing the path with the shortest driving distance and the least number of changes can reduce energy consumption, improve energy utilization, and reduce operating costs.

[0023] Specifically, after determining the set of shuttles that meet the scheduling conditions, the next task is to collect three key parameters for each candidate vehicle: the current power, the distance to the target point, and the number of switching nodes. The current power refers to the remaining energy level of the shuttle battery, which directly affects whether the vehicle can successfully complete the task, especially for long-term or long-distance tasks. In order to ensure that the selected shuttle has enough energy to perform the task until the end, the system must accurately know the real-time power status of each vehicle. Usually, this information is read directly by sensors installed on the shuttle, which can continuously monitor the status of the battery and report the latest data to the central control system.

[0024] Next, for each candidate shuttle, the system needs to calculate the estimated driving distance from its current location to the target location specified by the task. This process involves understanding the physical layout within the warehouse, including information such as shelf locations and aisle widths, in order to find a safe and economical route for each vehicle. Shorter driving distance means less energy consumption and faster task completion time, so this is a very important consideration when selecting the optimal vehicle. The method for calculating driving distance may rely on a pre-built map model or a dynamically generated map view, depending on the system's architectural design and technical implementation.

[0025] In addition, the system also needs to consider the number of switching points required from the starting point to the end point, that is, the number of switching nodes. Fewer switching times usually means a more direct path, reducing additional time and energy consumption, and reducing complexity. The number of switching nodes can be obtained by analyzing pre-defined driving rules and path planning algorithms, which are designed to optimize the shuttle's trajectory, reduce unnecessary turning actions, and improve operating efficiency.

[0026] Specifically, in step S120, a natural language description of the current demand is obtained. Accordingly, the natural language description of the current demand refers to specific task requirements or instructions for warehouse operations expressed in the language used by humans in daily life (such as Chinese, English, etc.). These descriptions usually include information such as the type of task (such as warehousing, outbound), target location, priority, special requirements (such as completion at a specific time, type of goods that require special handling), etc. For example: "Please take out the product numbered 12345 located in shelf A1 area and transport it to the shipping area." Or "Before 5 pm, store the newly arrived goods in the designated refrigerated area. Therefore, in order to ensure that the system accurately understands the uniqueness and complexity of each task and better selects the optimal shuttle vehicle for execution, in the technical solution of the present application, a natural language description of the current demand is obtained, so that a detailed task background and specific requirements can be provided to avoid operational errors caused by misunderstanding of the task.

[0027] When there is a new task requirement, workers usually enter the task instructions in some way. This may be filling out a form on a graphical user interface (GUI), entering via voice commands, or typing text instructions directly on a terminal device. Regardless of the method used, the system first needs to receive these instructions and save them as raw data.

[0028] Next, the system focuses on the process of actually obtaining natural language descriptions. At this stage, the core task is to extract information from the received instructions that can clearly express the task requirements. Since the task requirements in the warehousing and logistics environment often involve multiple aspects, such as cargo location, target location, priority, special requirements, etc., it is necessary to ensure that all relevant information is fully recorded. For example, a typical task instruction may be: "Please take out the product with product number 12345 located in shelf A1 area and transport it to the shipping area." This instruction contains key information of the starting location (shelf A1 area), cargo identification (product number 12345), and destination (shipping area).

[0029] In order to effectively obtain this information, the system has designed a special input interface to guide users to provide a complete task description. For text-based input, this interface may be a structured form that requires users to fill in each item explicitly; for voice input, it relies on high-performance speech recognition technology to convert verbal instructions into text in real time. In addition, considering that there may be a variety of different expressions in actual application scenarios, the system also has a built-in set of flexible parsing rules to deal with task instructions in different formats. This set of rules can not only handle standard language expressions, but also adapt to some informal or colloquial descriptions to ensure that various situations are covered as comprehensively as possible.

[0030] Once the initial reception and organization of the task instructions is completed, the next step is to ensure that this information is stored correctly for subsequent use. The system will save the preprocessed natural language description in a temporary database or memory cache for subsequent module calls. In order to ensure the security and reliability of the data, timestamps and other metadata will be added during the storage process to track the source and status changes of each instruction. At the same time, in order to support multi-task concurrent processing, the system must also have an efficient concurrency control mechanism to ensure that multiple instructions from different users at the same time will not conflict or be lost.

[0031] Specifically, in step S130, the current power of the target shuttle, the driving distance to the target point, the number of switching nodes, and the natural language description of the current demand are respectively embedded and encoded to obtain low-dimensional embedded coding features of the current power, low-dimensional embedded coding features of the driving distance, embedded coding features of the number of switching nodes, and semantic embedded coding features of the current demand. Specifically, in an embodiment of the present application, the current power of the target shuttle, the driving distance to the target point, the number of switching nodes, and the natural language description of the current demand are respectively embedded and encoded to obtain low-dimensional embedded coding features of the current power, low-dimensional embedded coding features of the driving distance, embedded coding features of the number of switching nodes, and semantic embedded coding features of the current demand, including: the current power of the target shuttle, the driving distance to the target point, and the number of switching nodes are respectively embedded and encoded to obtain a low-dimensional embedded coding vector of the current power as the low-dimensional embedded coding feature of the current power, a low-dimensional embedded coding vector of the driving distance as the low-dimensional embedded coding feature of the driving distance, and an embedded coding vector of the number of switching nodes as the embedded coding feature of the number of switching nodes; the natural language description of the current demand is semantically embedded and encoded to obtain a current demand semantic embedded coding vector as the current demand semantic embedded coding feature.

[0032] Specifically, the current power of the target shuttle, the driving distance to the target point, and the number of commutation nodes are embedded and coded respectively to obtain a low-dimensional embedded coding vector of the current power as the low-dimensional embedded coding feature of the current power, a low-dimensional embedded coding vector of the driving distance as the low-dimensional embedded coding feature of the driving distance, and an embedded coding vector of the number of commutation nodes as the embedded coding feature of the number of commutation nodes. It should be understood that considering that the original data (such as power value, distance, and number of nodes) are usually high-dimensional. Therefore, in order to map these features to a lower-dimensional space to significantly reduce the amount of data to be processed subsequently and capture the semantic information in each data, in the technical solution of the present application, the current power of the target shuttle, the driving distance to the target point, and the number of commutation nodes are embedded and coded respectively to convert the original physical quantity into a more abstract but more information-rich low-dimensional feature, and obtain a low-dimensional embedded coding vector of the current power, a low-dimensional embedded coding vector of the driving distance, and an embedded coding vector of the number of commutation nodes. In particular, in a specific embodiment of the present application, after the label information of the current power, the driving distance to the target point and the number of switching nodes (the meaning labels of the current power, the driving distance to the target point and the number of switching nodes) is input into the word embedding layer, the corresponding numerical data is added to the tail of each label embedding feature to obtain a low-dimensional embedding coding vector of the previous power, a low-dimensional embedding coding vector of the driving distance, and an embedding coding vector of the number of switching nodes.

[0033] More specifically, in an embodiment of the present application, the natural language description of the current demand is semantically embedded and encoded to obtain the current demand semantic embedding encoding vector, including: the natural language description of the current demand is processed by word segmentation and then input into a semantic embedding encoder containing a word embedding layer to obtain the current demand semantic embedding encoding vector. Accordingly, considering that the natural language description of the current demand contains specific demand information for inbound and outbound tasks such as moving goods from one location to another, in order to capture and extract key demand semantic information to better understand and analyze the specific semantic content of the demand, in the technical solution of the present application, the natural language description of the current demand is semantically embedded and encoded to obtain the current demand semantic embedding encoding vector. In particular, in a specific embodiment of the present application, the natural language description of the current demand is processed by word segmentation and then input into a semantic embedding encoder containing a word embedding layer to obtain the current demand semantic embedding encoding vector. In this way, the semantic information in the current demand description can be extracted more accurately to better understand the specific contextual meaning of the user's demand, so as to make the optimal choice of a more intelligent shuttle vehicle on this basis.

[0034] Specifically, in step S140, the current power low-dimensional embedded coding feature, the driving distance low-dimensional embedded coding feature, and the switching node number embedded coding feature are dynamically memorized and interacted with the current demand semantic embedded coding feature and weighted to obtain the score value of the target shuttle vehicle.

[0035] Figure 2 4 is a flowchart of step S140 in the automatic control method of the four-way shuttle according to an embodiment of the present application. Specifically, in the embodiment of the present application, Figure 2As shown, the S140 includes: S210, performing the attention interaction based on external dynamic memory on the current power low-dimensional embedded coding vector and the current demand semantic embedded coding vector to obtain the current power-current demand semantic collaborative interaction coding vector; S220, performing the attention interaction based on external dynamic memory on the driving distance low-dimensional embedded coding vector and the current demand semantic embedded coding vector to obtain the driving distance-current demand semantic collaborative interaction coding vector; S230, performing the attention interaction based on external dynamic memory on the switching node number embedded coding vector and the current demand semantic embedded coding vector to obtain the switching node-current demand semantic collaborative interaction coding vector; S240, obtaining the current power weight ratio, driving distance weight ratio and switching node weight ratio according to the current power-current demand semantic collaborative interaction coding vector, the driving distance-current demand semantic collaborative interaction coding vector and the switching node-current demand semantic collaborative interaction coding vector respectively; S250, obtaining the score value of the target shuttle according to the current power weight ratio, the driving distance weight ratio and the switching node weight ratio.

[0036] In step S210, the current power low-dimensional embedded coding vector and the current demand semantic embedded coding vector are subjected to the attention interaction based on external dynamic memory to obtain the current power-current demand semantic collaborative interaction coding vector. In step S220, the driving distance low-dimensional embedded coding vector and the current demand semantic embedded coding vector are subjected to the attention interaction based on external dynamic memory to obtain the driving distance-current demand semantic collaborative interaction coding vector. In step S230, the switching node number embedded coding vector and the current demand semantic embedded coding vector are subjected to the attention interaction based on external dynamic memory to obtain the switching node-current demand semantic collaborative interaction coding vector.

[0037] Accordingly, considering the mutual interaction between the current power low-dimensional embedding coding vector and the current demand semantic embedding coding vector, for example, the shuttle car has enough power to successfully complete the task under the current demand and will not stop working due to insufficient power in the middle of the task. Therefore, in order to more accurately clarify the complex semantic relationship between the current power and the current demand, in the technical solution of the present application, the current power low-dimensional embedding coding vector and the current demand semantic embedding coding vector are processed by an attention interaction method based on external dynamic memory to obtain the current power-current demand semantic collaborative interaction coding vector. In particular, this method not only retains the unique information of the original features, but also captures the complex relationship between them, thereby improving the overall perception ability of the system, and combines external knowledge (such as environmental conditions, historical data, warehouse layout rules, etc.) to help the system more accurately evaluate the specific situation of each task. For example, while considering the power limit, understand the user's requirements for task priority to ensure that the most appropriate shuttle car is selected to perform the task.

[0038] Similarly, considering that there is a mutual collaborative interaction relationship between the driving distance low-dimensional embedding coding vector, the reversing node number embedding coding vector and the current demand semantics embedding coding vector. Therefore, in order to understand the collaborative relationship between the two more carefully and accurately and provide data support for the subsequent optimal shuttle, the driving distance low-dimensional embedding coding vector and the current demand semantics embedding coding vector are subjected to the external dynamic memory-based attention interaction to obtain the driving distance-current demand semantics collaborative interaction coding vector; the reversing node number embedding coding vector and the current demand semantics embedding coding vector are subjected to the external dynamic memory-based attention interaction to obtain the reversing node-current demand semantics collaborative interaction coding vector as the reversing node-current demand semantics collaborative interaction coding feature.

[0039] In particular, the current power low-dimensional embedded coding vector is taken as an example for specific explanation.

[0040] Figure 3 Flowchart of step S210 in the automatic control method of the four-way shuttle according to an embodiment of the present application. Specifically, in the embodiment of the present application, Figure 3As shown, in step S210, the current power low-dimensional embedded coding vector and the current demand semantic embedded coding vector are subjected to attention interaction based on external dynamic memory to obtain the current power-current demand semantic collaborative interaction coding vector, including: S310, calculating the current power-current demand external knowledge fine-grained optimized feature interaction matrix between the current power low-dimensional embedded coding vector and the current demand semantic embedded coding vector; S320, according to the current power-current demand external knowledge fine-grained optimized feature interaction matrix, performing fine-grained modulation response interaction on the current power low-dimensional embedded coding vector and the current demand semantic embedded coding vector to obtain the current power-current demand semantic collaborative interaction coding vector.

[0041] More specifically, in the embodiment of the present application, in step S310, the current power-current demand external knowledge fine-grained optimization feature interaction matrix between the current power low-dimensional embedded coding vector and the current demand semantic embedded coding vector is calculated, including: performing fine-grained interaction on the current power low-dimensional embedded coding vector and the current demand semantic embedded coding vector to obtain the current power-current demand fine-grained feature interaction matrix, and the process can be expressed by the following formula: ;in, is the current power low-dimensional embedding coding vector, is the semantic embedding encoding vector of the current requirement, for The transposed vector of is matrix multiplication, It is the current power-current demand fine-grained feature interaction matrix.

[0042] According to the learnable memory parameter matrix, the current power-current demand fine-grained feature interaction matrix is ​​optimized based on external knowledge to obtain the current power-current demand external knowledge fine-grained optimized feature interaction matrix. The process can be expressed by the following formula: ;in, is the current power-current demand fine-grained feature interaction matrix, and They represent the learnable memory parameter matrix, for The transposed matrix of represents the normalization function, It is the fine-grained optimization feature interaction matrix of current power-current demand external knowledge.

[0043] It should be understandable that the current power low-dimensional embedding coding vector and the current demand semantic embedding coding vector are fine-grainedly interacted to deeply understand and capture the complex correlation information between the two, and obtain the current power-current demand fine-grained feature interaction matrix.

[0044] Accordingly, according to the learnable memory parameter matrix, the current power-current demand fine-grained feature interaction matrix is ​​optimized based on external knowledge attention to embed the features of external knowledge into the interaction matrix, and the current power-current demand external knowledge fine-grained optimized feature interaction matrix is ​​obtained. That is, by encoding external knowledge (such as logistics rules, warehouse layout, historical task data, etc.) into the memory parameter matrix, the system can not only rely on the direct correlation between the current power and task requirements when evaluating and selecting, but also combine these external factors to make a more comprehensive judgment.

[0045] Figure 4 is a flow chart of step S320 in the automatic control method of the four-way shuttle according to an embodiment of the present application. More specifically, in the embodiment of the present application, Figure 4 As shown, in step S320, according to the current power-current demand external knowledge fine-grained optimization feature interaction matrix, the current power low-dimensional embedded coding vector and the current demand semantic embedded coding vector are fine-grained modulated and responded to interact to obtain the current power-current demand semantic collaborative interaction coding vector, including: S410, the current power low-dimensional embedded coding vector is linearly transformed to obtain the current power query feature vector and the current power value feature vector, and the current power-current demand external knowledge fine-grained optimization feature interaction matrix is ​​used as the key matrix, and the current power query feature vector, the current power value feature vector and the key matrix are optimized and modulated based on the Transformer model to obtain the optimized current power low-dimensional embedded coding vector. The process can be expressed by the following formula: ;in, is the current power low-dimensional embedding coding vector, is the fine-grained optimization feature interaction matrix of current power and current demand external knowledge, and They are the current power query embedding matrix and the current power first bias vector respectively. is the current power query feature vector, and They are the current power value embedding matrix and the current power second bias vector respectively. is the current power value feature vector, for The transposed matrix of yes Length, yes function, is the optimized low-dimensional embedding coding vector of the current power.

[0046] S420, linearly transform the current demand semantic embedding coding vector to obtain a current demand query feature vector and a current demand value feature vector, and use the current power-current demand external knowledge fine-grained optimization feature interaction matrix as a key matrix, and perform the Transformer model-based optimization modulation on the current demand query feature vector, the current demand value feature vector and the key matrix to obtain an optimized current demand semantic embedding coding vector. This process can be expressed by the following formula: ;in, is the semantic embedding encoding vector of the current requirement, is the fine-grained optimization feature interaction matrix of current power and current demand external knowledge, for The transposed matrix of and They are respectively the current demand query embedding matrix and the current demand first bias vector, is the current demand query feature vector, and are the current demand value embedding matrix and the current demand second bias vector respectively, is the current demand value feature vector, yes Length, is the optimized current demand semantic embedding encoding vector.

[0047] S430, calculating the position response between the optimized current power low-dimensional embedding coding vector and the optimized current demand semantic embedding coding vector to obtain the current power-current demand semantic collaborative interaction coding vector. This process can be expressed by the following formula: ;in, is the optimized low-dimensional embedding coding vector of the current power, is the optimized current demand semantic embedding encoding vector, It is the current power-current demand semantic collaborative interaction encoding vector.

[0048] It should be understood that the current power low-dimensional embedding coding vector is linearly transformed to obtain the current power query feature vector and the current power value feature vector, and the current power-current demand external knowledge fine-grained optimized feature interaction matrix is ​​used as the key matrix, and optimized modulation based on the Transformer model is performed on it. In this way, using the self-attention mechanism in the Transformer model, the current power feature and the key matrix can be precisely interacted in detail to achieve deep information exchange and fusion of the current power feature with internal and external knowledge. This process enables the power feature to effectively absorb the beneficial information of external knowledge, thereby obtaining an optimized low-dimensional embedding coding vector of the current power.

[0049] Similarly, a linear transformation is performed on the current demand semantic embedding coding vector to obtain the current demand query feature vector and the current demand value feature vector, and the same processing flow is performed on them as the current demand query feature vector and the current demand value feature vector to utilize the features of the current demand obtained from external knowledge to enhance the representation capability of the current demand semantic features, and obtain the optimized current demand semantic embedding coding vector.

[0050] Finally, the position-based response between the optimized current power low-dimensional embedding coding vector and the optimized current demand semantic embedding coding vector is calculated to comprehensively evaluate the relationship between the shuttle vehicle's power status and task requirements, thereby providing a highly integrated and information-rich current power-current demand semantic collaborative interaction coding vector for the final task selection.

[0051] In step S240, the current power weight ratio, the driving distance weight ratio and the switching node weight ratio are obtained according to the current power-current demand semantic collaborative interaction coding vector, the driving distance-current demand semantic collaborative interaction coding vector and the switching node-current demand semantic collaborative interaction coding vector. Specifically, in the embodiment of the present application, the current power weight ratio, the driving distance weight ratio and the switching node weight ratio are obtained according to the current power-current demand semantic collaborative interaction coding vector, the driving distance-current demand semantic collaborative interaction coding vector and the switching node-current demand semantic collaborative interaction coding vector, respectively, for: inputting the current power-current demand semantic collaborative interaction coding vector, the driving distance-current demand semantic collaborative interaction coding vector and the switching node-current demand semantic collaborative interaction coding vector into the weight specifier based on the feedforward neural network model to obtain the current power weight ratio, the driving distance weight ratio and the switching node weight ratio. That is to say, by assigning appropriate weights to each feature (power, driving distance, switching node), the system can calculate the weight ratio of each shuttle according to the requirements of the specific task and environmental conditions, thereby dynamically adjusting the resource (shuttle) allocation strategy. This helps achieve more efficient resource utilization, ensuring that tasks are completed smoothly while maximizing energy and other resource savings. It should be understood that the feedforward neural network model is the most basic neural network architecture, characterized by a unidirectional flow of information from the input layer to the output layer, passing through one or more hidden layers in the middle, without feedback connections. Each neuron (or node) in a layer receives input from the previous layer and generates output through an activation function, which is passed to the next layer.

[0052] In particular, the current power-current demand semantic collaborative interaction coding vector is taken as an example for specific explanation: the specific process of inputting the current power-current demand semantic collaborative interaction coding vector into the weight specifier based on the feedforward neural network model to obtain the current power weight ratio is as follows: First, the current power-current demand semantic collaborative interaction coding vector is used as the input feature of the feedforward neural network model. These coding vectors not only contain the physical state information of the shuttle vehicle (such as power level), but also incorporate the semantic understanding of the task requirements (such as specific requirements and priorities for transporting goods).

[0053] Next, build a feedforward neural network (FNN) architecture. FNN consists of an input layer, one or more hidden layers, and an output layer. The number of nodes in the input layer matches the dimension of the collaborative interaction encoding vector to ensure that each element can be correctly passed to the next layer. The design of the hidden layer is critical, and the appropriate number of layers and the number of nodes per layer are usually selected to balance model complexity and computational efficiency. The neurons in each layer are connected to the next layer through linear transformations and nonlinear activation functions (such as ReLU, Sigmoid, Tanh, etc.) to introduce nonlinear characteristics, enabling the model to capture complex patterns and relationships. The output layer has only one node and uses an appropriate activation function (such as Sigmoid) to generate a value between 0 and 1, representing the weight ratio of the current power.

[0054] The training phase is the key to ensure that FNN can accurately generate weight ratios. To this end, it is necessary to collect historical task allocation data as labels. These labels can be manually set ideal weight ratios or empirical values ​​obtained by other methods. Define a loss function to measure the difference between the predicted weight ratio and the true label. Common loss functions include mean square error (MSE), cross entropy, etc. Use gradient descent or other optimization algorithms (such as Adam, RMSprop) to minimize the loss function and adjust the weights and bias parameters in the network. To avoid overfitting, you can also add L1 / L2 regularization terms or use Dropout technology. During batch training, data is input into the network in batches, and weights and bias parameters are updated each iteration. In addition, the validation set is used to evaluate the model performance, and the network structure or hyperparameters are adjusted based on the results to ensure that the model has good generalization ability.

[0055] The inference phase occurs in actual applications. The system calculates the current power-current demand semantic collaborative interaction encoding vector in real time and passes it to the trained FNN as a new input. During the forward propagation process, the input vector starts from the input layer, passes through the linear transformation and activation function processing of each hidden layer in turn, and finally reaches the output layer to generate the current power weight ratio. This weight ratio reflects the importance of the current power state in the context of a specific task. In order to ensure that the weight ratios of different factors (such as driving distance and number of commutation nodes) can work together, it may be necessary to normalize all weight ratios to ensure that they add up to 1.

[0056] Finally, in the application phase, the generated weight ratio is used for comprehensive scoring, combined with the weight ratios of other relevant factors, to calculate the score of each candidate shuttle. In this way, the feedforward neural network not only realizes intelligent weight generation, but also ensures that the system can dynamically adjust the importance of each parameter according to the actual situation, so as to make more scientific and reasonable decisions.

[0057] In step S250, the score value of the target shuttle vehicle is obtained according to the current power weight ratio, the travel distance weight ratio and the switching node weight ratio.

[0058] In particular, the calculation process of the score value is as follows: ;in, is the current power weight ratio, is the current power value, is the driving distance weight ratio, is the driving distance to the target point, is the commutation node weight ratio, is the number of commutation nodes, is the score value. In particular, This section is used to reflect the importance of battery level. The inverse form is used to ensure that the higher the battery level, the lower the score (i.e. better) for this item, because a high battery level means that the vehicle is more likely to successfully complete the task. This part represents the weighted score of the driving distance, reflecting the path length required to complete the task. The shorter the driving distance, the lower the score, indicating that the vehicle performs better in terms of energy consumption and time efficiency. This part represents the weighted score of the number of switching nodes, reflecting the complexity of the path. The fewer the number of switching nodes, the lower the score, indicating that the path is more direct and easier to execute. It is a total score that takes into account battery power, driving distance and number of reversals. The system will calculate this score for all shuttles that meet the scheduling conditions and select the one with the lowest score as the optimal shuttle to perform the task.

[0059] It should be understandable that, here, considering that the current power low-dimensional embedding coding vector and the current demand semantic embedding coding vector represent the low-dimensional embedding coding features of the current power and the semantic embedding coding features of the current demand respectively, when performing fine-grained feature interaction based on external knowledge modulation, due to the difference in external knowledge modulation caused by the difference in source data modal patterns, the current power-current demand semantic collaborative interaction coding vector will have a probabilistic convergence and divergence based on different feature interactivity, thereby affecting the accuracy of the current power weight ratio obtained by inputting the weight specifier based on the feedforward neural network model.

[0060] Preferably, inputting the current power-current demand semantic collaborative interaction coding vector into a weight designator based on a feedforward neural network model to obtain the current power weight ratio includes: determining the median eigenvalue, the maximum eigenvalue and the minimum eigenvalue in the current power-current demand semantic collaborative interaction coding vector, and dividing the median eigenvalue by the difference between the maximum eigenvalue and the minimum eigenvalue to obtain the current power-current demand semantic collaborative interaction coding distribution probability value, that is, ;in, , and are respectively the median eigenvalue, the maximum eigenvalue and the minimum eigenvalue in the current power-current demand semantic collaborative interaction encoding vector, It is the distribution probability value of the current power-current demand semantic collaborative interaction encoding.

[0061] The current power-current demand semantic collaborative interaction coding vector is normalized to the maximum value to obtain the current power-current demand semantic collaborative interaction coding probability feature vector .

[0062] The power function of each eigenvalue of the current power-current demand semantic collaborative interaction coding probability feature vector is calculated by subtracting the current power-current demand semantic collaborative interaction coding distribution probability value from one to obtain the first current power-current demand semantic collaborative interaction coding convergence feature vector. ;in, is the current power-current demand semantic collaborative interaction encoding probability feature vector, The power function of each eigenvalue of the current power-current demand semantic collaborative interaction coding probability feature vector is calculated by subtracting the current power-current demand semantic collaborative interaction coding distribution probability value from one as the exponent.

[0063] The power function of each eigenvalue of the difference eigenvalue of the unit eigenvector and the current power-current demand semantic collaborative interaction coding probability eigenvector is calculated with the current power-current demand semantic collaborative interaction coding distribution probability value as an exponent to obtain the second current power-current demand semantic collaborative interaction coding convergence eigenvalue. ;in, is the unit eigenvector, is a power function for calculating each eigenvalue in the spread eigenvector with the current power-current demand semantic collaborative interaction encoding distribution probability value as an exponent, It is subtracted by position point. It is the second current power-current demand semantic collaborative interaction encoding convergence feature vector.

[0064] The current power-current demand semantic collaborative interaction coding probability feature vector is multiplied by the difference between one and the current power-current demand semantic collaborative interaction coding distribution probability value to obtain a first current power-current demand semantic collaborative interaction coding limited feature vector. ;in, is the first current power-current demand semantic collaborative interaction encoding limited feature vector, It is the point product by position.

[0065] The point difference feature vector is multiplied by the current power-current demand semantic collaborative interaction coding distribution probability value to obtain a second current power-current demand semantic collaborative interaction coding limited feature vector ;in, It is the second current power-current demand semantic collaborative interaction encoding limited feature vector.

[0066] The first current power-current demand semantic collaborative interaction coding convergence feature vector is multiplied by the second current power-current demand semantic collaborative interaction coding convergence feature vector, and then the first current power-current demand semantic collaborative interaction coding limit feature vector and the second current power-current demand semantic collaborative interaction coding limit feature vector are multiplied to obtain the optimized current power-current demand semantic collaborative interaction coding vector ;in, is the optimized current power-current demand semantic collaborative interaction encoding vector, To add by location point.

[0067] The optimized current power-current demand semantic collaborative interaction encoding vector is input into a weight designator based on a feedforward neural network model to obtain a current power weight ratio.

[0068] Based on this, by using the cross entropy form power series of the distribution analysis probability of the current power-current demand semantic collaborative interaction coding vector as the probability distribution convergence limit, the current power-current demand semantic collaborative interaction coding vector is approximated by the quasi-probability convergence based on the combination of the feature set distribution and the probability density distribution of the current power-current demand semantic collaborative interaction coding vector, so as to avoid the conservative convergence control of the fatal potential divergence of the current power-current demand semantic collaborative interaction coding vector, further use the probability distribution cross entropy of the current power-current demand semantic collaborative interaction coding vector as the objective function to guide the limited recovery strategy, realize the common agility of the convergence of the feature set of the probability convergence and divergence of the current power-current demand semantic collaborative interaction coding vector to the probability density distribution space, and improve the accuracy of the current power weight ratio obtained by the weight specifier based on the feedforward neural network model input of the current power-current demand semantic collaborative interaction coding vector. In particular, the optimization process of the driving distance weight ratio and the commutation node weight ratio is consistent with the optimization process of the current power weight ratio.

[0069] In summary, the automatic control method of the four-way shuttle vehicle based on the embodiment of the present application is explained, which obtains the current power of the target shuttle vehicle, the driving distance to the target point, and the number of switching nodes, and obtains the natural language description of the current demand, and uses data analysis and coding technology based on deep learning to embed and encode the current power, the driving distance of the shuttle vehicle to the target point, and the number of switching nodes of the shuttle vehicle, and at the same time, the natural language description of the current demand is semantically embedded, so as to intelligently generate the current power weight ratio, driving distance weight ratio and switching node weight ratio according to the external dynamic collaborative interaction representation between the low-dimensional embedded coding features of the current power, the low-dimensional embedded coding features of the driving distance, the embedded coding features of the number of switching nodes, and the current demand semantic embedded coding features, and obtain the score value. In this way, the dependence on manual experience is reduced, and at the same time, a more intelligent weight ratio is generated to better balance factors such as power, driving distance and number of switching, so as to avoid the problem of waste of resources in other aspects due to excessive emphasis on certain factors.

[0070] In one embodiment of the present application, the four-way shuttle adopts the independently developed S-curve acceleration and deceleration speed and position planning algorithm, as well as the speed loop and position loop double closed-loop PID motion control algorithm. The use of the independently developed S-curve planning algorithm and double closed-loop PID motion control algorithm can get rid of the dependence on expensive logistics-specific inverters, and the use of standard general-purpose inverters can achieve the same control effect. At the same time, a shuttle with a hopper is used for granular material handling and automatic feeding; a flexible cloth cover is set on the receiving barrel to suppress the generation and flow of dust.

[0071] Figure 5FIG. 1 is a block diagram of a four-way shuttle automatic control system according to an embodiment of the present application. Figure 5 As shown, the four-way shuttle automatic control system 100 according to the embodiment of the present application is used to execute the above-mentioned four-way shuttle automatic control method, wherein the four-way shuttle automatic control system is used to: calculate the score value of each shuttle car among the shuttle cars that currently meet the scheduling conditions, and use the shuttle car with the smallest score value as the optimal shuttle car, and the four-way shuttle automatic control system also includes: a shuttle car working parameter acquisition module 110, which is used to obtain the current power of the target shuttle car, the driving distance to the target point, and the number of switching nodes; a current demand acquisition module 120, which is used to obtain a natural language description of the current demand; a data embedding module 130 , which is used to embed the current power of the target shuttle, the driving distance to the target point, the number of switching nodes and the natural language description of the current demand to obtain low-dimensional embedded coding features of the current power, low-dimensional embedded coding features of the driving distance, embedded coding features of the number of switching nodes and current demand semantic embedded coding features; the feature collaborative interaction module 140 is used to perform dynamic memory collaborative interaction and weight calculation on the low-dimensional embedded coding features of the current power, the low-dimensional embedded coding features of the driving distance and the embedded coding features of the number of switching nodes with the current demand semantic embedded coding features to obtain the score value of the target shuttle.

[0072] Here, those skilled in the art will appreciate that the specific operations of each step in the above-mentioned four-way shuttle automatic control system have been described in detail above. Figures 1 to 4 The four-way shuttle automatic control method has been described in detail, and therefore, its repeated description will be omitted.

[0073] As described above, the four-way shuttle automatic control system 100 according to the embodiment of the present disclosure can be implemented in various wireless terminals, such as a server with a four-way shuttle automatic control algorithm. In a possible implementation, the four-way shuttle automatic control system 100 according to the embodiment of the present disclosure can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the four-way shuttle automatic control system 100 can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the four-way shuttle automatic control system 100 can also be one of the many hardware modules of the wireless terminal.

[0074] Alternatively, in another example, the four-way shuttle automatic control system 100 and the wireless terminal may also be separate devices, and the four-way shuttle automatic control system 100 may be connected to the wireless terminal via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0075] The above is merely an example of the principles of the present disclosure, and various modifications may be made by those skilled in the art without departing from the scope of the present disclosure. The above embodiments are presented for the purpose of illustration rather than limitation. The present disclosure may also take many forms other than those explicitly described herein.

Claims

1. A four-way shuttle automatic control method, comprising: The BLOCK mechanism of the shuttle vehicle is preset according to the warehouse layout; when the system starts, the preset BLOCK mechanism is loaded and deadlock monitoring is started; In response to receiving a task requirement, the shuttle vehicles that currently meet the scheduling conditions are obtained and the optimal shuttle vehicle is determined therefrom; the minimum driving route of the optimal shuttle vehicle is calculated, and the minimum driving route is sent to the optimal shuttle vehicle; the running state of the optimal shuttle vehicle is monitored in real time through the deadlock monitoring, and whether a deadlock occurs with other shuttle vehicles; in response to the optimal shuttle vehicle being deadlocked with other shuttle vehicles, a route is replanned for the affected shuttle vehicle, and the newly planned driving route is sent to the affected shuttle vehicle; characterized in that, in response to receiving a task requirement, the shuttle vehicles that currently meet the scheduling conditions are obtained and the optimal shuttle vehicle is determined therefrom, including: Calculate the score of each shuttle vehicle among the shuttle vehicles that currently meet the scheduling conditions, and use the shuttle vehicle with the smallest score as the optimal shuttle vehicle, including: obtaining the current power of the target shuttle vehicle, the driving distance to the target point, and the number of switching nodes; obtaining the natural language description of the current demand; embedding and coding the current power of the target shuttle vehicle, the driving distance to the target point, the number of switching nodes, and the natural language description of the current demand to obtain low-dimensional embedded coding features of the current power, low-dimensional embedded coding features of the driving distance, embedded coding features of the number of switching nodes, and current demand semantic embedded coding features; respectively perform dynamic memory collaborative interaction and weight calculation on the low-dimensional embedded coding features of the current power, the low-dimensional embedded coding features of the driving distance, and the embedded coding features of the number of switching nodes with the current demand semantic embedded coding features to obtain the score of the target shuttle vehicle; The current power low-dimensional embedded coding feature, the driving distance low-dimensional embedded coding feature, and the switching node number embedded coding feature are respectively subjected to dynamic memory collaborative interaction and weight calculation with the current demand semantic embedded coding feature to obtain the score value of the target shuttle vehicle, including: Perform attention interaction based on external dynamic memory on the current power low-dimensional embedding coding vector and the current demand semantic embedding coding vector to obtain the current power-current demand semantic collaborative interaction coding vector; Performing the external dynamic memory-based attention interaction on the driving distance low-dimensional embedding coding vector and the current demand semantics embedding coding vector to obtain a driving distance-current demand semantics collaborative interaction coding vector; Performing the external dynamic memory-based attention interaction on the commutation node number embedding coding vector and the current demand semantics embedding coding vector to obtain a commutation node-current demand semantics collaborative interaction coding vector; According to the current power-current demand semantic collaborative interaction coding vector, the driving distance-current demand semantic collaborative interaction coding vector and the switching node-current demand semantic collaborative interaction coding vector, the current power weight ratio, the driving distance weight ratio and the switching node weight ratio are obtained respectively; Obtaining a score value of the target shuttle vehicle according to the current power weight ratio, the travel distance weight ratio and the switching node weight ratio; Among them, performing attention interaction based on external dynamic memory on the current power low-dimensional embedded coding vector and the current demand semantic embedded coding vector to obtain the current power-current demand semantic collaborative interaction coding vector includes: Calculate a current power-current demand external knowledge fine-grained optimization feature interaction matrix between the current power low-dimensional embedding coding vector and the current demand semantic embedding coding vector; According to the current power-current demand external knowledge fine-grained optimization feature interaction matrix, the current power low-dimensional embedded coding vector and the current demand semantic embedded coding vector are fine-grained modulated response interaction to obtain the current power-current demand semantic collaborative interaction coding vector; Among them, according to the current power-current demand external knowledge fine-grained optimization feature interaction matrix, the current power low-dimensional embedded coding vector and the current demand semantic embedded coding vector are fine-grained modulated and responded to interact to obtain the current power-current demand semantic collaborative interaction coding vector, including: linearly transforming the current power low-dimensional embedded coding vector to obtain a current power query feature vector and a current power value feature vector, and using the current power-current demand external knowledge fine-grained optimization feature interaction matrix as a key matrix, optimizing and modulating the current power query feature vector, the current power value feature vector and the key matrix based on the Transformer model to obtain the optimized The current power low-dimensional embedding coding vector; linearly transform the current demand semantic embedding coding vector to obtain the current demand query feature vector and the current demand value feature vector and use the current power-current demand external knowledge fine-grained optimized feature interaction matrix as the key matrix, perform the Transformer model-based optimization modulation on the current demand query feature vector, the current demand value feature vector and the key matrix to obtain the optimized current demand semantic embedding coding vector; calculate the positional response between the optimized current power low-dimensional embedding coding vector and the optimized current demand semantic embedding coding vector to obtain the current power-current demand semantic collaborative interaction coding vector; The process of calculating the position response between the optimized low-dimensional embedding coding vector of current power and the optimized semantic embedding coding vector of current demand to obtain the current power-current demand semantic collaborative interaction coding vector can be expressed by the following formula: ;in, is the optimized low-dimensional embedding coding vector of the current power, is the optimized current demand semantic embedding encoding vector, It is the current power-current demand semantic collaborative interaction encoding vector.

2. The automatic control method of the four-way shuttle according to claim 1 is characterized in that: The current power of the target shuttle, the driving distance to the target point, the number of switching nodes, and the natural language description of the current demand are respectively embedded and encoded to obtain low-dimensional embedded coding features of the current power, low-dimensional embedded coding features of the driving distance, embedded coding features of the number of switching nodes, and semantic embedded coding features of the current demand, including: The current power of the target shuttle, the travel distance to the target point, and the number of commutation nodes are respectively embedded and encoded to obtain a current power low-dimensional embedded coding vector as the current power low-dimensional embedded coding feature, a travel distance low-dimensional embedded coding vector as the travel distance low-dimensional embedded coding feature, and a commutation node number embedded coding vector as the commutation node number embedded coding feature; The natural language description of the current demand is semantically embedded and encoded to obtain a current demand semantic embedded encoding vector as the current demand semantic embedded encoding feature.

3. The automatic control method of the four-way shuttle according to claim 2 is characterized in that: The natural language description of the current demand is semantically embedded and encoded to obtain a semantic embedded encoding vector of the current demand, including: performing word segmentation processing on the natural language description of the current demand and inputting the result into a semantic embedding encoder including a word embedding layer to obtain the semantic embedded encoding vector of the current demand.

4. The automatic control method of the four-way shuttle according to claim 3 is characterized in that: Calculating a current power-current demand external knowledge fine-grained optimization feature interaction matrix between the current power low-dimensional embedding coding vector and the current demand semantic embedding coding vector, including: Performing fine-grained interaction on the current power low-dimensional embedding coding vector and the current demand semantic embedding coding vector to obtain a current power-current demand fine-grained feature interaction matrix; According to the learnable memory parameter matrix, the current power-current demand fine-grained feature interaction matrix is ​​optimized based on external knowledge to obtain the current power-current demand external knowledge fine-grained optimized feature interaction matrix.

5. The automatic control method of the four-way shuttle according to claim 4 is characterized in that: According to the current power-current demand semantic collaborative interaction coding vector, the driving distance-current demand semantic collaborative interaction coding vector and the switching node-current demand semantic collaborative interaction coding vector, the current power weight ratio, the driving distance weight ratio and the switching node weight ratio are obtained respectively, which are used to: input the current power-current demand semantic collaborative interaction coding vector, the driving distance-current demand semantic collaborative interaction coding vector and the switching node-current demand semantic collaborative interaction coding vector into the weight specifier based on the feedforward neural network model respectively to obtain the current power weight ratio, the driving distance weight ratio and the switching node weight ratio.

6. A four-way shuttle automatic control system, used to execute the four-way shuttle automatic control method according to claim 1, characterized in that: The four-way shuttle automatic control system is used to calculate the score of each shuttle among the shuttles that currently meet the scheduling conditions, and use the shuttle with the smallest score as the optimal shuttle. The four-way shuttle automatic control system also includes: The shuttle working parameter acquisition module is used to obtain the current power of the target shuttle, the travel distance to the target point, and the number of switching nodes; The current demand acquisition module is used to obtain the natural language description of the current demand; A data embedding module is used to embed and encode the current power of the target shuttle, the travel distance to the target point, the number of switching nodes, and the natural language description of the current demand to obtain low-dimensional embedding coding features of the current power, low-dimensional embedding coding features of the travel distance, embedding coding features of the number of switching nodes, and semantic embedding coding features of the current demand; The feature collaborative interaction module is used to perform dynamic memory collaborative interaction and weight calculation on the current power low-dimensional embedded coding feature, the driving distance low-dimensional embedded coding feature, the switching node number embedded coding feature and the current demand semantic embedded coding feature to obtain the score value of the target shuttle vehicle.

Images